Strategy Benchmarks

From the Investment Managers Perspective

David E. Kuenzi
Vice President, Investment Analytics
Nuveen Investments
333 W. Wacker Drive
Chicago, IL 60606
[email protected]
Phone (312) 917-8037
Fax (312) 917-7961

July 2002

Abstract
As the investment management industry becomes more sophisticated, investment managers are using
benchmark indices in an increasingly complex fashionas the baseline along which the manager intends to
add value and manage risk, for determining which factor bets have most influenced overall portfolio returns
(attribution analysis), and for determining the extent to which the manager added value (through use of the
information ratio, for instance). Managers typically use readily available, published indices for these
purposes. For many managers, these published indices are inadequate for the simple reason that they often
do not accurately reflect the managers true investment universe. Use of published indices in this context
leads to a multitude of distortions and to poor information. A solution is the use of strategy benchmarks.
While much work has been done showing the benefits of strategy benchmarks from the perspective of the
client and consultant, this article shows that the use of a strategy benchmark is often crucial from the
investment managers perspective as well.

In past decades, the use of benchmark indices was quite limited. Managers simply compared their total
returns to those of some best fit broad market index, such as the S&P 500 for an equity manager, or the
Lehman Aggregate for a fixed income manager. During the 1990s, managers began to use indices much
more rigorouslyfor risk comparisons, determination of the consistency of investment manager returns,
portfolio risk management, and portfolio attribution analysis. During the last 10 years this shift in the use
of benchmark indices has led to extraordinary growth among index providers, including MSCI, Frank
Russell, Dow Jones, and the index groups at S&P, Salomon Smith Barney, Lehman Brothers, FTSE, and
Ryan Labs, as well as to increased interest in size / style indices. 1

While these published indices provide good benchmarks for many investment strategies, they are not able
to provide all that is needed for sophisticated investment managers employing specialized investment
strategies and requiring thorough analysis of their portfolios versus a benchmark. In these cases, the use of
customized or strategy benchmarks is appropriate. 2 Past work on this topic has focused mostly on client
needs as an impetus for the use of strategy benchmarks. In this paper, I will make a case for the use of a
strategy benchmark with an emphasis on the investment process. Specifically, I will show that a strategy
benchmark is necessary in order for the investment process to work efficiently in any situation in which the
managers strategy produces a universe of securities that differs by rule from available published indices. I
will also show the importance of a strategy benchmark to the integrity of more recently employed
performance measures, such as the information ratio.

Normal (or Neutral) Weights

There is a large body of work suggesting that a managers benchmark should represent normal or
neutral portfolio weights. Brinson, Hood, and Beebower [1986] and Brinson, Singer, and Beebower
[1991] suggest that average allocations over time represent normal portfolio weights, and more recently
Smith [2002] and Ryan [2001] build on this notion. Kritzman [1987] gives guidance for creating normal
portfolios and for considering where one draws the line between style and skill. Divecha and Grinold
[1989] emphasize the importance of normal portfolios in evaluating portfolio performance. Dietz and

Kirshman [1990] consider normal portfolios in the more general context of portfolio performance. Most
notable to this end, however, is the work of Bailey, Richards, and Tierney [1990] and Grinold and Kahn
[1995].

Bailey, Richards and Tierney present a framework such that a managers portfolio holdings, P, are
represented as:

P = B + ( P B) = B + A ,
where B is the benchmark, or the portfolios normal weight. In this formula A is the active positionthe
portfolio exposures, P, minus the benchmark exposures, B. Defining M as a market index (an available
published index) they write:

P = M + (B M ) + A = M + S + A,
where S is the style exposure of the manager, which is equal to the managers benchmark minus some
published index. Here they make a distinction between the normal weights, represented by B, and a
published index, represented by M. This difference is the manager strategy, S. This final equation:

P=M +S+ A

(1)

provides powerful intuitions concerning the relationships among active management, a managers strategy,
and a published index. The idea is simply that the only time a strategy benchmark is unnecessary is when B
= M, or when the managers normal weights are equal to those of a published benchmark, so that B M =
0 and P = M + A. If a managers strategy is such that the managers average exposures through time, B,
differ from those of a published benchmark, M, then it is crucial to create a strategy benchmark, rather than
using a published benchmark and just assuming that B = M.

Grinold and Kahn suggest the use of factor exposures for describing the normal portfolio. 3 The active
manager will likely always be taking on some active risk due to differences in the number of securities in
the index and the number of securities in the portfolio, thus making benchmark security allocations a
somewhat clumsy, albeit highly accurate, way of measuring neutral exposures. In the case of equities, this
would involve sector exposures as well as exposure to other factors, such as average P/E, average price

momentum, average volatility, etc. In the case of fixed income, this would involve exposure to sectors,
maturity categories, rating categories, duration, convexity, etc.

Both Baily [1992] and Baily, Richards, and Tierney [1990] support the notion that a valid benchmark
should be: 1) unambiguous, 2) investable, 3) measurable, 4) appropriate, 5) reflective of current investment
opinions, and 6) specified in advance. Most investors and investment managers gravitate to benchmarks
that generally have these characteristics, although items 4 and 5 are often interpreted loosely. 4 Bailey
suggests that a benchmark is appropriate if it is consistent with the managers investment style, and that a
benchmark is reflective of current investment opinions if the manager has current investment knowledge
of the securities that make up the benchmark. In order for a benchmark to work well, managers and
investors must interpret these requirements rigorously, and such interpretation will often point to the use of
a separate strategy benchmark for which the securities are truly reflective of neutral weights of the manager
universe.

The relationship between neutral weights from the managers universe and the published benchmark may
differ, however, based on the investment management discipline.

Two Types of Investment Disciplines

For the purposes of benchmarking, it is helpful to think in terms of two types of investment disciplines:

1.

Published benchmark centered (PBC) disciplines are those that begin with a well established broadbased benchmark as the goal and are managed closely to this index on an ongoing basis. Managers
targeting pension funds may be likely to have PBC disciplines. If, for instance, a pension fund
benchmarked its U.S. equity assets to the Russell 1000, it might hire two managers with PBC
disciplines in growth and value, respectively. Each manager would attempt to provide returns
commensurate with or in excess of those provided by the Russell 1000 Growth and Value indices,
respectively. These managers might be expected to turn over some of their portfolios at the annual

rebalancing of the Russell indices in June of each year in order to maintain exposures reflective of
those of their benchmark, as minimizing tracking error to these indices is a primary concern. Referring
to Equation (1), this would be a situation in which B = M so that S = B M = 0, and P = M + A.
Enhanced indexing would be the most defining example of such a strategy.

2.

Manager strategy (MS) disciplines are those that begin with an investment strategy that is meant to
take advantage of a market anomaly, to exploit a particular competitive advantage of the manager, or
to otherwise provide for risk / return characteristics that may not be identical to those of some wellestablished index. A strategy that invests in stocks with low debt and a near term reason for an
improvement in those companies fortunes would be an example. A manager of an MS discipline
might choose a best fit benchmark (the Russell 1000 Growth, say) as an afterthought, based on a best
fit with his or her strategy. Again, using Equation (1), it is clear that BM so that P = M + S + A.

The published benchmark centered discipline has no need of a strategy benchmark, as the method for
generating returns is subjugated to the goal of tracking the stated external benchmark. This portfolios risk
exposures (sector exposures, etc.) will only differ from, say, the Russell 1000 Growth insofar as the
manager thinks that such risk-taking will lead to outperformance vis --vis this index. In this instance, the
established external benchmark provides the fund manager with neutral weights.

Such is not the case with the manager strategy discipline. In this case, the published index may contain
securities that would never be found in the portfolio. In the examp le, for instance, stocks with high debt-tocapital might by rule be excluded from possible inclusion in the portfolio. As such, the external benchmark
does not represent neutral weights for the portfolio and is therefore a poor benchmark for the purposes of
gauging active risk, performance attribution, and insights into the portfolio management process. Over- or
underperformance to the external benchmark in a particular period may be simply a function of the longterm, static strategy of the manager. In this case, a strategy benchmark is appropriate.

In general, whenever the managers strategy is such that his or her universe of investable securities differs
from the closest fit published index components by a quantifiable rule, or by a style preference that can be
expressed with a quantifiable rule, then a strategy benchmark is appropriate. The remainder of this paper
will be applicable exclusively to those types of investment disciplines whose specialized strategies make
the use of a strategy benchmark appropriate. In order to establish why a strategy benchmark is crucial for
this type of investment discipline, it is first important to review the basic elements of an institutional
investment process.

General Framework of an Institutional Process

The investment process for a long-only, unleveraged, active institutional investor has a number of essential
elements, as represented in Exhibit 1. 5 The four basic elements are 1) investment policy, 2) generation of
alpha estimates, investment decision-making, and security / factor over- / under-weighting, 3) execution,
and 4) performance measurement and attribution.

The investment management group begins with an investment philosophy (how they intend to add value in
a fairly efficient market over the intermediate to long term), client goals, benchmarks and measures to
evaluate performance, and risk bands around benchmark exposures (the amount by which, say, a fixed
income portfolio could be longer or shorter than the duration of the benchmark). The investment manager
then makes active decisions on which securities and risk factors to overweight or underweight in order to
generate active returns. These decisions are then implemented and perhaps modified based on the ease
with which they can be executed. Finally, the investment management group performs attribution analysis
in order to determine whether they have been able to add value along the intended dimensions. This
feedback is then considered in the context of refinements to the investment processboth in the short- /
intermediate-term for refining tactical strategies, and over the long term for the determination of
appropriate policy.

In the sections that follow I will show that in order for this process to work well, the benchmark must be
reflective of appropriate neutral weights, and that for a published benchmark centered discipline, a strategy
benchmark is absolutely necessary.

Investment Policy and the Strategy Benchmark

Two of the most important aspects of investment policy (the first box in Exhibit 1) are the choice of the
benchmark and the risk controls around benchmark factor exposures. Grinold and Kahn note that the
client bears the benchmark risk, and the active manager bears the active risk of deviating from the
benchmark. When the manager and client identify the benchmark, they are identifying the general risk /
return characteristics that the client will expect over time. If the investment manager deviates from the
benchmark by policy, due to an investment strategy that specifically excludes a large portion of the
securities in the benchmark, then the client is not, on average, bearing benchmark risk. The client will be
exposed to a different set of risks, which may be ill defined from the clients perspective. Additionally, the
concept of risk controls becomes distorted if the manager employs a benchmark that is not representative of
true neutral weights. A strategy benchmark enables the manager to accurately identify the risk / return
characteristics that the investor can expect over time (which may be very different from those of the closest
fit published index), and to use risk controls more meaningfully.

An example shows this best. Suppose that an investment manager has a strategy of finding companies with
low debt (a high degree of flexibility to pursue profitable projects), and which the manager believes are
undervalued versus other similar firms. To this end, the manager starts with the S&P 500, removes the 350
stocks with the highest debt-to-capital ratio, and then performs bottom-up analysis on the remaining 150
companies to find those stocks most undervalued relative to other low debt-to-capital stocks in that sector.
(Note that the first step above could be easily applied in the creation of a strategy benchmark.) Lets
suppose further that the manager benchmarks against the S&P 500. This decision is based on the fact that
at inception of the product, say five years ago, 66% of the capitalization of the 150 stock portfolio was in
the S&P 500 / Barra Growth Index and 34% of the capitalization was in the S&P 500 Barra Value Index,

with the number of stocks in each divided approximately evenly. 6 Additionally, we assume that the
manager set sector risk bands at +/-7.5% of the benchmark.

Exhibit 2 provides some detail. Column A of the exhibit represents the sector weights that will be
understood to be the neutral weights in the agreement between the client and the managerthe weights of
the S&P 500 Index. Column B shows how much the investment manager will be able to deviate from each
of these index sector weights. Columns C and D show the minimum and maximum allocation that the
manager can have to any one of these sectors. Column E contains the capitalization weights of the strategy
universethe 150 low debt-to-capital stocks. These are the sector weights of the stocks that the manager
really has access to given the stated strategy. Column F is the difference between the strategy and the S&P
500. This column shows the built-in sector underweights or overweights of the strategy universe versus
the published benchmark.

The most stark item in column E is the zero weight in utilities. These stocks are patently excluded from
consideration for investment by the manager, yet they constitute 7.28% of the S&P 500 and thus have a
neutral weight of 7.28%. The manager will therefore, by policy, have a continual underweight in utilities by
nearly the maximum allowable underweight. By contrast, technology is only 12.83% of the index but
34.84% of the strategy. At a maximum overweight, the manager can invest only 20.33% of assets in
technology. Given that a good portion of the 150 stocks the manager is looking at are technology stocks
(52, to be exact), it is likely that there will be a systematic overweight (close to the 20% limit) in this
sector. Generally, with strategy overweights in energy, health care, and information technology, and
underweights in industrials, financials, and consumer-oriented stocks, the portfolio will likely have a higher
standard deviation than the S&P 500. While the risk bands in this instance serve to reduce risk vis --vis the
S&P 500, the investors ongoing exposure is very different than indicated by the S&P 500 weights.
Additionally, the risk bands will largely serve in shaping the true long-term sector weights rather than as
extreme boundaries around a neutral weight. The net result is that the investor is taking on more risk than
expected and the risk bands are not serving their purpose.

Alpha Generation and Risk Management

The argument in this section is a continuation of the discussion of risk controls. The active management
process represented in box number 2 of Exhibit 1 is largely a process of deciding which factor exposures to
overweight or underweight vis --vis the index. In the case of our example, the limited opportunity set
(only 150 stocks) will force the manager to the outside limits of many risk controls on an ongoing basis. So
as the manager attempts to decide which factors to over- or under-weight, the decision in many cases will
already have been made by risk controls set around an inappropriate benchmark. The question of whether
to overweight technology, for instance, will be answered in a constant fashion at each portfolio review.
Thus, active management is artificially constrained due to poor benchmark selection. This could be a
severely damaging outcome from an investment process perspective.

If, however, the manager constructed a strategy benchmark consisting of the 150 lowest debt-to-capital
stocks in the S&P 500 index, the neutral weights would be fully reflective of the managers true
opportunity set. 7 The weights and risk bands for such an index are shown in Exhibit 3. Using column A of
Exhibit 3 as neutral weights, and risk limits of the same size, +/-7.5%, the managers neutral exposures
would now be reflective of the securities that the manager has access to. The problem with utility stocks is
overcome, as the neutral weight in utilities is now zero. Technology, on the other hand, has a large neutral
allocation equal to, on average 34.84% of the portfolio. Implementing the risk bands, the manager can
invest anywhere from 27.34% to 42.34% in technology. Health care exposure is also greater, with a
maximum overweight of 29.84%. These riskier neutral weights are a direct result of the managers
strategyto buy companies with significant business risk and very little financial risk. If investors are not
comfortable with this level of equity risk, then they should consider a different strategy. If the investment
management team is not comfortable with this level of risk, then they should consider alterations to their
strategy. The solution is not, however, to use neutral weights, and therefore risk bands, that are not
compatible with the strategy. (There are, however, some instances in which the weights of excessively
large holdings or large sectors can and should be reduced in order to better reflect the strategy.) 8

It is clear, then, that the use of a strategy benchmark enables active management to take place in the
rational context of decisions to over- or underweight factor exposures versus the benchmark. Without a
strategy benchmark, such decisions often dont make sense, which can lead to a general breakdown in the
investment process. In this sense, a strategy benchmark can be a crucial element of a well-honed
investment process. Such decisions to over- or underweight particular factors or securities lead to the
execution phase of the investment process (box 3 of Exhibit 1). The critical issues here are 1) the extent to
which the ideas, information, and decisions generated in box two of Exhibit 1 are efficiently transmitted or
executed, and 2) the extent to which the factor exposures decided on in the previous step can be executed
with sufficient ease, given liquidity and other constraints. Finally, we get investment results.

Performance Attribution

The last box in Exhibit 1 contains one of the most important elements of the active management process
the stage at which managers engage in self-evaluation in order to understand whats working and what they
might improve. 9 Are the factor exposures taken on by the investment team paying off? Are the
quantitative models working right? Are the bottom-up analysts looking at the right information? Is the
team executing on this information in an optimal fashion? What might the investment management team
be missing and how should they correct any glitches? The integrity of attribution information is crucial if
the manager is to make sensible and timely adjustments to the activities which determine how and when
risks are taken or neutralized vis --vis the benchmark, and this requires a benchmark that represents true
neutral weights.

Attribution analysis provides a manager with sources of return. Returns-based attribution indicates the
amount of the total return that can be attributed to asset allocation, timing, and security selection, and is
used mainly by investors. Investment managers generally use holdings based attribution models. This
analysis indicates the amount of benchmark outperformance that is attributable to various factor
exposures the portion of the outperformance attributable to exposures listed in box number 2 of Exhibit 1.

10

As such, returns-based attribution is highly dependent on which benchmark is used. Exhibit 4 provides an
example, using the simple mechanics for attribution described Dietz and Kirschman [1990]. Suppose the
fund exposure to technology is 25%, an allocation that the manager looks at as being about average over
time. If the manager is using a non-representative index, then the attribution model will interpret the
weight difference between the portfolio and the benchmark as an overweight, thereby attributing
outperformance to this active decision. From the managers perspective, this is not an active decision at all,
and the attribution results based on technology exposure are fully anticipated each month. This colors any
results concerning the overall value-added of tactical sector allocation decisions and renders the attribution
analysis almost worthless. If, however, a strategy benchmark that accurately represents neutral is used, any
overweight or underweight to a particular factor is either 1) an active decision, 2) a calculated by-product
of an active decision, or 3) an unanticipated by-product of an active decision and thus a failure to
effectively manage risk. Attribution based on these overweights and underweights then provides valuable
information to the manager concerning active investment decisions and risk management.

Performance Comparisons, Tracking Error, and Information Ratio

The below analysis, which compares the strategy benchmark to two published indices, can be viewed in
two ways. First, to the extent that we expect the managers risk exposures to be reflected in the more
narrow strategy benchmark, it drives home the notion that direct comparison of the managers results with
inappropriate benchmarks provides little useful information. Second, it provides a blueprint for the type of
analysis that might be considered for evaluating a managers strategyseparate and apart from the
managers efforts to add value around that strategyto some published benchmark that may be of interest
to the investor.

Performance comparisons can be grossly distorted when using unrepresentative benchmarks. Looking back
on five years (60 months) of performance for our low debt-to-capital strategy, (April 1997 through March
2002), it is clear that style is a significant factor influencing returns and that poor benchmark selection can
provide for poor comparisons. Exhibit 5 shows the composition of the 150 stock strategy benchmark. The

11

Exhibit shows that over time approximately half of the stocks in the strategy were in the S&P 500/Barra
Growth Index, with the other half in the S&P 500/Barra Value Index, while about 68% of the capitalization
of the strategy has been in growth, with the remainder in value. Given this result, it seems that one might
choose either the S&P 500 or the S&P 500/Barra Growth Index as a benchmark. In Exhibit 6, we compare
the returns of our strategy benchmark, our 150 stock portfolio of low debt-to-capital stocks, rebalanced
quarterly, to the returns of both the S&P 500 and the S&P 500 / Barra Growth Index. Over the full 5-year
period, the strategy benchmark marginally outperforms both indices, by 8 and 12 basis points, respectively.
The first 36 months of the period represent the last three years of the technology bull market, while the
remaining 24 months represent the market decline. During this first three-year period the strategy
benchmark outperforms both indices, and the S&P 500 quite dramatically. This is reversed for both indices
during the subsequent 2-year period. This can be seen graphically in Exhibit 7. Assuming that the
managers portfolio tracked closely with the strategy, it is clear that comparing performance to the S&P
500, and even to what seems to be a better-fitting S&P 500 / Barra Growth Index, 10 would likely have
given poor signals as to when the manager was adding value. On the simple basis of total return
comparisons, the manager would have looked quite good at the market peak.

Results are similar when we adjust these returns for risk. Exhibit 8 contains the returns, standard
deviations, and Sharpe ratios11 for the two benchmarks and the strategy for both the longer period and the
two sub-periods. Here again, given the results of the strategy benchmark, it is likely that the manager
would have shown a strong risk-adjusted return at the markets peak relative to the published indices.

Now lets consider value-added against the benchmarks. Exhibit 9 shows the beta, alpha, total tracking
error, residual tracking error, and information ratio for the strategy benchmark against the two published
indices for both the full period and the two sub-periods. 12 The first row of this table shows beta
coefficients. Note that during the first three years the betas are close to 1.0 and then shoot upwards to 1.41
and 1.14 during the last 24 months. Using an inappropriate benchmark might lead an investor to determine
that, in order to boost returns during the market downturn, the manager began taking on more market
timing risk by increasing the portfolios beta, when in actuality this shift in beta is purely a function of a

12

passive set of rules employed by the manager. Secondly, the strategy shows a strong alphaespecially
against the S&P 500during the first three years. For an investor comparing this manager to the S&P, the
peak in the market may have seemed an ideal time to increase exposure to this manager.

Perhaps most important is the tracking error. Both total tracking error and residual tracking error are very
large. Using an inappropriate benchmark leads to built-in tracking error, which will in turn distort the
information ratio and make evaluation of a manager much more difficult. Barton Waring, in a Journal of
Investing debate on this subject (Belden and Waring [2001]), notes that using benchmarks that represent
the managers normal as closely as possible has a solid basis in theory. We want the simple, investable
benchmark to explain in a statistical sense as much of the managers behavior as possibleIf part of the
investment result that could have been explained with a more general benchmark is left in the residual, it
will add tracking errorthat obscures the managers true trail. Our work here provides a stark example of
this phenomenon. All of the tracking error shown in the Exhibit 9 could be removed by using the strategy
benchmark rather than an inappropriate published benchmark.

Tracking error is also important when considering the calculation of the information ratio. The information
ratio is calculated as:

IR =

(2)

where is the beta-adjusted benchmark outperformance and is the residual risk (the portion of tracking
error unrelated to benchmark timing risk). The information ratio is being used more and more frequently to
evaluate both manager skill and proposed investment strategies. Equation (2) shows us that the lower the
residual risk, , all other things being equal, the higher the information ratio. Built in residual risk through
the use of an inappropriate benchmark could have the impact of reducing the information ratio of a very
skilled manageran unfavorable outcome from both the managers and the investors perspectives.
Additionally, all else being equal, a manager outperforms if is higher. If a strategy contains style risk
against an inappropriate benchmark, then a favorable environment for that style can lead to a high
information ratio when the supposed outperformance is really just a function of the managers peculiar

13

style being in favor. This is clearly the case in Exhibit 9. The information ratio of the strategy benchmark
is a significant 1.03 against the S&P 500 during the first three years but reverses significantly during the
subsequent two-year period. Goodwins research [1998] supports this conclusion. He found that
information ratios among managers differed by manager style. His final conclusion is that you should
always be cautious in interpreting a published information ratio, and you should discount any that uses an
inappropriate benchmark.

Overall the above analysis shows that this particular strategy is not at all neutral to either published
benchmark. The strategy has its own style that makes it perform differently. Given that manager style
drift tends to occur in the direction of ouperforming styles (see Arrington [2000]), another interpretation of
the Exhibit 9 is that the manager has engaged in significant style drift in order to boost returns. This
confusion can be broadly preempted through the use of a strategy benchmark. If the manager or investor
were to perform the above analysis using the strategy benchmark as the benchmark, the fit with the
portfolio would be extraordinarily high and the value-added of the managerabove and beyond a simple
rule for narrowing down the broader universe of stockswould be much more clear.

Conclusion

I have shown that in order for a benchmark to provide the investment manager with the qualities necessary
for the implementation of a robust investment process, the benchmark must represent neutral or normal
weights to both security and factor exposures. When there is no published index reflective of the
managers neutral, the manager must develop a strategy benchmark, either internally or with the help of a
third party. This strategy benchmark enables the investment process to maintain integrity with regard to:
mean sector / factor exposures, the use of risk bands around sector / factor exposures, the process of overand under-weighting exposures, and attribution analysis. Without a strategy benchmark, many investment
processes will tend to operate sub-optimally, as the neutral weights, ris k bands, over-weighting, and
attribution are skewed by differences between the chosen benchmark and the essential characteristics of the
investment strategy. Additionally, I have shown that some of the most useful measurestracking error and

14

information ratiocan give spurious results if the benchmark used in the analysis is not truly representative
of the managers peculiar style.

Where previous work on this subject suggests that it is mainly in the investors interests to use a strategy
benchmark, the above shows that it is equally and perhaps more important from the investment managers
perspective. Investment managers who are able to rigorously implement the framework detailed above
around an appropriate benchmark are well positioned to continuously improve their investment processes
based on the ever-shifting nature of the capital markets, and to thus have a higher probability of providing
investors with superior results.

15

ENDNOTES

Many of these groups also create strategy benchmarks for clients on a customized basis.

While many practitioners use the term custom benchmark, the term strategy benchmark is

used hereto emphasize its use in relation to a managers peculiar strategy and universe of securities.
3

Grinold and Kahn define the managers active position as hPA = h P hB , where h P and h B are

vectors of security weights in the portfolio and the benchmark, respectively, and h PA is the active position.
If the manager has no information, then the appropriate exposures are those of the benchmark (h P = h B ); if
the manager does have information, then h P hB .
4

To the extent that the portfolio is large and market impact in trading is a concern, the

requirement that an index be investable is also often loosely interpreted, as it may be difficult for the
manager to invest in the index due to the small float of some stocks. In such cases, float-weighting of
strategy benchmarks should be considered.
5

These essential elements can be seen both in textbook literature and in practice. See, for

instance, Maginn and Tuttle [1990] in the introductory chapter of their book, Managing Investment
Portfolios, Grinold and Kahn [1995] (their book is organized along the concepts in boxes 2, 3, and 4 of
Exhibit 1, and they explicitly assume that the items in box 1 have been established), and Agache [2001].
Thanks go to members of Nuveen Investments municipal research for useful input in the construction of
this graphic.
6

The S&P 500 / Barra Growth and Value indices are derived by simply dividing the total

capitalization of the S&P 500 equally into two bucketsthe first containing the stocks with higher price-tobook values and the second containing stocks with low price-to-book values.
7

For simplicity, we assume full quarterly rebalancing and capitalization-weighting of the stocks.

The reweighting of individual holdings is particularly applicable for concentrated strategy

benchmarks used by mutual fund managers. Managers must maintain compliance with regulated
investment company (RIC) guidelines in order to avoid double taxation. Such guidelines require that no
individual holding be greater than 25% of total assets and that no three holdings make up more than 50%.

16

With respect to the remaining 50% of the portfolio, no holding can be greater than 5% of total assets. If the
strategy benchmark violates these guidelines, then the benchmark is not investable and its weightings
must be modified. This can be accomplished by using iterative methods to reduce the weights of the largest
holdings and to then spread excess weight across the remainder of the portfolio on a cap-weighted basis.
These or similar processes can be used to reduce sector weights or to otherwise hone the strategy
benchmark. Managers who tend to equally weight their portfolios, and hence have size bias, should refer to
Divecha and Grinold [1989] for guidance.
9

While this analysis is also important fromthe perspectives of clients and consultants, this idea is

well developed in Baily, Richards, and Tierney; here, we choose to focus on its importance to the
investment manager.
10

Despite the inclusion of many value stocks in the strategy benchmark, the portfolio likely

tracks better against the growth index due to its strategy of choosing companies with low financial risk and
higher business risk. Value companies with this quality are likely to behave more like growth companies
than would the overall value universe.
11

Sharpe ratios for negative excess returns are not meaningful; in this case, the lower the excess

return, the higher the Sharpe ratio.

I use the notation and definitions of Grinold and Kahn. Beta is defined as the slope of the

regression rp(t) = p + p rB (t) + p . Alpha is defined from this equation as well; assuming that p is, on
average, equal to zero, we get p = rp (t) - p rB(t). Total tracking error, p , is defined as std{ rp (t) - rB (t)},
and residual tracking error, p , as std{rp (t) - p rB (t)}. The information ratio, IR, is then defined as p / p .

17

REFERENCES

Agache, Kristof. The European Equity Investment Process. Journal of Investing, Winter 2001, pp. 1723.

Arrington, George R. Chasing Performance Through Style Drift. Journal of Investing, Summer 2000,
pp. 13-17.

Bailey, Jeffery V. Are Manager Universes Acceptable Performance Benchmarks? Journal of Portfolio
Management, Spring 1992, pp. 9-13.

Bailey, Jeffery V., Thomas M. Richards, and David E. Tierney. Benchmark Portfolios and the Manager /
Plan Sponsor Relationship. In Frank J. Fabozzi and T. Dessa Fabozzi, eds., Current Topics in Investment
Management. Harper Collins, 1990, pp. 349-363.

Belden, Susan, and M. Barton Waring. Compared to What? A Debate on Picking Benchmarks Journal
of Investing, Winter 2001, pp. 66-72.

Brinson, Gary P., L. Randolph Hood, and Gilbert L. Beebower. Determinants of Portfolio Performance.
Financial Analysts Journal, July / August 1986, pp. 39-44.

Brinson, Gary P., Brian D. Singer, and Gilbert L. Beebower. Determinants of Portfolio Performance II:
An Update. Financial Analysts Journal, May / June 1991, pp. 40-48.

Dietz, Peter O., and Jeannette R. Kirschman. Evaluating Portfolio Performance. In John L. Maginn and
Donald L. Tuttle, eds., Managing Investment Portfolios. Warren, Gorham & Lamont, 1990, pp. 14.114.58.

18

Divecha, Arjun, and Richard C. Grinold. Normal Portfolios: Issues for Sponsors, Managers and
Consultants. Financial Analysts Journal, March/April 1989, pp. 7-13.

Goodwin, Thomas H. The Information Ratio. Financial Analysts Journal, July / August 1998, pp. 34-43.

Grinold, Richard C., and Ronald N. Kahn. Active Portfolio Management, Chicago, IL: Richard D. Irwin,
1995.

Kritzman, Mark. How to Build a Normal Portfolio in Three Easy Steps. Journal of Portfolio
Management, Summer 1987, pp. 21-23.

Maginn, John L. and Donald L. Tuttle. The Portfolio Management Process and Its Dynamics. In John L.
Maginn and Donald L. Tuttle, eds., Managing Investment Portfolios. Warren, Gorham & Lamont, 1990,
pp. 1.1-1.11.

Ryan, Timothy P. Separating the Impact of Portfolio Management Decisions. Journal of Performance
Measurement, Fall 2001, pp. 29-40.

Smith, Paul. Process AttributionMeasuring the Performance of the Investment Process. Journal of
Performance Measurement, Winter 2001/2002, pp. 21-28.

19

EXHIBIT 1
GENERAL INSTITUTIONAL INVESTMENT PROCESS FLOW

Set Investment Policy

Investment philosophy
Client goals / investment objectives
Benchmarks and measures
Risk tolerance

Investment Analysis:
Determine Risk Exposures

2
Equities:

Fixed Income:

Security selection
Sector selection
Size distribution
PE exposure
Div yield exposure
Momentum
exposure
Etc.

Security selection
Sector selection
Structure/Convexity
Yield curve
positioning
Duration
Rating allocation
Etc.

Implement Strategy and Execute

Buy/sell/hold decision
Market condition modifications
Cash management

Measure Results

Attribution analysis
Benchmark comparisons
Reassess strategy, data, tools,
decision making

20

F
e
e
d
b
a
c
k

L
o
o
p

EXHIBIT 2
COMPARISON OF PUBLISHED INDEX SECTOR WEIGHTS / RISK
CONTROLS WITH THOSE IMPLIED BY THE STRATEGY

Energy
Materials
Industrials
Consumer Discrt
Consumer Stpls
Health Care
Financials
Info Technology
Telecomm Svcs
Utilities

(A)
S&P 500
Weights
3.57%
5.78%
11.80%
15.27%
11.65%
11.60%
17.02%
12.83%
3.20%
7.28%

(B)
Risk
Limits
+/- 7.5%
+/- 7.5%
+/- 7.5%
+/- 7.5%
+/- 7.5%
+/- 7.5%
+/- 7.5%
+/- 7.5%
+/- 7.5%
+/- 7.5%

(C)
Maximum
Exposure
11.07%
13.28%
19.30%
22.77%
19.15%
19.10%
24.52%
20.33%
10.70%
14.78%

21

(D)
Minimum
Exposure
0.00%
0.00%
4.30%
7.77%
4.15%
4.10%
9.52%
5.33%
0.00%
0.00%

(E)
Strategy
Weight
12.15%
1.68%
4.55%
11.75%
6.70%
22.34%
5.07%
34.84%
0.92%
0.00%

(F)
Difference
Strat-S&P
8.58%
-4.10%
-7.25%
-3.52%
-4.95%
10.74%
-11.95%
22.00%
-2.28%
-7.28%

EXHIBIT 3
STRATEGY NEUTRAL WEIGHTS, RISK CONTROLS,
AND NUMBER OF STOCKS REPRESENTED

(A)

(B)

(C)

(D)

(E)

Strategy
Benchmark
12.15%

Risk
Limits
+/- 7.5%

Maximum
Exposure
19.65%

Minimum
Exposure
4.65%

Number
of Stocks
5

Materials
Industrials

1.68%
4.55%

+/- 7.5%
+/- 7.5%

9.18%
12.05%

0.00%
0.00%

6
20

Consumer Discrt
Consumer Stpls
Health Care

11.75%
6.70%
22.34%

+/- 7.5%
+/- 7.5%
+/- 7.5%

19.25%
14.20%
29.84%

4.25%
0.00%
14.84%

19
6
23

Financials
Info Technology
Telecomm Svcs

5.07%
34.84%
0.92%

+/- 7.5%
+/- 7.5%
+/- 7.5%

12.57%
42.34%
8.42%

0.00%
27.34%
0.00%

18
52
1

Utilities

0.00%

+/- 7.5%

7.50%

0.00%

Energy

22

EXHIBIT 4
ATTRIBUTION EXAMPLE USING A
PUBLISHED AND A STRATEGY BENCHMARK

Using
Using
Published Strategy
Benchmark Benchmark
Outperformance of
Technology Sector
Portfolio Technology
Exposure
Benchmark Technology
Exposure
Overweight (Portfolio
Exposure - Index
Exposure)
Outperformance
Attributable to
Technology Exposure*

10%

10%

25%

25%

15%

25%

10%

0%

1%

0%

*Calculated as Overweight x Outperformance

23

EXHIBIT 5
STYLE COMPOSITION OF STRATEGY BENCHMARKUSING
S&P / BARRA GROWTH AND VALUE INDICES

75%
70%
65%
60%
55%
50%
45%

Growth--% of Capitalization

Growth--% of Companies

24

Nov-01

Jul-01

Mar-01

Nov-00

Jul-00

Mar-00

Nov-99

Jul-99

Mar-99

Nov-98

Jul-98

Mar-98

Nov-97

Jul-97

Mar-97

40%

EXHIBIT 6
RAW RETURNS AND OUTPERFORMANCE OF STRATEG Y
BENCHMARK AGAINST PUBLISHED INDEXES

Full Period 4/97-3/02

Annualized

Ttl Return

Ttl Return vs S&P/

Ttl Return
10.27%
37.78%

vs S&P 500
0.08%
10.36%

Barra Growth
0.12%
2.02%

-21.05%

-9.66%

-1.55%

25

EXHIBIT 7
RETURN COMPARISONSS&P 500, S&P/BARRA GROWTH,
AND STRATEGY BENCHMARK

280

240
220
200
180
160
140
120
100
M
ar97
Ju
l-9
7
No
v-9
M 7
ar
-9
8
Ju
l-9
8
No
v-9
M 8
ar99
Ju
l-9
9
No
v9
M 9
ar
-0
0
Ju
l-0
0
No
v-0
M 0
ar
-0
1
Ju
l-0
1
No
v-0
M 1
ar
-0
2

Value of $100 Invested

260

S&P 500

SP/Barra Growth

26

Strategy

EXHIBIT 8
RETURN, STANDARD DEVIATION, AND SHARPE RATIO COMPARISONS&P 500,
S&P/BARRA GROWTH, AND STRATEGY BENCHMARK

(Annualized from
Monthly Data)

Full Period 4/97-3/02

3Years 4/97-3/00
SP500
Growth Strategy

Geometric Ttl Return

10.19%
17.72%
0.31

27.43%
17.43%
1.07

10.15%
20.55%
0.24

10.27%
22.69%
0.24

27

35.76%
20.89%
1.29

37.78%
19.42%
1.69

2 Years 4/00-3/02
SP500
Growth Strategy
-11.39%
16.52%
NM

-19.50%
20.89%
NM

-21.05%
24.28%
NM

EXHIBIT 9
RISK, RETURN, AND OUTPERFORMANCE STATISTICS FOR STRATEGY
BENCHMARK AGAINST S&P 500 AND S&P/BARRA GROWTH INDICES

Beta

Full Prd 4/97-3/02

3 Yrs 4/97-3/00
Vs 500 Vs Growth
1.04
1.03

2 Yrs 4/00-3/02
Vs 500 Vs Growth
1.41
1.14

Alpha
Total Tracking Error
Residual Rtrn Error

-0.97%
8.55%
7.75%

-0.27%
5.31%
5.08%

7.27%
7.12%
7.09%

0.80%
5.15%
5.13%

-4.68%
9.64%
6.82%

1.69%
5.61%
4.82%

-0.13

-0.05

1.03

0.16

-0.69

0.35

Information Ratio

28