RiskMetrics⢠âTechnical Document
RiskMetrics⢠âTechnical Document
RiskMetrics⢠âTechnical Document
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
72 Chapter 4. Statistical and probability foundations<br />
We can model each underlying return as a random walk that is similar to Eq. [4.17]. This yields<br />
[4.48a]<br />
[4.48b]<br />
[4.48c]<br />
,<br />
= µ 1<br />
+ σ 1 t<br />
r 1 t<br />
r 2 t<br />
,<br />
= µ 2<br />
+ σ 2 t<br />
r 3 t<br />
,<br />
= µ 2<br />
+ σ 3 t<br />
,<br />
ε 1,<br />
t<br />
,<br />
ε 2,<br />
t<br />
,<br />
ε 3,<br />
t<br />
Now, since we have three variables we must account for their movements relative to one another.<br />
These movements are captured by pairwise correlations. That is, we define measures that quantify<br />
the linear association between each pair of returns. Assuming that the ε t ’s are multivariate normally<br />
(MVN) distributed we have the model<br />
⎛<br />
⎞<br />
ε 1,<br />
t ⎜ 0<br />
1 ρ 12, t<br />
ρ 13,<br />
t ⎟<br />
[4.49] ε 2,<br />
t<br />
∼ MVN<br />
⎜<br />
0 ,<br />
⎟<br />
⎜ ρ 21, t<br />
1 ρ 23,<br />
t ⎟ , or more succinctly, ε∼<br />
MVN ( µ t<br />
, R t<br />
)<br />
⎜<br />
ε 3, t ⎝<br />
0<br />
⎟<br />
ρ 31, t<br />
ρ 32, t<br />
1 ⎠<br />
where parameter matrix R t<br />
represents the correlation matrix of ( ε 1, t<br />
, ε 2, t<br />
, ε 3,<br />
t<br />
) . Therefore, if we<br />
apply the assumptions behind Eq. [4.49] (that the sum of MVN random variables is normal) to the<br />
portfolio return Eq. [4.47], we know that r pt is normally distributed with mean µ p,t and variance<br />
2<br />
. The formulae for the mean and variance are<br />
σ pt ,<br />
[4.50a]<br />
µ pt<br />
,<br />
= w 1<br />
µ 1<br />
+ w 2<br />
µ 2<br />
+ w 3<br />
µ 3<br />
[4.50b]<br />
2<br />
σ pt ,<br />
=<br />
2 2<br />
w 1σpt<br />
,<br />
2 2<br />
w 2σpt<br />
,<br />
2 2<br />
w 3σpt<br />
,<br />
+ + + 2w 1<br />
w 2<br />
σ 12,<br />
t<br />
+ 2w 1<br />
w 3<br />
σ 13,<br />
t<br />
+<br />
2<br />
2<br />
2<br />
2w 2<br />
w 3<br />
σ 23,<br />
t<br />
2<br />
σ ij t<br />
where the terms ,<br />
represent the covariance between returns i and j. In general, these results<br />
hold for ( N ≥ 1 ) underlying returns. Since the underlying returns are distributed conditionally<br />
multivariate normal, the portfolio return is univariate normal with a mean and variance that are<br />
simple functions of the underlying portfolio weights, variances and covariances.<br />
4.5.3 The lognormal distribution<br />
In Section 4.2.1 we claimed that if log price changes are normally distributed, then price, , conditional<br />
on P t – 1<br />
is lognormally distributed. This statement implies that P t<br />
, given P t – 1<br />
, is drawn<br />
from the probability density function<br />
P t<br />
[4.51]<br />
f<br />
( P t<br />
)<br />
=<br />
1 –( lnP ---------------------------<br />
t – 1<br />
– µ ) 2<br />
exp -------------------------------------- P<br />
2<br />
t 1<br />
><br />
P t – 1<br />
σ t<br />
2π 2σ t<br />
–<br />
0<br />
where<br />
[4.52]<br />
[4.53]<br />
P t<br />
follows a lognormal distribution with a mean and variance given by<br />
E[ P t<br />
] =<br />
⎛ 2⎞<br />
exp⎝µ + 5σ t ⎠<br />
V ( P t<br />
) =<br />
⎛ 2⎞<br />
⎛ 2⎞<br />
exp2µ t<br />
⋅ exp⎝2σ t ⎠–exp⎝σ t ⎠<br />
RiskMetrics —Technical <strong>Document</strong><br />
Fourth Edition