getting the drift - northfield · the first period after a style shift from value to growth...

1

Getting the Drift Ron Surz, President, PPCA Inc.

Dan Thatcher, President, Thatcher Resources 1/12/01

Summary

With knowledge and awareness of style drifts, consultants can be proactive in protecting their clients’ assets, rather than reactive to the harm after it is done. Point-in-time attribution analysis incorporating style was not available a few years ago, so consultants could not be proactive, but today they can be. “I didn’t know” is not an acceptable explanation to the client anymore. With the new tools that are now available, consultants can help their clients much more today than they could a few years ago. Get the drift?

Style Drift

Growth stock investing had been in favor through the first quarter of last year, but that all changed with the recent correction, especially the correction in Technology stocks. We’re all reading now about some investment managers losing business because they succumbed to the resurgence of growth stocks at precisely the wrong time. Many of these unfortunates felt they had to make the leap into growth to stay in business, but the real unfortunates are the clients. Style drift is a serious problem for clients because it distorts asset allocation and undermines performance when styles rotate. Value managers who have drifted over the past three years toward more favored growth stocks are regretting those moves recently, but not as much as their clients. Most sophisticated investors hire style specialists and, most importantly, design their policies around style stability. To guard against the deleterious effects of style drift, consultants need to become more proactive, rather than reactive, to the problem. They need to protect their clients. Firing a manager after the harm is done does not bring back the loss.

2

Being Proactive A consultant’s foremost responsibility is to protect a client's assets. It’s the “do no harm” counterpart to the medical doctor’s creed. To provide this protection a consultant must be diligently aware of the current and prospective consequences of an investment manager's actions. This due diligence is best accomplished with point-in-time attribution analysis, which can provide insights into: • Current allocation across styles and economic sectors − Do the present allocations

to styles conform to the manager’s style mandate? What economic sectors are being emphasized or de-emphasized relative to the style mandate?

• Skill in picking stocks and sectors − Is the investment manager selecting superior stocks? Superior industries? Why is he adding or subtracting value?

• Sources of investment return − The market, the manager’s style, his individual stock picks, and her industry concentrations all contribute to performance. How much each contributes is an indication of the manager’s strengths and weaknesses.

• Success or failure − Performance is evaluated relative to the manager’s assignment, which is usually style-related. Care must be exercised to evaluate skill and not style. A manager whose style is in favor should beat his style benchmark, not just the market.

• Style drift − Is the manager’s style profile similar to what it was in the previous quarter? Is it consistent with the client’s policy for this manager? If not, why not?

Point-in-time attribution analysis incorporating style is to financial consulting what early detection testing is to medicine. Despite the superiority of point-in-time analysis, the most popular approach currently used to detect style drift is returns-based style analysis, which uses mathematical regression of investment returns to estimate the manager's effective style mix. Since returns-based style analysis requires only returns, it can do a lot with very little, but the price paid for this convenience is a significant delay in identifying style shifts. For example, consider a returns-based style analysis that uses 30 observations to ascertain style fit. The first period after a style shift from value to growth produces 29 weightings of value versus one weighting of growth (29/1). The next period will provide weightings of 28/2, the next 27/3, and so forth. As you can see, identifying style drift

3

using returns-based style analysis takes a long time. Consequently, this method can only be reactive. Waiting for performance to deteriorate and then confirming that style drift is the culprit means we are too late. The damage has already been done. Point-in-time style analysis, on the other hand, requires portfolio holdings. This approach takes more time and effort to set up, but the needed information is readily available. The reward for making these additional data entries is immediate identification of style drifts and sector changes and, just as importantly, evaluation of the sources of value added by the manager, enabling consultants to be proactive. For these reasons, Dr. William F. Sharpe, Nobel economics laureate and creator of returns-based style analysis, has stated that point-in-time style analysis is superior to the method he created. This isn’t to say that point-in-time analysis should always be used instead of returns-based analysis. Returns-based analysis is very good at capturing the history of on-average effective style allocation. The resulting style profile is usually quite good for establishing the benchmark in point-in-time attribution. In other words, the point-in-time backdrop should rarely be a standard off-the-shelf index, but rather ought to be a blend of style indexes derived from a returns-based regression analysis. The returns-based analysis establishes the norm against which point-in-time attribution measures value added or subtracted. In this way, the two methods are in fact complementary, and both should be used. Continuing with the medical analogies, returns-based analysis establishes the acceptable range for the test, like your blood pressure, while point-in-time analysis gives you your current reading.

Stock Classifications But, you may say, "It isn’t easy to classify stocks into styles, and these classifications probably change over time", and you’d be right. In the November 1999 issue of Senior Consultant, one such classification method was described in "Incorporating Style in Attribution Analysis”. The appendix to the November article, which details classification rules, is repeated at the end of this article. There are certainly alternative schemes, and the industry may never agree on one single classification methodology, but similar concerns haven’t stopped us in the past from using economic sectors, such

4

as Health Care and Technology. The point is: We need to start doing something that makes sense and to perfect it as we learn. We encourage you to revisit the November article and to peruse the following examples of how some S&P500 stocks were classified at the beginning of the third quarter of 2000. Sample Security Classifications Large Value Large Core Large Growth BANK OF AMERICA AMERICAN INTL GRP AMERICA ONLINE ANHEUSER-BUSCH GENERAL ELECTRIC CISCO SYS CHEVRON MICROSOFT DISNEY CO PHILIP MORRRIS PFIZER INC AT&T YAHOO Mid-Cap Value Mid Core Mid Growth ARCHER-DANIELS BED BATH & BEYOND ALLIED WASTE ALCAN ALUMINUM ECOLAB PHELPS DODGE BLACK & DECKER PERKIN ELMER SAPIENT COOPER INDUSTRIES STARBUCKS NCR CORP TEKTRONIX Small Value Small Core Small Growth GREAT A&P TEA RUSSELL ARMSTRONG SPRINGS Classifications like the ones above are used to determine portfolio style allocations at a point in time and to perform attribution analyses that are far superior to the old-fashioned sector-based approaches that we’ve been using for the past 30 years. Unlike sector classifications that seldom change, style classifications change frequently. Philip Morris was a growth stock not too long ago, now it’s value. Microsoft and General Electric have been hovering on the cusp between core and growth. The world of styles is very dynamic.

Doing the Job Right It may appear that getting the drift is much too complicated, but like the skilled surgeon, the job can be routine with the best tools. The best tool for point-in-time attribution analysis incorporating style is StokTrib, from PPCA, Inc. (www.PPCA-Inc.com). Consulting professionals like yourself have been using StokTrib for more than a year now to be proactive to their clients’ needs. Importantly, this very sophisticated tool distills the complex down to the easily understood. You don’t need to worry about

http://www.ppca-inc.com)/

http://www.ppca-inc.com)/

5

calculating sources of return, style classifications, and style drift. It’s all done for you. Like the medical doctor, you just need to read the report, confident that the technology used is best-of-class. It’s your expert opinion that the client is paying for. Isn’t it time you got the drift? Don’t your clients deserve it?

APPENDIX: Style Groupings

Style groupings are based on data provided by Compustat. Two security databases are

used. The U.S. database covers more than 8,000 firms, with total capitalization

exceeding $14 trillion. The non-U.S. database coverage exceeds 10,000 firms, 20

countries, and $18 trillion − substantially broader than EAFE.

To construct style groupings, we first break the Compustat database for the region into size groups based on market capitalization, calculated by multiplying shares outstanding by price per share. Beginning with the largest capitalization company, we add companies until 60% of the entire capitalization of the region is covered. This group of stocks is then categorized as "large cap" (capitalization). For the U.S. region, this group currently comprises 180 stocks, all with capitalizations in excess of $16 billion. The second size group represents the next 35% of market capitalization and is called "mid cap". Finally, the bottom 5% is called "small cap" or "mini cap".

Then, within each size group, a further breakout is made on the basis of orientation.

Value, core, and growth stock groupings within each size category are defined by

establishing an aggressiveness measure. Aggressiveness is a proprietary measure that

combines dividend yield and price/earnings ratio. The top 40% (by count) of stocks in

aggressiveness are designated as "growth," while the bottom 40% are called "value,"

with the 20% in the middle falling into "core."

Stok

Trib

Hold

ings

bas

ed p

oint

-in-t

ime

attr

ibut

ion

anal

ysis

ag

ains

t a s

tyle

cus

tom

ized

ben

chm

ark

ww

w.P

PCA-

Inc.

com

Inve

stm

ent P

erfo

rman

ce

Know

ledg

e

Data

Facts

Measu

rement

Rate o

f Retu

rnInfo

rmatio

n

Evalu

ation

Succe

ss ver

sus Fa

ilure

Judgm

ent

Attrib

ution

Know

ledge

Skill v

ersus

Luck

Sour

ces

of R

etur

n

•M

arke

t : H

abit

at

•Se

ctor

or S

tyle

Allo

cati

on

•St

ock

Sele

ctio

n

•Ac

tivi

ty a

nd C

urre

ncy

4

Perf

orm

ance

Attr

ibut

ion

Usin

g St

yles

Port

folio

Mar

ket

Sele

ctio

nPC

T O

FM

AR

KET

VA

LUE

A

PCT

OF

MA

RK

ETV

ALU

EC

RA

TEO

F R

ETU

RN

B

RA

TEO

FR

ETU

RN

D

STO

CK

STY

LETO

TAL

21.5

%0.

0%0.

0%

73.4

%0.

0%0.

0%5.

1%0.

0%0.

0%

100.

0%

21.5

%0.

0%0.

0%

73.4

%73

.4%

0.0%

0.0%

5.1%

0.0%

0.0%

100.

0%

26.3

%12

.8%

20.9

%

16.3

%16

.3%

7.2%

11.5

%2.

6%1.

3%1.

1%

100.

0%

7.4

0.0

0.0

23.6

23.6 0.0

0.0

25.0 0.0

0.0

20.2

8.5

2.5

-1.8

20.4

20.4

18.5 8.4

29.3 4.7

0.6

8.6

-0.2 0.0

0.0

2..3 0.0

0.0

-0.2 0.0

0.0

1.9

-0.0 0.7

2.0

7.3

7.3 -0.8

-0.1 0.5

0.0

0.1

9.7

-0.2 0.7

2.0

9.6

-0.8

-0.1 0.3

0.0

0.1

11.6

LAR

GE

VA

LUE

CO

RE

GR

OW

TH

MID

VA

LUE

CO

RE

GR

OW

TH

SMA

LLV

ALU

EC

OR

EG

RO

WTH

STO

CK

= A

x (B

-D) /

100

STY

LE =

(A

-C)

x (D

-TO

TAL

IND

EX R

ETU

RN

) / 1

00

Attr

ibut

ion

Appr

oach

es

Tot

alM

arke

tSt

yle-

Spec

ific

Mar

ket

Styl

e Gr

oups

Econ

omic

Sect

ors

Econ

omic

Sect

ors

(

new

)

(old

)(n

ewes

t &be

st)

6

Perf

orm

ance

Attr

ibut

ion

Usin

g St

yles

Port

folio

Mar

ket

Sele

ctio

nPC

T O

FM

AR

KET

VA

LUE

A

PCT

OF

MA

RK

ETV

ALU

EC

RA

TEO

F R

ETU

RN

B

RA

TEO

FR

ETU

RN

D

STO

CK

STY

LETO

TAL

21.5

%0.

0%0.

0%73

.4%

0.0%

0.0%

5.1%

0.0%

0.0%

100.

0%

26.3

%12

.8%

20.9

%16

.3%

7.2%

11.5

%2.

6%1.

3%1.

1%

100.

0%

7.4

0.0

0.0

23.6 0.0

0.0

25.0 0.0

0.0

20.2

8.5

2.5

-1.8

20.4

18.5 8.4

29.3 4.7

0.6

8.6

-0.2 0.0

0.0

2.3

0.0

0.0

-0.2 0.0

0.0

1.9

-0.0 0.7

2.0

7.3

-0.8

-0.1 0.5

0.0

0.1

9.7

-0.2 0.7

2.0

9.6

-0.8

-0.1 0.3

0.0

0.1

11.6

STO

CK

= A

x (B

-D) /

100

STY

LE =

(A

-C)

x (D

-TO

TAL

IND

EX R

ETU

RN

) / 1

00

LA

RG

EV

ALU

EC

OR

EG

RO

WTH

MID

VA

LUE

CO

RE

GR

OW

THSM

AL

LV

ALU

EC

OR

EG

RO

WTH

Tota

l Mar

ket

Styl

e Gr

oups

Styl

e-Sp

ecifi

c M

arke

t

Indu

stry

sec

tors

Hist

ory

Abou

t the

Uni

vers

es

Port

folio

Opp

ortu

nity

Dis

trib

utio

ns

(POD

s)

The

Prob

lem

s w

ith

Benc

hmar

ks

and

Peer

Gro

ups

•Pe

er G

roup

Bia

ses

–Su

rviv

or

–Cl

assi

ficat

ion

–Co

mpo

siti

on•

size

•m

embe

rshi

p

•Be

nchm

arks

--De

finin

g

--W

aiti

ng

(Com

pute

rized

Uni

vers

es)

(Com

pute

rized

Uni

vers

es)

Port

folio

Opp

ortu

nity

Dis

trib

utio

nsPo

rtfo

lio O

ppor

tuni

ty D

istr

ibut

ions

(( PO

D’s

POD

’s))

Inve

stm

ent

Inve

stm

ent

Sub

Sub --

Uni

vers

eU

nive

rse

Port

folio

Con

stru

ctio

nPo

rtfo

lio C

onst

ruct

ion

RU

LES

RU

LES

Port

folio

Opp

ortu

nitie

sPo

rtfo

lio O

ppor

tuni

ties

POD

Regi

ons

Cou

ntrie

sC

ount

ries

��U

SAU

SA��

Japa

nJa

pan

��U

KU

K��

Can

ada

Can

ada

��Ire

land

Irela

nd��

Aus

tral

iaA

ustr

alia

��G

erm

any

Ger

man

y��

Fran

ceFr

ance

��Sw

itzer

land

Switz

erla

nd��

Net

herla

nds

Net

herla

nds

��O

ther

Wor

ldW

orld

��G

loba

lG

loba

l��

Glo

bal e

x G

loba

l ex

Japa

nJa

pan

��G

loba

l ex

UK

Glo

bal e

x U

K��

Glo

bal e

x G

loba

l ex

USA

USA

��G

loba

l ex

Glo

bal e

x

U

SA &

Jap

anU

SA &

Jap

an

Con

tinen

tsC

ontin

ents

��N

orth

Am

eric

aN

orth

Am

eric

a��

Far E

ast

Far E

ast

��Fa

r Eas

t ex

Far E

ast e

x Ja

pan

Japa

n��

Euro

peEu

rope

��Eu

rope

ex

UK

Euro

pe e

x U

K��

Emer

ging

Em

ergi

ng

Mar

kets

Mar

kets

Oth

er

Surz

Sty

le D

efin

itio

ns

Capi

taliz

atio

n

•La

rge

: Top

65%

of $

s

•M

id : N

ext 3

0%

•Sm

all :

Bot

tom

5%

E/P

+ y

•Va

lue

: Top

40%

•Co

re : M

iddl

e 20

%

•Gr

owth

: Bot

tom

40%

Mob

ius

Rese

arch

Cand

idat

es: R

usse

ll, S

urz,

Pru,

Wils

hire

,Ves

tek

•Fu

ndam

enta

l : A

ll w

in e

xcep

t Wils

hire

•Tr

acki

ng e

rror

: Sur

z & R

usse

ll

•M

ulti

colli

near

ity

: Sur

z

•Co

vera

ge : S

urz &

Rus

sell

•W

inne

r : R

usse

ll???

?

SENIOR CONSULTANT November 1999

© 1999 PCT Publishing � Circulation and Editorial Offices � 1457 Crystal Springs Lane � Richmond, VA 23231 � 804-795-1642 � Fax 804-795-7703

Performance attribution determinesthe reasons for performance beinggood or bad. Attribution goes

beyond performance evaluation touncover manager strengths and weak-nesses, thus empowering us to makeinformed hire/retain/fire decisions. Thisarticle describes a popular attributionapproach that has been used for the pasttwo decades and recommends two supe-rior approaches. We know much moretoday than we did two decades ago andtoday's computer technology makes itpossible to do much more.

The Need to ModernizePerformance Attribution

Attribution attempts to differentiateluck from skill and to identify the typesof skill. Erroneous inferences based onantiquated methods can be quite costlyto investors when they lead to wrongfulfirings and/or misguided hirings.Investment managers benefit most fromenlightened attributions throughimproved client communications anddecision making. Managers use attribu-tion to see what works and why, and tomake appropriate adjustments.

To properly separate luck from skillwe know today that style is usually animportant consideration. Ignoring stylecan lead to false conclusions such asattributing to skill when it’s really anin-favor style, or damning the skillfulmanager whose style is out of favor.Antiquated attribution methods ignorestyle effects entirely. Contemporaryattributions properly separate theeffects of style from skill in sector andstock selection. Except for sector fundmanagers, all domestic managersmanage to a style, either explicitly orimplicitly. Even managers who professstyle neutrality tend to exhibit an iden-tifiable and consistent style profile.Accordingly, the performance of mostmanagers should be attributed to sectorand stock selection within their style.The assessment of luck or skill withinthe style is then based on the probabil-ity that these selections are non-random, or statistically significant.

Old-Fashioned Sector-BasedAttribution

Table 1 shows an example of a popu-lar attribution approach that useseconomic sectors to differentiate sectorselection from stock selection. The firsttwo columns of the table show thesubject portfolio's commitment andreturn to each economic sector, respec-tively. The second column also shows

return rankings in parentheses with 1being the best and 100 the worst. Thisranking is the probability that return isdue to skill or lack thereof. The deriva-tion of these rankings is discussedbelow. The next two columns show thecorresponding market commitmentsand returns, where the market in thetable is the entire Compustat database(over $11 trillion comprising approxi-mately 8,000 stocks). The last twocolumns use differences in commit-ments to attribute sector selectioneffects and differences in performanceto attribute stock selection effects.

The example is for a real live mid-cap growth manager in the thirdquarter of 1998. As can be seen in thetable, the manager's buy-and-holdreturn of -17.69% significantly lags themarket's -12.25% return, and most ofthis underperformance (-5.48%) isattri0butable to poor stock selection.Compounding this problem, activityduring the quarter subtracted 1.52%,bringing the actual return (-19.21%)ranking down to the 91st percentile, orthe bottom decile. Activity is thedifference between the actual returnand the buy-and-hold return, and is sonamed because it arises from tradingduring the quarter. The arithmetic forarriving at these conclusions isstraightforward and is shown at the topof the last two columns.

It would seem that a serious conver-

Performance Evaluation:Incorporating Style in Attribution Analysis

Ronald J. Surz

Ronald J. Surz is a Managing Directorof Roxbury Capital Management, aninvestment management firm in SantaMonica. He has earned an MBA inFinance at the University of Chicagoand an MS in Applied Mathematics atthe University of Illinois, and holds aCIMA (Certified InvestmentManagement Analyst) designation.Ron can be reached at 949-488-8277 [email protected].

Table 1.Old-Fashioned Sector-Based Performance Attribution

Portfolio Market Attribution(D-m)

A B C D (A-C)* A(B-D)Cmtmt Return (Rk) Cmtmt Return Sectr Slctn

Non-Dur 21.6 -22.89 (95) 23.2 -16.03 .06 -1.48Durabls 1.9 -40.53 (99) 9.3 -19.33 .53 -.40Mat&Ser 11.1 -31.28 (99) 6.5 -17.41 -.24 -1.53Cap Gds 4.1 19.50 (1) 3.1 -21.58 -.09 1.67Tchnlgy 38.0 -7.48 (81) 29.9 -2.60 .77 -1.85Energy 2.2 -46.68 (99) 6.5 -8.14 -.18 -.84Transpo 3.4 -29.44 (99) 2.5 -19.46 -.07 -.34Utility .0 .00 3.3 4.22 -.55 .00Finance 17.8 -24.93 (74) 15.6 -21.00 -.20 -.70

-17.69 (86) -12.25 .05 -5.48Activity -1.52 m

-19.21 (91)Source: PPCA

November 1999 SENIOR CONSULTANT Page 13


sation with this manager is in order,and perhaps more. But, you will seethat this manager, who is actually quiteskillful, did not mess up nearly as badlyas it would appear.

Style-Based Attribution

As mentioned above, we know thatthis manager is mid-cap growth. Couldthis be a possible explanation for hisapparent abysmal stock selection?More importantly, shouldn't his knownstyle of management be incorporatedinto the attribution analysis? We knowsomething today that we didn't know 20years ago when sector-based attributionwas developed: manager style is apowerful determinant of investmentperformance.

Table 2 presents style-based attribu-tion for the same manager and timeperiod shown in Table 1. The analysisproceeds along the same lines as Table1, but now stocks are classified by stylerather than by economic sector. Thedefinitions used for these style classifi-cations are provided in Appendix 1.

Table 2 tells a completely differentstory than Table 1. The style-basedattribution indicates that virtually all ofthe underperformance is attributable tothe manager's style − mid-cap growthwas out of favor in the quarter.Actually, only large cap was in favor forthe quarter. This manager's stock selec-tion is actually average rather thanabysmal.

In addition to superior and accurateinferences, style-based attributionmonitors manager adherence to style.In Table 2, we see that the manager ispredominantly mid-cap growth, with a61.3% allocation, but that 37% of hisportfolio is allocated between mid-capvalue and mid-cap core. This is quiteacceptable since very few managersallocate 100% to their stated style. Ofcourse, there would be cause for concernif a pattern of style divergence hademerged or if there were a significantdeparture from the style mandate.

Style-Specific Attribution

We could stop with style-based attri-bution as a superior alternative tosector-based attribution, but there's stillroom for improvement. We can combinestyles with sectors to achieve the bestinferences of manager strengths andweaknesses. To do so, we redefine"market" to be the market for themanager's style. Here we can define the

style independently (e.g. 100% mid-capgrowth), or we can use the profile fromthe style-based attribution (61.3% mid-cap growth, 22.8% mid-cap core, 15.6%mid-cap value, and 1.2% small-capvalue).

In Table 3, we use the style profilederived in Table 2 to analyze themanager's sector and stock selectionwithin his style universe. As can beseen, the style-specific return of -17.19%is fairly close to the manager's buy-and-hold -17.69 return, ranking it in the52nd percentile against its style-specificuniverse. Adjusting for the 1.52%subtracted from performance by activityduring the quarter, the manager's rank-ing declines somewhat to the 58thpercentile. These style-specific rankingsare substantially better than the bottomdecile rankings suggested by compari-son to the broad market. In otherwords, this manager underperformedrelative to his style, but not by much.

You can also see how style matterswithin sectors. For example, considerthe technology sector. As shown inTable 1, the total technology sectorreturned -2.6% in the quarter. Bycontrast, the mid-cap growth portion ofthe technology sector returned themuch lower -13.55% shown in Table 3.You needed to be in the big caps toavoid the hit to technology in the thirdquarter of 1998.

About the Rankings

Rankings are shown in the tables(under the "Rk" columnar heading) forperformance in each sector/style and fortotal performance. These rankings arederived from Portfolio OpportunityDistributions (PODs) where thecomputer generates all the possibleportfolios that could be held in thestocks comprising the sector, style orrelevant market.1 Among the features of

Table 2.New Style-Based Performance Attribution



LrgeVal .0 .00 21.7 -8.68 -.78 .00LrgeCor .0 .00 10.8 -8.02 -.46 .00Lrge Gro .0 .00 27.5 -9.80 -.67 .00MidlVal 15.5 -26.14 (96) 13.7 -13.87 -.03 -1.91MidlCor 21.9 -21.48 (72) 8.0 -18.45 -.86 -.67MidlGro 61.3 -14.62 (34) 13.4 -17.90 -2.71 2.01SmlVal 1.3 3.65 (2) 2.3 -15.24 .03 .24SmlCor .0 .00 1.2 -21.66 .11 .00SmlGro .0 .00 1.5 -29.38 .25 .00

-17.69 (86) -12.25 -5.11 -.32Activity -1.52 m

-19.21 (91)

Table 3.New Style-Specific Performance Attribution



Non-Dur 21.6 -22.89 (70) 23.9 -17.50 .00 -1.16Durabls 1.9 -40.53 (99) 10.6 -22.22 .42 -.34Mat&Ser 11.1 -31.28 (91) 11.0 -18.52 .00 -1.41Cap Gds 4.1 19.50 (1) 4.6 -22.90 .03 1.72Tchnlgy 38.0 -7.48 (28) 26.4 -13.55 .44 2.30Energy 2.2 -46.68 (99) 4.7 -21.16 .10 -.56Transpo 3.4 -29.44 (91) 2.8 -19.46 -.01 -.34Utility .0 .00 3.8 -5.17 -.46 .00Finance 17.8 -24.93 (69) 12.1 -18.40 -.06 -1.16

-17.69 (52) -17.19 .46 -.95Activity -1.52 m

-19.21 (58)Source: PPCA

SENIOR CONSULTANT November 1999


PODs are their lack of bias and theirearly availability. POD universes, orcyberclone peer groups, are free fromnumerous biases that plague traditionalpeer groups. Also, since PODs relysolely on security data, they are avail-able within days of quarter-end, so eval-uations can be performed much soonerthan those approaches that wait for livemanager databases to be assembled,and they are constructed over cumula-tive time periods as well as the singleperiods shown here.

A Summary So Far

Table 4 summarizes the results ofTables 1 through 3. Here we see thatthe manager's overall ranking is nearthe bottom decile, but his style-basedranking is near median. We also seethat his style returned -17.9% in thequarter versus the market's -12.25%return, resulting in much of his under-performance being attributable to style.

Extensions

This discussion has focused ondomestic equity performance attribu-tion before taxes for a single quarter.

Extending the analysis to other assetclasses over time after taxes is straight-forward. Cumulative period analysesare complicated by the fact that returnsare compounded, but it is desirable tohave attribution components added.Algorithms developed to solve this prob-lem have been applied in the past toold-fashioned sector-based approaches,2

and these algorithms have now beenapplied to the style-based approachesdescribed here.

Tax effects are assimilated into theattribution analysis by subtracting thedifference between actual taxes andthose taxes that would have been duefrom holding the manager's stylepassively.

Non-U.S. attributions follow a simi-lar style-based framework with an addi-tional component added for currencyeffects. These currency effects are eval-uated using an appropriate style-basedPOD for currency returns. Some arguethat non-U.S. managers have no style,but there's increasing evidence thatdespite statements to the contrary, non-U.S. managers do, in fact, frequentlyexhibit an identifiable and consistentstyle profile. For U.S. managers invest-ing abroad, this style profile tends to be

a mix of large cap value and large capcore (surprised?).

Bond attributions proceed alongsimilar style lines too, with differentstyle definitions, namely low, middleand high quality, segmented by long,intermediate and short durations.

The Cross of Changes

Mark Twain said "Everyone hateschange but a wet baby." So it is with theideas presented in this article. I askedeveryone I could think of what theythought of style-based and style-specificattributions. All agreed that style-specific attribution is indeed a new idea.Most thought it was a pretty good idea,but some felt otherwise. Here are someof the criticisms and my brief responsesto these misconceptions.

1. Evidence

Criticism: There is no evidence thatstyles explain any more or less thansectors.

Response: Zephyr Associates and TheMöbius Group have both conductedresearch on the relative explanatorypowers of styles versus sectors. Using

Table 4.Summary of Tables 1 Through 3

Overall Rank Style-Specific RankBuy & Hold Return -17.69 86 52Activity -1.52Actual Return -19.21 91 55Attribution Market Return Sector/Style Allocation Stock SelectionOverall Market

Style-Based -12.25 -5.48 .05Sector-Based -12.25 -.32 -5.11

Style-Specific MarketSector-Based -17.90 -.22 .47

Allocations LrgeVal LrgeCor LrgeGro MidlVal MidlCor MidlGrov SmlVal SmlCor SmlGroActual .0 .0 .0 16.0 22.3 60.4 1.3 .0 .0Market 21.7 10.8 27.5 13.7 8.0 13.4 2.3 1.2 1.5StylMkt .0 .0 .0 .0 .0 100.0 .0 .0 .0

Non-Dur Durbl Mat&Ser Cap Gds Tchnlgy Energy Transpo Utility FinanceActual 22.1 1.9 11.2 4.2 36.5 2.2 3.5 .0 18.3Market 23.2 9.3 6.5 3.1 29.9 6.5 2.5 3.3 15.6StylMkt 22.6 10.2 10.3 3.2 38.4 4.6 1.8 1.5 7.4Allocation Attribution LrgeVal LrgeCor LrgeGro MidlVal MidlCor MidlGro SmlVal SmlCor SmlGroMarket -.78 .46 -.67 -.04 -.89 -2.66 .03 .11 .25

Non-Dur Durbl Mat&Ser Cap Gds Tchnlgy Energy Transpo Utility FinanceMarket .04 .53 -.24 -.10 .64 -.18 -.07 -.55 -.24StylMkt .01 .21 .00 -.03 -.05 .08 -.02 -.18 -.24Selection Attribution LrgeVal LrgeCor LrgeGro MidlVal MidlCor MidlGro SmlVal SmlCor SmlGroMarket .00 .00 .00 -1.95 -.63 -1.09 .24 .00 .00

Non-Dur Durbl Mat&Ser Cap Gds Tchnlgy Energy Transpo Utility FinanceMarket 1.53 -.40 -1.56 1.74 -2.52 -.85 -.35 .00 -.70StylMkt -.72 -.38 -1.48 1.72 2.17 -.56 -.35 .00 -.87

November 1999 SENIOR CONSULTANT Page 15


returns-based optimizations to explainthe performance of a large cross-sectionof funds, both firms conclude that styleindexes do a much better job of explain-ing performance that do sector indexes.

2. S&P Rules

Criticism: At the end of the day, theclient only cares about beating theS&P 500.

Response: The whole focus of perfor-mance evaluation ought to be on devel-oping expectations for the future. Foran individual manager these expecta-tions evolve from an understanding ofthe manager's form and substancewhich, in today's world, translates intostyle and skill. It's hard to know inadvance if the manager's style will be infavor, but you can still develop expecta-tions − many today are expecting largecap growth to end soon. The skill part isthe manager's crystal ball. Bill Sharpeand others have demonstrated thateven the best crystal balls can go unap-preciated when style effects are ignoredbecause they are so powerful.Comparisons to the S&P 500 can easilymiss a skillful manager whose style isout of favor, and this can happen for anextended period of time − long enoughto result in unwarranted termination.

3. Agreeing on Classifications

Criticism: No one will ever agree onthe classification of stocks into stylesegments.

Response: It is true that there willprobably never be a universallyaccepted mapping of stocks into stylegroups, but there has never been such amapping into economic sectors either.Yet, sector-based analysis is theaccepted way of doing things. The styleallocations used here are described inthe Appendix. They have beencompared to other approaches by theMöbius Group and found to be superior.

4. The Generalist

Criticism: Style-based analyses don’tmean much to a manager with a broadmarket mandate.

Response: Style-based analysis maybe less meaningful for a generalistmanager, but it never hurts to observehow the manager implements his broadmandate. Most evaluators use sectoranalysis to make these observations.Why not use styles as well for addi-tional insights.

Conclusion

Manager style really is a very power-ful determinant of performance.Ignoring this fact, by using old-fash-ioned sector-based attribution analysiscan lead to significantly misleadinginferences, as shown in our real-lifeexample.

Style-based attribution producesmuch more reliable indicators becausemost investment managers manage to astyle rather than to some mix ofeconomic sectors. Style-specific attribu-tion, which combines styles witheconomic sectors, is even better since wegain a better understanding of themanager's unique market as well as themanager's successes and failures withinthis security habitat. Investmentmanagers stand to benefit most from

modernized investment performanceattribution as a result of the superiorinsights obtained from style-basedanalyses. Since superior insights lead tosuperior decisionmaking the clientbenefits as well.

Free software for the approachdescribed here can be obtained atwww.PPCA-inc.com so you can experi-ence the difference. �

1Surz, Ronald J. (summer 1994),"Portfolio Opportunity Distributions:An Innovation in PerformanceEvaluation", Journal of Investing. Surz,Ronald J. (winter 1996), "PortfolioOpportunity Distributions: A Solutionto the Problems with Benchmarks and

Peer Groups," Journal of PerformanceMeasurement. Surz, Ronald J., (winter1998), "Cyberclone Peer Groups,"Journal of Investing.

2Carino, David R. (summer 1999),"Combining Attribution Effects OverTime", The Journal of PerformanceMeasurement.

Many thanks to the following fortheir helpful comments and sugges-tions: William F. Brock of EvaluationAssociates, Dr. Michael Edesess ofLockwood, Steve Hardy of ZephyrAssociates, Dr. Robert Haugen ofHaugen Systems, Rex Macey of Macey-Holland, Jack Marco of MarcoAssociates, Kenneth Phillips of MoneyManagement Partners, Troy Schell ofJeffrey Slocum & Associates, Dr.William F. Sharpe of Financial Engines,and two anonymous reviewers from TheJournal of Performance Measurement.

Appendix. Style Groupings

Style groupings are based on dataprovided by Compustat. Two securitydatabases are used. The U.S. databasecovers more than 7,500 firms, with totalcapitalization exceeding $11 trillion.The non-U.S. database coverage exceeds5,500 firms, 20 countries, and $10 tril-lion − substantially broader than EAFE.

To construct style groupings, we firstbreak the Compustat database for theregion into size groups based on marketcapitalization (calculated by multiply-ing shares outstanding by price pershare). Beginning with the largest capi-talization company, we add companiesuntil 60% of the entire capitalization ofthe region is covered. This group ofstocks is then categorized as "large cap"(capitalization). For the U.S. region,this group currently comprises 150stocks, all with capitalizations in excessof $13 billion. The second size grouprepresents the next 35% of market capi-talization and is called "mid cap".Finally, the bottom 5% is called "smallcap", or "mini cap".

Then, within each size group, afurther breakout is made on the basis oforientation. Value, core and growthstock groupings within each size cate-gory are defined by establishing anaggressiveness measure. Aggressive-ness is a proprietary measure thatcombines dividend yield and price/earn-ings ratio. The top 40% (by count) ofstocks in aggressiveness are designatedas "growth," while the bottom 40% arecalled "value," with the 20% in themiddle falling into "core."

Style-basedattribution

produces muchmore reliable

indicators becausemost investment

managers manageto a style rather

than to some mixof economic

sectors

getting the drift - northfield · the first period after a style shift from value to growth...

Documents