How We Analyze Risk

July 2015

1 Introduction
With Axioma Risk, you choose the way you analyze risk. Axioma Risk enables you to compare analyses and
identify the best one for your circumstances. For a quick assessment of specific risks, Axioma Risk offers beta
and measures of sensitivity. For short-term risk measures on actively traded portfolios and for other analyses
that are closely aligned with pricing factors, Axioma Risk offers granular models. For longer-term analyses,
Axioma Risk offers factor models. With Axioma Risk, you can also:
• Measure tracking error
• Calculate various versions of value at risk
• Run a large selection of preconfigured stress-tests
T
• Customize existing risk statistics
AF

• Create your own risk statistics

DR

2 Beta
Beta (β) indicates how one asset’s value has moved in relation to another’s. An asset with a (β) of 1 has
moved the same amount and in the same direction as the benchmark has moved. An asset with a (β) of -1
has moved the same amount as the benchmark has moved, but in the opposite direction. An asset with a (β)
of 2 has moved in the same direction as the benchmark, but twice as much. Other values for (β) continue
this pattern. β is computed as
Covariance(A, B)
β(Asset[A], Benchmark[B]) = .
Variance(B)

2.1 Calculating Beta

Cov (rasset , rindex )

β=
Var (rindex )
α (rasset , rindex )
β= , (1)
α (rindex , rindex )
where rasset is the return for the asset, rindex is the return for the index, Cov is the covariance function, and
V ar is variance. Both covariance and variance can be computed using the general form
n
X
αx,y = (xi − xµ )(yi − yµ ),
i=1

where xi and yi are the i’th return for asset x and y, respectively, and µ and yµ are their mean returns.
In Axioma Risk, β can be computed with vectors of either historical or relative returns on the asset and
index.

Axioma 1
2.2 Included Versions of Beta
Axioma Risk is delivered with the ability to calculate the following types of Beta:
• Beta, Linear-Parametric (Flat)
• Beta, Linear-Parametric, Decay Factor

• Beta, Linear-Parametric, Fixed Start Date

• Beta
• Beta, Historical

• Beta, Monte-Carlo
You can also edit these types of Beta or create your own.

3 Greek Sensitivities
3.1 Sensitivities Summary

Sensitivity Symbol Definition Black-Scholes Call Formulation

∂V
Delta ∆ ∂S Φ(d1)
T

∂V Φ(d1)
AF

Gamma Γ ∂∆
√
Sσ T

Rho ρ ∂V
KT e−RT Φ(d2)
DR

∂R

Theta Θ ∂V
∂T − SΦ(d1)σ
√
2 T
− RKe−RT Φ(d2)

Driftless Theta θ ∂V
∂T − SΦ(d1)σ
√
2 T
√
Vega ν ∂V
∂σ Ke−RT Φ(d2) T
∂V S
Lambda λ ∂S V
∂∆
Vanna (none) ∂σ
∂ν
Volga (none) ∂σ
∂∆
Charm (none) ∂T
∂Γ
Speed (none) ∂S
∂Γ
Zomma (none) ∂σ
∂Γ
Color (none) ∂T
∂ν
DVegaDTime (none) ∂T

Ultima

Dual Delta

Dual Gamma

Table 1: Sensitivity summary

Axioma 2
3.2 Delta
Delta (∆) is the sensitivity of an instrument to a change in the value of the derivative’s underlying asset, or

∂V
∆= .
∂S
For European options priced in the Black-Scholes framework, delta can be computed as

∆Call = e−qt Φ(d1)

or
∆Put = e−qt Φ(−d1),
where q is the continuous dividend yield, t is the time to maturity, Φ is the normal cumulative distribution
function, and d1 is from the Black-Scholes equation.

3.3 Gamma
Gamma (Γ) is the sensitivity of an instrument to a change in delta,

∂V ∂V
Γ= = 2 .
∂∆ ∂ S
For European options priced in the Black-Scholes framework, gamma can be computed as

φ(d1)
Γ = e−qt
T
√ .
Sσ t
AF

3.4 Rho
DR

Rho (ρ) is the sensitivity of an instrument to a change in interest rate regimes.

For European Black-Scholes options, ρ can be computed as

ρCall = Kte−rt Φ(d2),

or
ρPut = −Kte−rt Φ(−d2),
where K is the strike price for the option.

3.5 Theta
Theta (Θ) is the sensitivity of an instrument to a change in time to maturity.
For European Black-Scholes options, Θ can be computed as

Sφ(d1)σ
ΘCall = −e−qt √ − rKe− rtΦ(d2) + qSe−qt Φ(d1),
2 t
or
Sφ(d1)σ
ΘPut = −e−qt √ + rKe−rt Φ(−d2) − qSe−qt Φ(−d1).
2 t

3.6 Vega
Vega (ν) is the sensitivity of an instrument to a change in volatility regime. Note that there is no Greek
letter vega, so the symbol for nu is substituted to approximate the appearance of a Greek v.
For European Black-Scholes options, ν can be computed as
√ √
ν = Se−qt φ(d1) t ≡ Ke−rt φ(d2) t.

Axioma 3
3.7 Included Greek Sensitivities
Axioma Risk is delivered with the ability to calculate the following Greek sensitivities:
• Delta
• Gamma
• Rho
• Theta
• Vega

4 Other Sensitivities: Duration and Convexity

Duration is a first-order interest-rate sensitivity, important for fixed-income securities. This section describes
the three most common ways to derive it. Convexity is a second-order interest-rate sensitivity. This section
describes a way to compute it.

4.1 Duration
Effective Duration
To evaluate the sensitivity of an instrument’s price to interest rates, effective duration uses finite difference to
shift all interest rates by a fixed amount. Unlike Macaulay and modified duration, effective duration allows
the likelihood of exercise and the cash flows for bonds with optionality to change. The default shift size is
T

one basis point.

AF

Effective duration can be computed given the present value (P V ) of the instrument (based on its defined
pricing formula). It can be computed in an unshifted scenario and in two shifted scenarios—one in which
DR

interest rates have been shifted up by s basis points and one in which interest rates have been shifted down
by s basis points.
PV−s − PV+s
Effective Duration = .
2 × PV × s

Fisher-Weil (Or Macaulay-Weil) Duration

Fisher-Weil duration measures the weighted average time to maturity for a bond and adjusts for a nonflat
term-structure of yields. For a bond that uses continuous discounting, this statistic can be computed as
n
X τi Ci e−ri τi
Fisher-Weil Duration = ,
i=1
PV

where Ci , ri , and τi are the i’th cash flow, discount rate, and term to payment, respectively, and P V is the
present value of the instrument.

Macaulay Duration
Like Fisher-Weil duration, Macaulay duration is a weighted-average sensitivity to interest rates, but Macaulay
duration ignores the possibility of a varying yield curve and uses the yield to maturity instead.
n
X τi Ci e−ytm×τi
Macaulay Duration = ,
i=1
PV

where Ci , ytm, and τi are the i’th payment, yield to maturity, and term to payment for the i’th payment,
respectively, and P V is the present value of the instrument.
If the compounding is anything but continuous, the equation above becomes
n n
X Ci /(1 + ytm/f )f ×τi X Ci
Macaulay Duration = τi = τi ,
i=1
PV i=1
PV × (1 + ytm/f )f ×τi

Axioma 4
where f is the compounding frequency for the instrument, and ytm is the associated discretely compounded
interest rate.

Modified Duration
Modified duration extends Macaulay duration by including the number of coupon payments per year to
account for a change in interest rates over time. Macaulay duration and modified duration are equivalent
when compounding is continuous. Otherwise, modified duration is computed as
Macaulay Duration
Modified Duration = , (2)
1 + ytm/f
where ytm is the yield to maturity of the instrument.

4.2 Convexity
Convexity measures the rate of change of duration. As with duration, there are several versions of this
statistic. Here we describe two.

Effective Convexity
Effective convexity (like effective duration) takes into account the impact that interest rates would have on
optionality. It can be computed with a finite-difference approach as
PV−s − PV+s − 2PV
Effective Convexity = .
PV × s2
T

Convexity as Second-Order Macaulay Duration Sensitivity

AF

The variant of convexity that is based on Macaulay duration can also be computed with the continuous yield
to maturity for the instrument, as
DR

Pn
Ci τi2 e−ytmτi
Convexity = i=1 .
PV

Portfolio Aggregation
Both measures M , duration and convexity, can be aggregated in a portfolio by using a weighted average, as
n
X
Magg = wi mi ,
i=1

where
n
X
wi = PVi ÷ P Vi ,
i=1

and mi is the duration or convexity measure for the i’th risk factor.

4.3 Included Non-Greek Sensitivities

Axioma Risk is delivered with the ability to calculate the following non-Greek sensitivities:
• CM01
• CS01
• DV01
• EQ01
• FX01

Axioma 5
 • VG01
• Convexity
• Convexity, Effective
• Duration, Effective

• Duration, Key Rate

• Duration, Macaulay
• Duration, Macaulay-Weil

• Duration, Modified
• Duration, Spread
• Duration, Standard
• Effective Duration

You can also edit these sensitivities or create sensitivities of your own.

5 Risk Models
5.1 The Granular Model
T
AF

The granular model’s risk factors are closely aligned with pricing factors. These risk factors may include
zero-coupon bond prices derived from yield curves for the currencies to which a portfolio is exposed, price
DR

returns for the stocks in the portfolio, and points from the implied volatility surface of options in the portfolio.
The granular model is especially useful for short-term risk measures on actively traded portfolios that include
bets on relative-value trades.

5.2 Multifactor Risk Modeling

Multifactor risk models include less-granular risk factors and accommodate longer-term risk management.
The particular factors may reflect the way an institution makes investment decisions or a manager’s belief
that certain fundamental factors drive asset returns.
Factors in multifactor models fall into two broad categories:

• Fundamental factors:
– Industry and country factors reflect a company’s line of business and country of domicile.
– Style factors encapsulate the financial characteristics of an asset—a company’s size, debt levels,
liquidity, etc. They are usually calculated from a mixture of market and fundamental (i.e., balance
sheet) data.
– Currency factors represent the interplay between local currencies of the various assets within the
model.
– Macroeconomic factors capture an asset’s sensitivity to variables such as GNP growth, bond yields,
and inflation.
• Statistical factors are mathematical constructs responsible for the observed correlations in asset returns.
They are not directly connected to any observable real-world phenomena, and may change from one
period to the next.

Axioma 6
An asset’s return is decomposed into a portion driven by these factors (common factor return) and a residual
component (specific return) to produce the following model at time t:
   b · · · b1m 
  
r1 11  u1
 r2   b21 b2m  f 1
..  +  u2 
   
 ..  = 
  .. . . ..   .
. .  .. 
 . .     
 .  
fm

rn .. un
bn1 . bnm

or, more succinctly, in matrix form:

r = Bf + u,
where r is the vector of asset returns at time t, f the vector of factor returns, u the set of asset-specific
returns, and B the n × m exposure matrix, whose elements denote each asset’s exposure to a particular
factor.
In essence, the multifactor model is a dimension-reduction tool that reduces the problem of calculating an
n × n asset-returns covariance matrix to calculating the variances and covariances of a much smaller number
of factors and n specific variances.
For a discussion of the various parts of a risk model and the stages in its construction, see the Axioma
Robust Risk Model Handbook.

5.3 Axioma Equity Risk Models

The Factor Risk Model for Equities
For its equity risk models as for other risk models, Axioma first ensures that returns are modeled correctly
T

and robustly. Then, if the model has been sensibly constructed with no important factors missed, the specific
AF

returns are uncorrelated with themselves and with the factors, and the factor risk model can be derived thus:
DR

var(r) = var(Bf + u)
Q̂ = BΣB T + ∆2 .

The asset returns covariance matrix Q̂ is a combination of a common factor returns covariance matrix Σ and
a diagonal specific variance matrix ∆2 .

The Factor Covariance Matrix

The factor covariance matrix is calculated directly from the time-series of factor returns. Regression models
estimate a set of factor and specific returns at each time period and eventually build a returns history.
Statistical models, on the other hand, generate the entire time-series anew with each iteration.
 
f1,1 f1,2 · · · f1,T
 f2,1 f2,2 · · · f2,T 
F = . ..  .
 
.. ..
 .. . . . 
fm,1 fm,2 ··· fm,T

The more recent the event, the more it should influence the model. But because the history must be long
enough so that covariances can be estimated reliably, less recent observations cannot be excluded. Therefore,
Axioma weights the returns with an exponential weighting scheme:

−(T − t)
wt = 2 t = 0, ..., T,
λ
where T is the most recent time period, and λ is the half-life parameter, the value of t at which the weight
is half that of the most recent observation.

Axioma 7
 Thus, given a half-life, weights W = diag(w1 , w2 , ...wT ) are computed, and the factor returns history is
weighted to yield a new matrix of values:
 1 1 1 
w12 f1,1 w22 f1,2 · · · wT2 f1,T
1 1 1
 
w 2
f w 2
f · · · w 2
f
 
1  1 2,1 2 2,2 T 2,T 
F̃ = F W 2 =  .. .. .. ..
.

 . . . . 
 1 1 1

w12 fm,1 w22 fm,2 ··· wT2 fm,T
From this, the factor covariance matrix is simply calculated as
FWFT
Σ = var(F̃ ) = .
T −1
Most Axioma risk models use a longer half-life for estimating the factor correlations and a shorter one for
the variances. The correlations are more stable than the variances, and a longer half-life enables the large
number of relationships to be estimated. The resulting correlations are then scaled by the corresponding
variances to obtain the full covariance matrix.

Content
Axioma Risk’s content team sources market data from leading data providers. The team then processes,
validates, assimilates, and updates the data. Corporate actions such as splits, stock dividends, rights issues,
and spin-offs are all processed as an adjustment factor that is used when calculating total returns. Mergers
and acquisitions are reviewed, and the histories of the “old” and “new” securities are linked when appropriate,
in which case the “new” security inherits the AxiomaID and the fundamentals from the “old.”
T
AF

Global/Regional Models
DR

• Asia-Pacific Model
• Asia-Pacific (ex-Japan) Model
• Emerging Markets Model
• Europe Model
• North America Model
• World-Wide Model
• World-Wide (ex-US) Model

Single-Country Models
• United States Model
• United States Macro Model
• Australia Model
• Canada Model
• China Model
• UK Model
• Japan Model
• Taiwan Model
• The Goldman Sachs Shortfall Model
For more information on the Axioma Equity Risk Models, see the Axioma Robust Risk Model Handbook.

Axioma 8
5.4 The Axioma Fixed-Income Factor Model
Like any factor model, Axioma’s fixed-income factor model is designed to separate and quantify systematic
and idiosyncratic components of risk. The systematic components in a typical fixed-income portfolio include
yield curve, volatility, credit, currency, inflation, and prepayment risks.
Without a risk model, a fixed-income asset manager can measure the risk in a portfolio with various
metrics. For instance, exposures measured via sensitivities to key risk factors provide a useful snapshot of
risk. However, exposures do not capture all the risks in a portfolio. For example, exposures cannot determine
whether a 0.5 spread duration in an MBS position is more volatile than a 0.1 spread duration in a corporate
bond position. In contrast, a risk model captures not only the risks of all individual positions, but also the
interactions or co-movements of different risk factors.
In its simplest form, the risk of a position to a market factor equals the position’s exposure multiplied
by the return volatility of the factor. Among multiple positions and risk factors, the interplay depends upon
portfolio weights and the covariance of risk factors. Our factors are either directly observable, like yield
curves, or latent factors derived from a cross-sectional regression. Thus our model combines parametric and
cross-sectional approaches for various risk factors.

Sensitivities as Exposures Axioma Risk models each return driver, such as interest rates or spreads, with
its own set of risk factors. Unlike equity exposures, which are descriptors or dummy variables, most fixed-
income exposures are price-based sensitivities. For example, interest-rate exposures require the computation
of key-rate durations, which are sensitivities to different tenors of the yield curve. When individual key rates
are selected as factors, exposures are simply the durations, and when “shape” factors such as level or slope
are selected, the exposures are combinations of durations.
The price (as well as risk) of a fixed-income security depends on factors such as interest rates, embedded
T

options such as calls and puts, and the creditworthiness of the issuer. Since standard pricing models for
AF

fixed-income securities are functions of the variables mentioned above, we can perform a Taylor expansion to
compute sensitivities:
DR

n
∆P X 1 ∂P 1 ∂2P 1 ∂P 1 ∂P 1 ∂P
= ∆yi + 2
(∆y)2 + ∆s + ∆σ + ∆t, (3)
P i=1
P ∂yi 2P ∂y P ∂s P ∂σ P ∂t

where P is the bond price, yi is the zero rate at tenor τi , σ is the implied volatility, and s is the spread.
The convexity term utilizes the parallel shift of the entire term structure ∆y, and the last term represents
temporal changes such as security aging. We can rewrite this equation as
n  
∆P X 1 ∂P ∆σ 1 ∂P
= (−Di )∆yi + 12 Γ(∆y)2 + (−Ds )∆s + σ + ∆t (4)
P i=1
P ∂σ σ P ∂t
n  
X
1 2 ∆s 1 ∂P ∆σ 1 ∂P
= (−Di )∆yi + 2 Γ(∆y) + (−Ds s) + σ + ∆t, (5)
i=1
s P ∂σ σ P ∂t

where
1 ∂P
Di = − is the key rate duration, (6)
P ∂yi
1 ∂P
Ds = − is the spread duration, and (7)
P ∂s
1 ∂2P
Γ= (∆y)2 is the convexity. (8)
P ∂y 2

The exposures (sensitivities) are computed with Axioma’s suite of multiasset pricing functions. For
example, given a corporate bond with embedded options, the duration and spread duration are computed
from full pricing models where interest rates and spreads are shifted.

Axioma 9
The Formula for the Fixed-Income Factor Model Our fixed-income factor model has the following form
for security i in country c:
(carry) (ir) (vol) (s) (mbs)
r̃i = ri − ri = ri + ri + ri + ri + εi (9)
X (ir) (ir) X (vol) (vol) X (s) (s) X (mbs) (mbs)
= Xi,k fk + Xi,k fk + Xi,k fk + Xi,k fk + εi , (10)
k k k k

where r̃i has been adjusted for future (deterministic) returns due to carry, and εi is the idiosyncratic portion
of return that is independent of all other sources of risk. For a discussion of each term on the right-hand side
of (10), see the Fixed Income Factor Model document.

5.5 The Axioma Commodity Model

The Axioma commodity model extends the coverage of Axioma’s renowned equity models to commodity
futures and their derivatives. Like other Axioma models, the commodity model:
• Is built using a transparent methodology.
• Is estimated daily based on daily data.
• Estimates factor returns cross-sectionally.
The commodity model allows portfolio managers with exposure to commodity futures and their derivatives
to examine their portfolios with Axioma Portfolio Optimizer and backtest in a manner consistent with other
T
Axioma models.
AF

Axioma Risk provides constant-maturity forward (CMF) curves as well as other market data for a wide
array of commodities.
DR

Submodels
The commodity model is a collection of submodels, each dealing with a particular underlying commodity or
family of commodities. Each submodel has a number of factors, and each futures contract has a time- and
contract-dependent exposure to each factor.
Together the submodels cover all the constituents of the major commodity indices:
• Energy
– Oil (such as crude oil, heating oil, gasoline)
– Natural gas
• Metals
– Industrial metals (such as copper and aluminum)
– Precious metals (such as gold and silver)
• Agriculture
– Grains and oilseeds (such as wheat and corn)
– Softs (such as cotton and coffee)
– Livestock (such as cattle and hogs)
The submodels for widely-traded commodities have several factors. For example, the crude oil model has
three major factors (shift, tilt, and curve) and two basis factors, which cover relative movements between
Brent and WTI (West Texas Intermediate). Other submodels have one or two factors. A futures contract
has a (cross-sectional) exposure to each relevant factor. This exposure is a function of the contract’s time to
maturity and of its expiry month (for seasonal commodities). A subset of these futures forms the estimation
universe from which the factor returns are estimated.

Axioma 10
Major Features of the Model
The model captures the essential characteristics of the futures markets:
• Cross-Sectionality. The exposure of each futures contract to each factor is a deterministic function of
its time to maturity and its expiry month. Because the model avoids time-series regressions, contracts
respond quickly to rising volatility in their final months.
• Term Structure. An entire curve of futures references any given spot price (such as for WTI). For a
heavily traded commodity, the curve may extend more than a decade into the future and may contain
more than one hundred contracts.
Many commodity futures exhibit a significant increase in volatility as they approach expiry. This
increase is called the Samuelson effect. The Axioma commodity model captures this effect in two
related ways. Each submodel has a shift factor. Submodels for many commodities (such as natural
gas) also have a tilt factor. The shift and tilt factors reflect the increasing volatility of contracts as
they approach expiry.
• Seasonality. For many commodities, the price curve (as observed on a single day) has a clear seasonal
pattern. For example, gasoline prices are higher in summer than in winter. Volatility is also seasonal.
Gasoline prices are not only higher, but also more volatile in summer. In the Axioma commodity model,
the exposures of contracts depend, in general, on both time to expiry and season, and thus capture this
effect.
• Exposures. The exposure βC,F of a contract C to some factor F .
• Derived Products. The effect that the price of one commodity has on the price of another.
For more information on the Commodity Factor Model, see the Axioma Commodity Factor Model Hand-
T

book.
AF
DR

6 Value at Risk
Value at Risk (VaR) is a statistic that attempts to answer the question “Over a specified time and with
a particular confidence level, how much should I expect to lose?” VaR is a percentile of a profit-and-loss
distribution. More precisely, it is the (predicted) maximum portfolio loss over a specified horizon at a
confidence level C. Equivalently, VaR is also the (predicted) minimum portfolio loss over a specified horizon
with probability 1 − C. See (13)–(14) for a mathematical definition. For example, VaR 95% is the maximum
expected loss with confidence level 95%, or equivalently, the minimum expected loss with a probability of
5%.
Axioma Risk provides three methods for computing VaR: parametric, historical, and Monte Carlo. This
section describes each method with its advantages and disadvantages.

Definition of VaR
Let q ∗ denote the percentile of the profit and loss distribution of future portfolio values v with confidence
level C and time horizon T :
Z q∗
1−C = f (v) dv = Pr(v ≤ q ∗ ) min loss with prob 1 − C (11)
−∞
Z ∞
or equivalently, C = f (v) dv = Pr(v ≥ q ∗ ) max loss with prob C. (12)
q∗

Here f (v) is the marginal distribution for the simulated future value v of the portfolio. Two versions of VaR
are available in the system, one in terms of an absolute loss and the other in terms of a relative loss:
VaR = −q ∗ (13)
∗
VaR(mean) = −(q − ET [v]). (14)
Although we defined VaR at the portfolio level, it is applicable to individual positions and subportfolios.
VaR is available in terms of a dollar loss or as a percentage of the portfolio loss.

Axioma 11
Confidence Level and Computation of VaR
As outlined above, the confidence level defines for the reader the value below which a percentage of values
fall. VaR is a loss measure, meaning that 95% VaR tells the reader the value below which 5% of the losses
will fall.
Given a confidence level and a vector of portfolio values (v1 , . . . , vn ), ordered from lowest to highest, we
can define the index for the percentile as

η = (1 − C) × (nobs + 1), (15)

where nobs is the number of return observations in the vector. Note that there are many approaches available
for computing a percentile. This approach, described separately by Gumbell and Weibull in 1939, divides
1
the distribution into n + 1 regions as a first step. This ensures that on average each has a probability of n+1 .
If you have a vector of 11 integers i = {1,2,3,4,5,6,7,8,9,10,11}, the median would be 6 (the midpoint), and
the first quartile would be 3 (0.25 × [11 + 1]). This is consistent with the statement “On average, 25% of the
distribution is below 3, 50% of the distribution is below 6, and so on.”
Given this index, we can now compute the profit/loss value at the specified confidence level. If η is an
integer, the value is simply the return observation at that index. If η is a noninteger, we would linearly
interpolate between the two values surrounding η to compute the value for our specified confidence level.
If η has a value less than 1 or greater than nobs , we will effectively take the value at index 1, or the value
corresponding to nobs , respectively; we will not extrapolate.
Thus we have: 
 v1
 η<1
qη∗ = wvi + (1 − w)vi+1 i ≤ η ≤ i + 1 , (16)

vnobs η > nobs
T

AF

where w = i + 1 − η and VaR is given by (13) or (14).

DR

6.1 Historical VaR

Historical VaR makes no distribution assumptions while parametric and Monte Carlo VaR do. The cost of
computing historical VaR is equal to the number of return observations multiplied by the number of risk
factors to be simulated. Generally, it is less computationally expensive than Monte Carlo VaR.
With historical VaR, there is currently no concept of a decay factor. All risk factor returns will be equally
weighted.
Simply put, historical VaR is computed by reading the result at the specified confidence level from the
sorted vector of P&Ls (profit and losses) for the position of interest. The results can be aggregated across
positions if the original sort order is preserved prior to the aggregation step.
In other words, absolute VaR is given by
∗
VaRC
historical = −qη , (17)

where C is the user-specified confidence level, η is the index in the vector of returns sorted in ascending order,
defined in (15), and qη∗ is given by (16).

6.2 Parametric VaR

Parametric VaR is the least computationally expensive VaR statistic to compute because it does not require
full valuation when computed, but rather transforms the pricing function to linear. This methodology assumes
that returns are log-normally or normally distributed. Generally, this methodology would be applied to
equity portfolios or other portfolios in which the underlying assets are assumed to have linear or mostly
linear behavior. The calculation of parametric VaR is closed-form and explicitly requires the calculation of a
covariance matrix. Thus no simulations are required.

Axioma 12
Methodology
To calculate Parametric VaR, we must compute first-order price sensitivities for each of the risk factors.
Suppose we have a position with a risk factor Si whose returns follow a log-normal or normal distribution.
The sensitivity of this position is given by either its delta equivalent or its first-order derivative:
∂PV ∆Si


 Si × if Si is log-normal
∂Si Si

∆PV = . (18)
 ∂PV

 × ∆Si if Si is normal
∂Si
We rewrite (18) succinctly as
∆PV = δi Ri . (19)
Where possible, we use a closed-form solution to compute the sensitivity δ. For example, we know that the
price sensitivity for equity positions is 100% with respect to the underlying equity risk factor. That is, if
the equity price goes up $1.00, the value of the asset in the portfolio also goes up by $1.00. Likewise, for
positions that are not FX derivatives, the sensitivity for the currency is 1.0. For vanilla options, closed-form
solutions exist for each of the risk factors: e.g., delta can be computed as φ(d1) for a European option.
When a closed-form solution is not available, a finite-difference approach can be applied to find a price
sensitivity. When using the finite-difference approach to compute first-order sensitivities, we apply a small
shift ∆ to the risk factor. For example, the computation for a delta equivalent is
PVSi +∆ − PVSi −∆
δi = Si . (20)
∆
T

The δs for each risk factor are additive across instruments. This additiveness is particularly useful for
AF

aggregation within drill-down dimensions. Thus the P&L at the portfolio is

DR

X
∆PVportfolio = δi Ri = δ t R. (21)
i

Parametric VaR Computation

If a covariance matrix is supplied, the standard deviation of portfolio returns is
p √
σportfolio = E[δ t RRt δ] = δ t Σδ. (22)

VaR can then be estimated at a specified confidence level by using

VaR = −zα σ, (23)

where zα is the inverse cumulative distribution function at the specified confidence level.

6.3 Monte Carlo VaR

Monte Carlo VaR is potentially the most computationally expensive risk measure to compute: it can require
far more simulations than historical VaR requires. For each simulated risk factor, we reprice all portfolio
positions, aggregate the results, and sort them in ascending order (v1 , . . . , vn ). VaR is then computed using
(13)–(16).

6.4 Risk Contribution

Introduction
Risk Contribution is an additive version of Value at Risk. It could also be described as the change in the
value of a portfolio when the value of a single asset changes.
Mathematically, the risk contribution of asset A is defined as
∂VaR
wA , (24)
∂wA

Axioma 13
where wA is the value of the asset.
Ultimately, in this measure, the sum of a set of subportfolio VaR measures equals the portfolio’s total
VaR, which must itself be equal to the original (usual) VaR measure.
There is some confusion regarding the name of this statistic. Some call it Marginal VaR. Others call it
Incremental VaR, which we define as the change in total VaR when an asset is added to a portfolio. We call
the statistic described in this section Risk Contribution to avoid the confusion regarding the other two terms.

Approach for Simulation-Based Methodologies

To compute the total P&Ls for a portfolio of assets (or subportfolios), we could simply add the P&Ls for
each of the assets in the portfolio. If we have i P&Ls and n subportfolios s (s = 1 : n), the ith P&L for the
portfolio could be described as
n
X
riportfolio = ris . (25)
s=1

The remainder of this section refers to position and portfolio P&Ls simply as returns.
A single vector of returns represents each subportfolio, and the portfolio returns are then the sum of the
subportfolio vectors. Conventional VaR treats these vectors as independent of one another. Risk Contribution
forces a dependency between total VaR and subportfolio VaR. It takes the observation of equation (25) and
uses the additivity that it describes. It sorts all subportfolio returns with respect to the total portfolio
returns.
For example, suppose we have the table of returns shown in Table 2 for two subportfolios A and B, along
with the total return T.
T

A B T = A+B Index
AF

0.08 -0.11 -0.03 1

-0.04 0.02 -0.02 2
DR

0.02 0.01 0.03 3

-0.12 0.04 -0.08 4
0.14 -0.18 -0.04 5

Table 2: Unsorted returns for two assets and the total portfolio returns

Risk Contribution sorts all rows according to total returns, as shown in Table 3. The subportfolio returns
in this table may at first glance appear to be unsorted, but comparing the Index column in Tables 2 and 3
shows that the elements are the same, but that Table 3 has been sorted by the items in the Total column.

A B T = A+B Original Index

-0.12 0.04 -0.08 4
0.14 -0.18 -0.04 5
0.08 -0.11 -0.03 1
-0.04 0.02 -0.02 2
0.02 0.01 0.03 3

Table 3: Returns sorted with respect to the total portfolio

Axioma Risk does not allow drilldowns in which a single position could fall into multiple buckets. Such
drilldowns include some of the most granular drilldowns, such as risk factor or risk type, but other drilldowns
such as the currency drilldown would also be disallowed (because an FX option would have two curren-
cies). These drilldowns are disallowed because a statistic can be additive in only one dimension, not across
dimensions.

Noise
Noise is inherent in simulations. A simple equation like (25) ignores the possibility that the return at the
index of interest does not represent the possible returns for a given subportfolio. For example, let’s say that

Axioma 14
1000 simulations are being used to compute VaR for a given portfolio. The portfolio’s returns will all be
sorted, and the index for the quantile of interest will be identified—say the 51st worst loss for 95% VaR. This
value will be based on the entire distribution of returns. If we then compute MVaR for the subportfolios,
we will be using a set of returns that does not represent the distribution of returns of the subportfolio, but
rather the distribution of returns for the portfolio. This incompatibility introduces noise that needs to be
corrected. With Monte Carlo in particular, the single return observation may not represent the asset’s typical
relationship to the portfolio. To correct this, we simply expand from a single return observation to a range
of return observations.

Averaging
With the averaging approach, we simply sample the return observations surrounding the index for the quan-
tile of interest. We sample ± whichever is larger—10 return observations or 1% of the number of return
observations.
Our procedure for computing Risk Contribution with this approach follows.
We begin with a set of n + 1 return vectors, where n is the number of drilldown results to be returned
and the n + 1th vector is the portfolio return. We sort all return vectors with respect to the total return as
described above.
Once we have a set of sorted vectors, we identify the index η corresponding to the quantile of interest, as
described in (15). We next define ν, as the offset above/below the index, to use in computing the average as

ν = max(10, 0.01 × nSims). (26)

Given this observation, we compute an average return µ for the observations surrounding the portfolio’s
VaR return as
T
AF

η+ν
1 X portfolio
µportfolio = s . (27)
ν × 2 + 1 i=η−ν i
DR

We now define a ratio, α, between the original portfolio VaR and the average VaR as
µportfolio
α= . (28)
VaRportfolio
We now use (27) to compute the average returns for each of the k subportfolios as
η+ν
1 X
µk = sk , (29)
ν × 2 + 1 i=η−ν i

and scale the result by α to compute Risk Contribution for the k’th subportfolio as
µk
RCVaRk = . (30)
α

6.5 Incremental VaR

The Incremental VaR of a position is the amount of risk that a position adds to or removes from a portfolio.
To compute this statistic with any methodology, one simply removes a given position from the portfolio and
recomputes VaR, as

IVaRi = VaRportfolio − VaR(portfolio−Asseti ) . (31)

6.6 Conditional VaR

Conditional VaR (CVaR), which is also referred to as Expected Shortfall, describes the expected loss if the
VaR threshold is met or exceeded. In other words, “What is the loss conditional on meeting VaR?”

Axioma 15
Linear Parametric Conditional VaR
In a closed-form approach, Parametric CVaR can be computed as
σ 2
CVaR = √ e−0.5zα . (32)
(1 − α) 2π

Conditional VaR Using Simulation-Based Approaches

For both historical and Monte Carlo methodologies, CVaR can be computed by taking the average loss for
all observations exceeding the VaR threshold. Explicitly, this can be computed as
κ
1X
CVaR = ri , (33)
κ i=1

where κ = bηc | bηc =

6 η & κ = bηc − 1 | bηc == η.

6.7 Conditional VaR Risk Contribution

Just as with VaR, the impact of increasing the weight of a particular position can be evaluated with CVaR.
We call this measure CVaR Risk Contribution, which we define succinctly as
∂CVaR
CVaR Risk Contribution = wA . (34)
∂wA

CVaR Risk Contribution Using Simulation-Based Methodologies

Computing CVaR Risk Contribution using simulation-based methodologies is not unlike computing VaR risk
T

contribution.
AF

The first step, just as with VaR Risk Contribution, is to sort the vectors of returns with respect to the
total return.
DR

The second step is to compute the average of the tail for the subportfolios.

6.8 Included Types of Value at Risk

Axioma Risk is delivered with the ability to calculate the following types of Value at Risk:

• VaR, Linear-Parametric
• VaR, Linear-Parametric (decay)
• CVaR, Monte-Carlo
• CVaR, Monte-Carlo, Flat

• Expected Shortfall (CVaR), Historical

• Expected Shortfall, Historical
• Expected Shortfall, Monte-Carlo

• VaR MC 0.94 decay

• VaR, Historical
• VaR, Historical (Abs Dates)
• VaR, Monte-Carlo

You can also edit these types of Value at Risk or create your own measure of Value at Risk.

Axioma 16
7 Tracking Error
Tracking Error compares the risk of a portfolio with the risk of a benchmark asset or index. In other words,
this measure answers the question “How much additional risk does my portfolio have over a benchmark?”
To compute Tracking Error, first scale the initial market value of the benchmark position (or set of
positions, via a benchmark account) to be the same as the market value of the portfolio of interest. Note
that

PVPortfolio = γPVBenchmark . (35)

Given this, solve for the scalar γ as

γ = PVPortfolio ÷ PVBenchmark , (36)

and apply it to all returns in the benchmark.

Subtract these scaled returns from the portfolio for which relative VaR is being computed:

r̂iPortfolio = riPortfolio − γriBenchmark . (37)

After computing the set of returns r̂iPortfolio , calculate tracking error at a given confidence level by using
equations (13)–(16).

7.1 Included Measures of Tracking Error

Axioma Risk is delivered with the ability to calculate the following measures of tracking error:
• Historical Tracking Error (2Y)
T
AF

• Tracking Error (MAC Global)

DR

• Historical TE VaR (2Y)

• TE VaR (MAC Global)
• TE Active
• TE VaR Contribution
You can also edit these measures or create measures of your own.

8 Stress Tests
A stress test shifts underlying risk factors or model parameters to simulate the effect of real-world devel-
opments on the value or risk of assets and portfolios. Stress tests are used in strategic planning, hedging,
evaluation of models, and regulatory compliance. Unlike risk models, stress tests do not weight the prob-
ability of different scenarios, but a well-designed stress test can isolate the consequences of changes to a
particular risk factor or to the macroeconomic environment. A well-designed stress test can also account for
the causes and effects of those changes.

Stress Test Categories

Stress tests can be classified into the following types:
• Instantaneous Stress Tests
• Horizon Stress Tests
• Historical Stress Tests
• Model Parameter Stress Tests
Each type has applications and constraints, which will be outlined next.

Axioma 17
8.1 Instantaneous Stress Tests
Definition
Instantaneous stress tests do not have a horizon, but instead gage the impact that shifting one or more risk
factors would have on the present value of an asset or assets. This kind of stress test would be used when
computing sensitivities such as effective duration or DV01.
This kind of stress test ages positions and changes cash flows only insofar as moneyness for the asset (or
its underlying) changes with the stressed levels. Those changes to moneyness would then affect the likelihood
of optionality being exercised or payment regimes changing.
An instantaneous stress test can be coupled with aging, which would then answer the question “What
would happen to the value of my asset(s) or portfolio next month if a given risk factor (or risk factors) went
up or down now?”

Inputs
The inputs for an instantaneous stress test are:

• Risk factor

• Shift amount
• Aging scheme (including none)

8.2 Horizon-Based Stress Tests

Definition
T

Horizon-based stress tests require two inputs—the risk factor or factors to stress and the horizon over which
AF

the stress should be applied. This kind of stress test answers the question “What would happen to the value
of my asset(s) or portfolio if a given risk factor or risk factors changed by a specified amount over the next
DR

n days, weeks, or months?” In this kind of stress test, one can evaluate the impact of a gradual change of
a risk factor, as opposed to a shock to the system. Note that if the horizon over which the stress should be
applied is 0 days, the result would be the same as for an instantaneous shock.
Horizon-based stress tests can be applied with or without aging.

Inputs
The inputs for this kind of stress test are:

• Risk factor
• Shift amount list (which would include timing / path)
• Interpolation / gap-filling method (linear, Brownian bridge, etc.)

• Horizon over which to apply the stress (including instantaneous)

• Aging scheme (including none)

8.3 Historical Stress Tests

Definition
Historical stress tests apply the events of a historical period to current observations of the same time series.
Usually, historical stress tests apply events that had a significant impact on financial markets—for instance,
the demise of Long-Term Capital Management, the implosion of the dot-com bubble, or the Flash Crash of
2010.
With historical stress tests, the user specifies the date range or date ranges to consider when computing
market movements and then applies the selected dates to the current levels of risk factors.

Axioma 18
Inputs
The required inputs to this kind of stress test are:

• Date range
• Sampling frequency
• Horizon over which to apply the stress (including instantaneous)
• Aging scheme (including none)

This methodology requires especially careful implementation when the number of days in the observation
period does not equal the number of days in the expectation period.

8.4 Model Parameter Stress Tests

Definition
Model parameter stress tests shift a pricing-model parameter that is not market observable. Examples of such
parameters include the recovery rate for an instrument with credit risk, the time to maturity for instruments
that have one, the refinancing rate for a mortgage pass-through, or any other feature that would be reasonable
to shift. (Features that could not reasonably be shifted, such as accrued interest, would not fall into this
category.)

Inputs
Required inputs to this kind of stress test are:
T
AF

• Model parameter
• Cohort over which to apply this stress
DR

• Aging scheme

8.5 Stress Test Cohorts

Stress tests can be applied to individual time series or to groups of time series via:

• Time series stress tests

• Grouped time series stress tests

Time Series Stress Tests

A time series stress test is applied only to a specific time series, such as IBM or the 10-Year Treasury rate.

Grouped Time Series Stress Tests

A grouped time series stress test is applied to a group of time series, such as equities or interest rates.
This sort of stress test could be more specific. For instance, one could stress only technology stocks down
by 10%, while leaving the rest of the market unchanged.

8.6 Stress Test Specification

To specify a stress test, a shift amount must be provided. The shift amount can be provided in one of three
forms, depending on the kind of stress test:

• Absolute stress
• Relative stress
• Standard deviation or standard deviations stress

Axioma 19
Absolute Stress
With an absolute stress, the user simply shifts a risk factor or group of risk factors up or down by a specified
amount. This would typically be a stress in basis points to an underlying risk factor. An example would be
a 50bp shock to the 10-year rate for a corporate issuer, whose current rate of 4.00% would become 4.50%.

Relative Stress
With a relative stress, the user shifts a risk factor or group of risk factors up or down by a specified relative
amount. An example would be a 5% shock to the 10-year rate for a corporate issuer, whose current rate of
4.00% would have a stressed rate of 4.20%.

Standard Deviation or Standard Deviations Stress

With a standard deviation stress, the user shifts a risk factor or group of risk factors up or down by a specified
number of standard deviations.
The standard deviation of the factor or factors must first be computed. Therefore, this kind of stress
requires more parameterization than the other kinds require. The following parameters need to be supplied
to compute the standard deviation:

• Statistics term
• Return horizon

• Decay factor (or half-life)

• Return type (if an override to default)
T

The user can also specify the number of standard deviations.

AF

With these parameters, the standard deviation of Return horizon returns would be computed and then
applied to the specified risk factor. For example, 0.5 standard deviations based on one year of equally
DR

weighted weekly absolute returns could be applied to the 10-year rate for a corporate issuer with a current
rate of 4.00%. If the weekly standard deviation were 0.1%, the shifted rate would be 4.05%.

8.7 Transitive Stress Tests

A transitive stress test can be created as either an Axioma factor model transitive stress test or as a time
series transitive stress test. The Axioma factor model transitive stress test is preferred. In it, the Axioma
factor model moves risk factors that change the portfolio’s value. This approach shows how a stress to the
value factor affects the portfolio.
Required inputs for an Axioma factor model transitive stress test are:
• Factor to stress
• Stress amount

The time series transitive stress test derives time series correlations among a set of independent risk
factors and uses the correlations between these independent factors and a set of dependent factors to predict
movements. This approach shows, for instance, how a stress to the price of oil affects a long-only equity
portfolio.
Required inputs for a time series transitive stress test:

• Factor to stress

• Stress amount
• Correlation matrix construction parameters (if time series)—date range, sampling, etc.

Axioma 20
8.8 Path Dependency Effects Associated with a Stress Test
Any stress test to a risk factor that affects the pricing of a path-dependent instrument may change the payoff
of the instrument. If a historical period is replayed, the impact would extend beyond the replay horizon.
Stress tests would affect American options, range accrual notes, barrier options, variance swaps, and other
positions.

8.9 Stress Test Examples

To put the above information together in a relevant manner, we now describe the impact of stress tests on
asset pricing models and risk factors.

8.10 Instantaneous Stress Test

Time Series
To show how an instantaneous stress test to a time series can be useful, use a stress test to estimate delta
for a European long call option on an equity priced using the Black-Scholes formula.
The values shown are for Apple, as of June 1, 2011.
First, from the values below, compute the value of the European option absent any stress:

S0 = 345.51
K = 380.06(S0 × 1.10)
r = 0.18%
T

σ = 21.9%
AF

T = 1.0
DR

You can now compute the variables used in the option pricing formula. To begin, compute d1 as

ln(S0 /K) + (r + σ 2 /2)(T )

d1 = √ (38)
σ T
ln(345.51/380.06) + (0.0018 + 0.2192 /2) × 1.0
d1 = √ = −0.31748.
0.219 × 1.0
Next, compute d2 as

√
d2 = d1 − σ T (39)
√
d2 = −0.31748 − 0.219 × 1.0 = −0.53648.

Finally, given these inputs, compute the value for the unstressed option as

c = S0 Φ(d1 ) − Ke−rT Φ(d2 ) (40)

−0.0018×1.0
c = 345.51Φ(−0.31748) − 380.06e Φ(−0.53648) = 17.49351.

Given current market data, you now know the value of the option. To compute delta, which is the
∂V
sensitivity of the derivative’s value to a change in the value of the underlying ( ∂S0
), introduce a small change
to the value of the underlying by increasing its value through a relative shift of 1% to 348.965.1
Repeat the steps above, first computing a stressed d1 as

ln(348.965/380.06) + (0.0018 + 0.2192 /2) × 1.0

dˆ1 = √ = −0.27204,
0.219 × 1.0
1 Note that you could have applied an absolute shift of 3.4551 with the same effect.

Axioma 21
 then computing a stressed d2 as
√
dˆ2 = −0.27204 − 0.219 × 1.0 = −0.49104,

and finally computing the value of the derivative given these stressed parameters as

ĉ = 348.965Φ(−0.27204) − 380.06e−0.0018×1.0 Φ(−0.49104) = 18.82066.

Given these two values, we can use a finite difference to compute delta (∆) as

ĉ − c
∆Finite Difference = (41)
ˆ
S0 − S0
18.82066 − 17.49351
∆Finite Difference = = 0.384124. (42)
348.965 − 345.51
We can compare this value to the closed-form solution for ∆ from the Black-Scholes equation,

∆Black-Scholes = Φ(d1 ) (43)

∆Black-Scholes = Φ(−0.31748) = 0.375442.

Note that as the size of the instantaneous shift applied approaches zero, the value for ∆ obtained through
approximation converges to the true (closed-form) value.

Time Series Group

T

A time series group can also be shifted. For example, DV01 could be computed via a parallel shift of an
AF

interest rate curve with a corresponding effect on the value of a bond with two cash flows.
The parameters needed to price a bond that matures in two years and pays coupons annually are:
DR

N = 1000
C = 4%
R1 Year = 0.2%
R2 Year = 0.35%,

where N is the notional for the bond, C is its coupon, and Rt is the discount rate corresponding to each
cash-flow date.
The value of the bond can be computed as
n
X
PV = N e−Rt Ci + N e−Rn , (44)
t=1

which for this bond becomes

P V = 1000e−1×0.002 × 0.04 + 1000e−1×0.0035 × 0.04 + 1000e−1×0.0035 = 1072.667.

DV01 implies a 1bp absolute shift to the interest rate. Accordingly, when we compute it, R1 Year becomes
0.21% and R2 Year becomes 0.36%. Given these new values, we recompute the value for our bond and then
take the difference between the stressed value and the unstressed value as DV01.

PˆV = 1000e−1×0.0021 × 0.04 + 1000e−1×0.0036 × 0.04 + 1000e−1×0.0036 = 1072.455.

DV01, therefore, is 1072.455 - 1072.667, or -0.212. By convention, DV01 is often computed as − ∂V ∂R , and
would accordingly be 0.212 in this instance. The interpretation here should be that a 1bp shift in the interest
rate curve would yield a $0.21 loss on a notional of $1000.

Axioma 22
8.11 Horizon-Based Stress Tests
Time Series Group—Linear Horizon Stress
To illustrate the impact of a horizon-based stress test on the value of an instrument, consider a portfolio
of two bonds. Assume that coupon payments are made at intervals of exactly one-half of one year (i.e.,
Ti − Ti−1 = 0.5). Also assume that the coupon schedules of the bonds are three months apart. Each bond
matures after three coupons have been paid (i.e., at 1.5Y and 1.75Y).

Bond 1

N = 1000
C = 4%
R0.25 Year = 0.15%
R0.75 Year = 0.18%
R1.25 Year = 0.23%

Bond 2

N = 2000
C = 2.5%
R0.5 Year = 0.17%
R1 Year = 0.2%
R1.5 Year = 0.26%
T
AF

From equation (44), the PV of Bond 1 is 1116.945, and the PV of Bond 2 is 2141.878.
Apply a horizon-based stress test. For instance, stress interest rates up linearly by 200bp over the next
DR

year, and for later periods apply a 200bp flat stress.

The yield curve has six points defined:

R0.25 Year = 0.15%

R0.5 Year = 0.17%
R0.75 Year = 0.18%
R1 Year = 0.2%
R1.25 Year = 0.23%
R1.5 Year = 0.26%
(45)

Conveniently, the four points during which the horizon stress is applied are distributed at intervals of 25%
of the distance from the origin to the 1Y point on the yield curve. (The 1.25Y and 1.5Y points occur after
the interpolation scheme is applied.)
Because this is a linear scheme, use linear interpolation to apply the stress, as

(x − x0 )y1 − (x − x0 )y0
y = y0 + . (46)
x1 − x0
To compute the stress for the 0.25 year node, apply equation (46):

(0.25 − 0) × 200bp − (0.25 − 0) × 0bp

y =0+ = 50bp.
1−0
Apply this stress to our yield curve, and the points will be as follows:

Axioma 23
 R0.25 Year = 0.65%
R0.5 Year = 1.17%
R0.75 Year = 1.68%
R1 Year = 2.2%
R1.25 Year = 2.23%
R1.5 Year = 2.26%
(47)
Bond 1 would then have a PV of 1090.845, a loss of 26.1. Bond 2 would have a PV of 2080.29, a loss of
61.59.

Time Series Group—Exponential Horizon Stress

You can also stress over a horizon via exponential growth. First compute the exponential growth rate:
ln(1 + sh )
g= , (48)
h
where sh is the shift amount to be applied at horizon h.
Once the growth rate is computed, apply this stress at any time step t:
st = etg − 1. (49)

8.12 Defined Stress Scenarios

T
AF

This section lists the stress tests that Axioma Risk includes at install.
DR

Instantaneous Stresses
• Bond Spreads
• CDS
• Commodity
• Credit Markets
• Equity and Equity Future Markets
• Equity Implied Volatilities
• Equity Index
• Equity Markets
• Inflation
• Interest Rates
• IR Curve shift (Market data shock)
• IR Implied Volatilities
• Transitive 2 Indices (Rel. Date)
• Transitive 2 IR Key Nodes
• FX Shock
• FX Shock (USD)
• Volatility

Axioma 24
Horizon
• Transitive, Equity
• Transitive, Equity (Rel. Date)
• Transitive, Index

• Transitive, Index (Rel. Date)

• Transitive, IR Key Node
• Transitive, IR Key Node (Rel. Date)

• Transitive, Oil
• Transitive, Oil & Stock Index (Rel. Date)

Model Parameter
• Equity Markets (Risk model shock)
• IR Key Node

• Risk Resolution, Equity Factor

• Risk Resolution, Interest Rate Curve
T

• Risk Resolution, Two Interest Rate Curves

AF

• Transitive Risk Resolution, Equity Factor

DR

• Transitive Risk Resolution, Interest Rate Factor

• Transitive IR, RM Shock, Flat (Rel. Date)

Historical Stresses
• Historical
You can also edit these stress tests or create your own.

9 Risk Resolution
With its wealth of carefully designed models and its Risk Resolution functionality, Axioma Risk has the
unique ability to help you find the best analysis for a specific portfolio, a specific strategy, a specific economic
environment, and a specific institutional role.
Risk or portfolio managers who are interested in short-term risk measures on an actively traded portfolio
that may include bets on relative value trades typically use risk factors that are closely aligned with the
pricing factors. As the investment horizon increases, risk managers typically use less granular risk factors.
They may choose particular risk factors because of the way their institution makes investment decisions or
because they believe that certain fundamentals drive asset returns.
Instead of requiring you to decide on risk factors beforehand, Axioma Risk allows you to specify how risk
factors are related to pricing factors. For instance, you can choose to drive risk with fine-grained factors,
with one of the many Axioma risk models, or with econometric risk factors. This approach provides flexible,
nuanced control of risk dimensions.

Axioma 25
United States and Canada: +1 212-991-4500
Europe: +44 (0)20 7856 2424
Asia: +852-8203-27904
New York Atlanta Chicago
Axioma, Inc. Axioma, Inc. Axioma, Inc.
17 State Street 400 Northridge Road 2 N LaSalle Street
Suite 2700 Suite 850 Suite 1400
New York, NY 10004 Atlanta, GA 30350 Chicago, IL 60602
Phone: 212-991-4500 Phone: 678-672-5400 Phone: 312-448-3219
Fax: 212-991-4539 Fax: 678-672-5401 Fax: 877-385-8919
T
AF

Frankfurt Geneva Hong Kong

Axioma Deutschland GmbH Axioma CH Axioma, (HK) Ltd.
DR

Mainzer Landstrasse 41 Rue du Rhone 69, 2nd Floor Unit B, 17/F, Entertainment Bldg
D-60329 Frankfurt am Main 1207 Geneva, Switzerland 30 Queen’s Road Central
Germany Phone: +41-22-700-83-00 Hong Kong
Phone: +49-(0)-69-95-925-162 Phone: +852-8203-2790
Fax: +852-8203-2774

London Melbourne Paris

Axioma, (UK) Ltd. Axioma (AU) Ltd Axioma (FR)
30 Crown Place 31st Floor 19 Boulevard Malesherbes
London, EC2A 4EB 120 Collins Street 75008, Paris, France
Phone: +44-207-856-2424 Melbourne, VIC 3000 Phone: +33-(0)-1-55-27-38-38
Fax: +44-(0)-20-3006-8747 Australia
Phone: +61-(0)3-9225-5296

San Francisco Singapore Tokyo

Axioma, Inc. Axioma, (Asia) PTE Ltd. Axioma, Japan
201 Mission Street 30 Raffles Place Tekko Building 4F
Suite 2150 #23-00 Chevron House 1-8-2 Marunouchi, Chiyoda-ku
San Francisco, CA 94105 Singapore 048622 Tokyo 100-0005
Phone: 415-614-4170 Phone: +65-6233-6835 Japan
Fax: 415-614-4169 Fax: +65-6233-6891 Phone: +81-6870-7766

Sales: [email protected] Client Support: [email protected] Careers: [email protected]