Essential Linear Algebra for Finance 55

I.2.4 APPLICATIONS TO LINEAR PORTFOLIOS

To measure the market risk of a large portfolio it should first be mapped to its risk factors9 .
Copyright © 2008. Wiley. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or

The portfolio mapping depends on the type of assets in the portfolio. It is different for equity
portfolios compared with bond portfolios, and quite different again for portfolios containing
options. In Volume III we shall cover portfolio mapping separately for each asset class.
A linear portfolio is one whose return may be represented as a linear function of its
underlying risk factor returns. Portfolios with only cash, futures or forward positions are
linear. But options portfolios are non-linear. Bond portfolios are also non-linear functions
of interest rates; however, we shall see in Chapter III.1 that the cash flow representation of
a bond (or a loan, or indeed any interest rate sensitive portfolio) can be regarded as a linear
function of interest rates. The non-linear relationship between prices and interest rates is
captured in the risk factor sensitivities.
As we have not yet dealt with the intricacies of portfolio mapping we shall here only
consider linear portfolios that are small enough to be represented at the asset level. Then,
for the portfolio to be linear, its return must be a linear combination (i.e. a weighted sum) of
the returns on its constituent assets. In this case it is convenient to represent the portfolio’s
risk and return in matrix form. The remainder of this section shows how this is done.

I.2.4.1 Covariance and Correlation Matrices

The covariance matrix is a square, symmetric matrix of variance and covariances of an m × 1
vector of m returns, where the variances of the returns are displayed along the diagonal and
their covariances are displayed in the other elements.10 We write
⎛ 2 ⎞
1 12   1m
⎜ 21 22   2m ⎟
⎜ ⎟
⎜
V = ⎜ 31 32 2
3  3m ⎟
⎟ (I.2.19)
⎝       ⎠
m1    m2
We use the notation V for an arbitrary covariance matrix of returns, and in particular when
the returns are on individual assets. The notation  is reserved for a covariance matrix of
risk factor returns as used, for instance, in Chapters II.1, IV.2.
Since
⎛ 2 ⎞ ⎛ ⎞
1 12       1m 12 12 1 2       1m 1 m
⎜ 21
⎜ 22       2m ⎟ ⎜
⎟ ⎜ 21 2 1 22       2m 2 m ⎟
⎟
⎜ 31 32 32 ⎟ ⎜
   3m ⎟ = ⎜ 31 3 1 32 3 2 32    3m 3 m ⎟
⎜ ⎟
⎝               ⎠ ⎝        ⎠
m1          m2 m1 m 1    m2
any covariance matrix can also be expressed as
V = DCD (I.2.20)
applicable copyright law.

9
Only in this way we can attribute risk to different sources.
10
The variance, standard deviation, covariance and correlation operators are introduced formally in the next chapter, in Sections I.3.2
and I.3.4.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 1/11/2018 2:38 PM via UNIVERSIDAD DE LOS ANDES
AN: 323283 ; Alexander, Carol.; Market Risk Analysis, Quantitative Methods in Finance
Account: undeloan
 56 Quantitative Methods in Finance

where D is a diagonal matrix of returns standard deviations and C is the returns correlation
matrix. That is,
⎛ 2 ⎞ ⎛ ⎞⎛ ⎞⎛ ⎞
Copyright © 2008. Wiley. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or

1 12   1m 1 0   0 1 12   1m 1 0   0
⎜ 12  2   2m ⎟ ⎜ 0 2 0  0 ⎟ ⎜ 12 1   2m ⎟ ⎜ 0 2 0  0 ⎟
⎜ 2 ⎟ ⎜ ⎟⎜ ⎟⎜ ⎟
⎜      ⎟ ⎜ ⎟⎜     ⎟ ⎜ ⎟
⎜ ⎟=⎜ 0 0    ⎟⎜  ⎟⎜ 0 0    ⎟
⎝      ⎠ ⎝     0 ⎠⎝      ⎠⎝     0 ⎠
1m 2m   m2 0   0 m 1m 2m   1 0   0 m

Example I.2.11: Covariance and correlation matrices

Find the annual covariance matrix when three assets have the volatilities and correlations
shown in Table I.2.1.

Table I.2.1 Volatilities and correlations

Asset 1 volatility 20% Asset 1 – asset 2 correlation 08

Asset 2 volatility 10% Asset 1 – asset 3 correlation 05
Asset 3 volatility 15% Asset 3 – asset 2 correlation 03

Solution We have
⎛ ⎞ ⎛ ⎞
02 0 0 1 08 05
D=⎝ 0 01 0 ⎠  C = ⎝ 08 1 03 ⎠ 
0 0 015 05 03 1

So the annual covariance matrix DCD is

⎛ ⎞⎛ ⎞⎛ ⎞ ⎛ ⎞
02 0 0 1 08 05 02 0 0 004 0016 0015
⎝ 0 01 0 ⎠ ⎝ 08 1 03 ⎠ ⎝ 0 01 0 ⎠ = ⎝ 0016 001 00045 ⎠ 
0 0 015 05 03 1 0 0 015 0015 00045 00225

I.2.4.2 Portfolio Risk and Return in Matrix Notation

In Section I.1.4 we proved that the percentage return R on a linear portfolio is a weighted
sum of the percentage returns on its constituent assets, i.e.

k
R= wi Ri  (I.2.21)
i=1

It is convenient to write this relationship in matrix form. Let r denote the k × 1 vector of
asset returns and let w denote the k × 1 vector of portfolio weights. That is
applicable copyright law.

w = wt     wk  and r = R1      Rk  

Then the matrix equivalent of (I.2.21) is

R = w r (I.2.22)

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 1/11/2018 2:38 PM via UNIVERSIDAD DE LOS ANDES
AN: 323283 ; Alexander, Carol.; Market Risk Analysis, Quantitative Methods in Finance
Account: undeloan
 Essential Linear Algebra for Finance 57

The risk of an asset or a portfolio can be measured by its standard deviation, i.e. the
square root of its variance. Taking the variance of (I.2.21) and using the properties of the
variance operator (see Section I.3.2.6), we obtain the following expression for the variance
Copyright © 2008. Wiley. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or

of the portfolio returns:

k
k
k  
VR = w2i V Ri  + wi wj Cov Ri  Rj  (I.2.23)
i=1 i=1 j=1

Hence the variance of a portfolio is a quadratic function of the variances and covariances
on the constituent assets. However, (I.2.23) is a very cumbersome expression. It is much
more convenient to write the portfolio variance in matrix form and when we do this we see
that the portfolio variance is a quadratic form, with the vector being the vector of portfolio
weights and the matrix being the covariance matrix of the asset returns. That is,
VR = w Vw (I.2.24)
Note that (I.2.20) gives another way to write the portfolio variance, as
VR = x Cx x = Dw (I.2.25)
Thus the variance of the portfolio returns can be expressed as either
• a quadratic form of the covariance matrix and the portfolio weights vector w, or
• a quadratic form of the correlation matrix and the vector x where each weight is
multiplied by the standard deviation of that asset return.
The result will be the same, as shown in Example I.2.12 below. In this example we use the
term volatility. Volatility is the same as standard deviation, except it is quoted in annualized
terms. Under a constant weights assumption the volatility of the portfolio P&L is the returns
volatility times the current value of the portfolio.

Example I.2.12: Volatility of returns and volatility of P&L

Let the return on asset 1 have volatility 10% and the return on asset 2 have volatility 20%
and suppose the asset returns have a correlation of 0.5. Write down their correlation matrix
and their annual covariance matrix and hence calculate the volatility of the return and the
volatility of the P&L for a portfolio in which E2 million is invested in asset 1 and E3 million
is invested in asset 2.

Solution With these volatilities the annual variances of the assets are 0.01 and 0.04 and
their annual covariance is 05 × 01 × 02 = 001. Hence the annual covariance matrix is
   
001 001 1 1
V= = × 10−2 
001 004 1 4
 
04
The portfolio weights vector is w = and so the annual variance of the portfolio
06
return is
applicable copyright law.

    

  1 1 04 −2
  04
w Vw = 04 06 × 10 = 04 + 06 04 + 06 × 4 × 10−2
1 4 06 06
 
  04
= 1 28 × 10−2 = 04 + 28 × 06 × 10−2 = 208 × 10−2 
06

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 1/11/2018 2:38 PM via UNIVERSIDAD DE LOS ANDES
AN: 323283 ; Alexander, Carol.; Market Risk Analysis, Quantitative Methods in Finance
Account: undeloan
 58 Quantitative Methods in Finance

Equivalently we could use (I.2.25) with

     
04 × 01 004 1 05
x= = and C = 
06 × 02
Copyright © 2008. Wiley. All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or

012 05 1
Then     

  1 05 004   004
x Cx = 004 012 = 01 014
05 1 012 012
= 0004 + 00168 = 00208
Either way the portfolio return’s volatility is the square root of 0.0208, i.e. 14.42%, and the
portfolio’s P&L has volatility 01442 × E5 million = E721,000.

I.2.4.3 Positive Definiteness of Covariance and Correlation Matrices

A quadratic form x Vx represents the variance of a portfolio when V is a covariance matrix
and the nature of x is linked to that of V. For instance, if V is the covariance matrix of
percentage returns on the asset then x is a vector of portfolio weights, or if V is the covariance
matrix of absolute returns (i.e p  L) on risk factors then x is a vector of portfolio sensitivities
measured with respect to changes in risk factors. The relationship between x and V depends
on the asset type and whether we are measuring all the risk or just the ‘systematic’ risk,
i.e. the risk attributed to variations in risk factors. This complex relationship will be the
focus of many chapters in Volume IV.
The elements of x do not need to be positive, but the variance of every portfolio must be
positive. Hence the condition that all quadratic forms are positive merely states that every
portfolio, whatever the weights, must have positive variance. Obviously we should require
this. In other words, every covariance matrix should be positive definite.11
A covariance matrix is positive definite if and only if its associated correlation matrix is
positive definite. To see this, recall that we can write a covariance matrix of a set of returns
(or P&Ls) as
V = DCD (I.2.26)
where D is a diagonal matrix of standard deviations and C is the associated correlation
matrix. Then
w Vw = w DCDw = x Cx with x = Dw
That is, x is the vector whose ith element is wi multiplied by the standard deviation of the
ith asset return.
Since the standard deviations along the diagonal are all positive, w = 0 ⇒ x = 0. Hence,
w Vw > 0 for any w = 0 ⇒ x Cx > 0 for any x = 0
The converse implication can also be proved, on noting that
C = D−1 VD−1  (I.2.27)
In summary, the relationship V = DCD, where D is the diagonal matrix of standard deviations,
applicable copyright law.

implies that V is positive definite if and only if C is positive definite.12

11
Or at least we should require that all portfolios have non-negative variance. Hence the condition of positive definiteness that we
apply to covariance matrices is sometimes relaxed to that of positive semi-definiteness.
12
Note that D will always be positive definite since it is a diagonal matrix with positive elements on its diagonal.

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 1/11/2018 2:38 PM via UNIVERSIDAD DE LOS ANDES
AN: 323283 ; Alexander, Carol.; Market Risk Analysis, Quantitative Methods in Finance
Account: undeloan