Вы находитесь на странице: 1из 22

The Use and Abuse of

Tracking Error
Jason MacQueen

R-Squared Risk Management

Definition of Tracking Error


Tracking Error is defined as the standard deviation of
the differences (the relative returns) between a
portfolios returns and its benchmarks returns.
Tracking Variance is the square of the Tracking Error.
It is common practice in the investment management
industry to judge risk models by how closely their
ex ante tracking error forecast for a portfolio
corresponds to its ex post observed tracking error.
Typically, the ex ante value will be somewhat lower
than the ex post value.

R-Squared

Definition of Tracking Error


Let rt be a vector of the rates of return at time t with
mean vector E (rt +1 ) = r and covariance matrix r
Let the portfolio weights at time t be given by the
vector at and benchmark weights by vector bt.
We may then define the portfolio relative weights as :
wt = at - bt.

R-Squared

Ex Ante Tracking Error


Ex ante tracking error is calculated as follows :

TE =

R-Squared

var(a t rt b t rt )

(a t b t ) var(rt )(a t b t )

(a t b t ) r (a t b t ).

( w t ) r (w t ).

Ex Post Tracking Error


Ex post tracking error is calculated from the actual portfolio
relative returns rpt, where
rpt = wt-1rt
Hence, a time series calculation of ex post tracking error
would involve, over a period from t = 1, T, the terms
rp1, rp2, rpT, or equivalently,
w0r1 , w1r2 ,..., wT-1rT
The crucial point is that the portfolio weights wt, which are
assumed to be fixed, (or non-stochastic) ex ante, will vary (be
stochastic) ex post.
Ex post tracking error is therefore given by :
TE =

R-Squared

1
T
2
(
r

r
)
pt p
T 1 t =1

Stochastic Weights
Consider an investment strategy such that at t = 0, we fix
the weights w0.
For randomness in the weights w0 not to enter into the
calculation, we would require that each wt should be
rebalanced back to w0 for all periods from t = 1 to t = T.
However, barring the above special case, all other
investment strategies, including buy and hold, or
quarterly re-balancing, will involve the weights wt being
stochastic.
The same will apply (obviously) to capitalisationweighted strategies, and, of course, to all active
strategies.

R-Squared

Ex Ante vs Ex Post
The formulae for calculating ex ante and ex post
tracking error are different, and they make
different assumptions about whether the weights
are fixed or not
Ex Ante and Ex Post Tracking Errors are:
THEREFORE NOT DIRECTLY COMPARABLE
WITH EACH OTHER.

R-Squared

Active Management & Tracking Error - 1


Let e be a vector of ones, i.e. e = (1, 1, 1). For conventional
portfolio calculations, we have the portfolio budget constraint

ew t = 1
However, when we are computing portfolio values relative to a
benchmark, we have

ew t = 0.

Otherwise, the calculations are the same.


In the conventional calculation, weights wt are given at time t
and the portfolio return rpt +1 can therefore be written as:

rpt +1 = w t rt +1
R-Squared

Active Management & Tracking Error - 2


Portfolio tracking variance becomes, by definition :

var( rpt +1 ) = var( w t rt +1 )


= w t r w t
where r is the conditional (or unconditional)
covariance matrix of rt +1 , w t being treated as fixed,
and E (rt +1 ) = r is again being interpreted
conditionally or unconditionally.
We now propose an important theorem describing the
fundamental relationship between ex post and
ex ante tracking variance (and also tracking error).

R-Squared

Theorem
If a set of portfolio weights wt that satisfies ew t = 0
is stochastic, i.e., w t = w + t , where v t ~ (0, w ) ,
then the ex-post tracking variance of the difference
between portfolio returns and benchmark returns,
can be decomposed as follows :
2

TE SD = r w r + tr ( r w ) + w r w .

R-Squared

Proof of Theorem
We have:
2

TE SD = var(rpt +1 )
= var( E (rpt +1 | w t )) + E (var(rpt +1 ) | w t )
= var(wt r ) + E (wt r w t )
= var(wt r ) + E[( w + v t ) r ( w + v t )]

= var(wt r ) + E[ v t r v t + 2 v t r w + w r w ]
= r w r + tr ( r w ) + w r w

R-Squared

Remarks
In the special case of fixed (i.e. non-stochastic)
weights, we have w t = w , w = 0, and var(rpt +1 ) = wr w
This is the ex post tracking variance with fixed weights,
which will be directly comparable to the ex ante
forecast tracking variance.
Since ew t = 0 for all t, then we must have e w = 0 and
also var(ewt) = e we = 0
Lawton-Browne (2000) establishes empirically that
r w r is, indeed, very small in the cases she examines
(see Tables below).

R-Squared

Lawton-Browne Research
We now digress briefly to discuss Carola LawtonBrownes research.
She provides two estimates of r , one based on
individual stock returns, and the other based on
industry returns.
Her universe is an active UK portfolio of 187 stocks,
with a benchmark of the FT-SE 100.
In her paper, Lawton-Browne reports the following
tables.

R-Squared

Table 1 : TE results with stock returns


Component
Term

Actual
Variance

% Variance
contribution

% Standard
deviation

'r w r

0.168

1.6

0.41

tr ( r w )

4.847

45.8

2.20

'w r w

5.566

52.6

2.36

10.581

100.0

3.25

Total

R-Squared

Table 2 : TE results with industry returns


Component
Term

Actual
Variance

% Variance
contribution

% Standard
deviation

'r w r

0.091

2.9

0.30

tr ( r w )

1.971

63.0

1.40

'w r w

1.068

34.1

1.03

Total

3.130

100.0

1.77

R-Squared

Results of Lawton-Browne
These results show that, in this case at least, the
overall contribution from stochastic weights accounts
for between 45% to 65% of the ex post tracking error.
Furthermore, the contribution of
both cases, as anticipated.

R-Squared

'r w r

is tiny in

Conclusions - 1
Tracking error is becoming increasingly influential in
the investment management business. Sponsors of
defined benefit plans increasingly pay attention to risk
budgeting that purports to allocate an allowable
budget of tracking error for the whole plan across
managers of different asset classes; see Gupta,
Prajogi, & Stubbs (1999).
Another example is that last year Barclays Global
Investors agreed to return back a portion of its
management fee to the Sainsburys Pension Scheme if
the portfolio exceeded its agreed tracking error limits.

R-Squared

Conclusions - 2
If tracking error is used to help judge fund performance,
the crucial difference between ex ante and ex post
tracking errors described in this study must be taken into
account.
More generally, both portfolio managers and their clients
should take particular care to avoid confusing the two
different kinds of tracking error.
Finally, it should now be clear that risk models cannot be
judged on a spurious comparison of the ex ante tracking
error based on the initial relative weights - with the
observed ex post tracking error, which reflects, inter
alia, the changing relative weights through time.
Ex ante and Ex post tracking error are subtly different

R-Squared

References - 1
Baierl, G. T. and P. Chen, 2000, ChoosingManagers and
Funds, Journal of Portfolio Management 26(2), 47-53.
Gardner, D. and D. Bowie, M. Brooks, M. Cumberworth, 2000,
Predicted Tracking Errors : Fact or Fantasy? Faculty and
Institute of Actuaries, Investment Conference paper, 25-27
June 2000.
Grinold, R. and R. Kahn, 1995, Active Portfolio Management,
Irwin.
Gupta, F., R. Prajogi, and E. Stubbs, 1999, The Information
Ratio and Performance, Journal of Portfolio Management
(Q3 1999), 33-39.

R-Squared

References - 2
Larsen, G. A. and B. G. Resnick,, 1998, Empirical Insights on
Indexing, Journal of Portfolio Management 25(1), 51-60.
Lawton-Browne, C. L. 2000, Masters dissertation, Dept. of
Economics, Birkbeck College, London University
Markowitz, H. M., 1959, Portfolio Selection, 1st Edition, New
York: John Wiley & Sons.
Pope, P. and P. K. Yadav, 1994, Discovering Error in Tracking
Error, Journal of Portfolio Management (Q4 1994), 27-32.
Roll, R., 1992, A Mean/Variance Analysis of Tracking Error,
Journal of Portfolio Management, 13-22.
Sharpe, W, 1964, Capital Asset Prices : A Theory of Capital
Market Equilibrium under Conditions of Risk, Journal of
Finance 19, 425-442.

R-Squared

Summary
The essence of investment management is the
management of risks, not the management of
returns - Ben Graham
It is also often said that you can manage risks, but
you cant manage returns
Portfolio managers obviously need to come up with
expected returns, or forecasts of what will do well or
badly, or just pick stocks
But this is just the first step; the true skill in portfolio
management lies in managing its risk

R-Squared

20

Contact Information
R-Squared Risk Management Limited
The Nexus Building, Broadway, Letchworth Garden City,
Hertfordshire, SG6 3TA, United Kingdom
+44 1462 688 325 +44 7768 068 333
455 Lakeland Street, Grosse Pointe, MI 48230, U. S. A.
+1 313 469 9960 +1 646 280 9598
Email: info@rsqrm.com

R-Squared Risk Management

Вам также может понравиться