Professional Documents
Culture Documents
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Taylor & Francis, Ltd. is collaborating with JSTOR to digitize, preserve and extend access to
Financial Analysts Journal
PERSPECTIVES
Why Market-Valuation-Indifferent
Indexing Works
Jack Treynor
B y the end of the 20th century, even casual value with error u is (1 + u)v(u); then, the error
investors had become comfortable with distribution satisfies
the idea of index funds. The idea of a better
index fund (see Arnott, Hsu, and Moore (l+u)v(u)>(1-u)v(-u). (2)
2005), however, is mind-boggling. This article
Unlike the distribution of the pricing error that
offers one man's view of why it will actually work.
uses true value, the error distribution for market
He defines market-valuation-indifferent (MVI)
values is skewed to the right. This lack of symmetry
indexing to be indexing in which the index is built
is the problem with capitalization weighting: By
on any weights that avoid the problem with mar-
using market values to determine its weights, a
ket capitalization.
cap-weighted index fund will invest more money
The bad news about stock markets is that they
in overpriced stocks than in underpriced stocks.
price stocks imperfectly. The good news is that the
Consider a symmetrical distribution of market
mispricings are always relative. Not only will over-
errors u around a mean error u. For each stock
priced stocks be counterbalanced by underpriced
whose error exceeds the mean by u - u-, there will
stocks, but the distribution of error at any point in
tend to be a stock whose error falls short of the
time will be symmetrical. We can picture this dis-
mean by u- - u. Expressed in terms of a frequency
tribution as a bell-shaped curve with "error" on the
function v( ) of true values, the original symmetry
horizontal axis and some measure of "frequency"
condition is obviously satisfied by
on the vertical axis. Because it reflects both the
number of companies and their size, aggregate
v(u- u)=v(ui-u), (3)
value is the appropriate measure of frequency.
But which measure of aggregate value-true because the second argument is indeed minus the
value or market value? If we use market value, first, as we specified. On the other hand, if we
then, alas, it will make bigger bets on overpriced expect market errors to be symmetrical around a
stocks and smaller bets on underpriced stocks. mean error of zero, we need to add the following
To get a handle on how much error, we begin condition: weighted by the true values, the mean
by defining of the errors in market price is zero. In terms of our
u = relative error (expressed as a fraction of symbols, we can express the new condition:
true value) and
v(u) = amount of true value with error.
Xuv (u) = O. (4)
When we consider the thousands of stocks in the
market, the randomness of particular stocks is sub- Obviously, the sum over all stocks-underpriced
merged in a density function that associates a rel- and overpriced-is zero.
atively stable amount of density function v (u) with
relative error u to satisfy
The Basic Equation
v(u)=(-v(-U))1) How does MVI indexing avoid cap-weighted
But 1 + u is the market value of $1.00 of true value indexing's problem? The key is a simple equation
with relative error u. So the amount of market linking the covariances of portfolio weights with
* market price per share,
Jack Treynor is president and CEO of Treynor Capital* true value per share, and
Management, Inc., Palos Verdes Estates, California. * errors in market price per share.
If, as before, u is the relative error, then 1 + u is the benchmark portfolios, hence more sensitivity to
ratio of market to true value v and any systematic small-cap factor (as discussed in, for
example, Fama and French 1973).
v(1+u)=v+vu (5)
The equation relating the three covariances can
is market price. So, to a common divisor equal to be applied in other ways. For example, instead of
the number of stocks, the covariance of portfolio demonstrating empirically that a given set of
weights w with share prices is weights has zero covariance with market prices, we
can appeal to a priori reasons why certain sets of
wv(l +u) - wXu(1 + u) weights will have a zero covariance with the errors.
=Xwv +, wvu -X,w1 ,v (6) We have seen that if the portfolio gives the same
weight to underpricing errors it gives to overpric-
=[Ewv- wXv]+Xwvu. ing errors, the third covariance vanishes.
The expression in brackets is the covariance of port- But then the other two covariances in the equa-
folio weights with the true share values. Now, con- tion must be equal. So we can use market values,
sider the covariance of portfolio weights with which are observable, rather than true values,
dollar errors in share price, which are not, to estimate the small-cap bias in such
sets of weights.
cov w(uv) = X wuv - E WEUV
investments will tend to be equal-in contrast to smallest companies in the Wilshire 5000 Index. It
the cap-weighted index fund, which pays more for will have
the overpriced segment and less for the under- * no alpha resulting from MVI indexing and
priced segment. Alas, a scheme that weights large- * lots of small-cap bias.
cap and small-cap stocks the same is going to have A short position in this portfolio will offset a lot of
small-cap market bias, however, relative to many small-cap bias without reducing the MVI alpha.
ments will tend to be the same. investors own more true value than the cap-
Of course, at a point in time, real stocks won't weighted investors, with a difference that depends
only on the relative size of the aggregate pricing
oblige the author by falling into exactly counterbal-
error. The gain for the whole market sums across
ancing pairs. But the easiest way to explain how
errors occurring with a wide range of frequencies.
MVI capitalizes on the tendency for pricing errors
to be symmetric is to focus on such an idealized pair. If the frequency function is fl(e/ v), then the gain can
be expressed as
Because of the errors in market price, the cor-
responding underpriced or overpriced stocks in a
I f (elv)d(elv)
cap-weighted portfolio will have different market (12)
values even if they have the same true values. Let _0 1_- (e lv)2
the true values of those stocks be v, and let the For small errors, we can approximate this integral by
aggregate pricing errors be +e and -e.
If cap-weighted investors spend v + e dollars
on the former and v - e dollars on the latter, they
[Ll +v) QeY] H v) d(13)
will spend a total of
(v+e)+(v-e)=2v (9) f d fQ v) + e H2 fH dQ d
dollars and get
The value of the first integral is 1. If, as we have
assumed for the frequency distribution of true val-
(v+e)( )+(v-e)( )=2v (10) ues, the mean of the errors is 0, then the second
~v+e v- e
integral is the variance of the errors.
worth of true value. When stocks are accurately priced, the MVI
On the other hand, the MVI investors spend portfolio realizes no gain relative to the price-
the same number of dollars on the underpriced as weighted portfolio. But when the error in market
they spend on the overpriced stocks. But a dollar prices is expressed as a fraction of the true value,
spent on overpriced securities buys less true value then the gain from MVI is the square of the stan-
than a dollar spent on underpriced securities. For dard error, 6. Table 1 displays a range of possible
example, a dollar spent on a stock with true value values of c, c2, 1 + Y2, and (for reasons to be
v and market price v + e buys v/(v + e) of the true explained) 1/(1 - c2). MVI investors realize this
value; a dollar spent on a stock with true value v benefit even if mispriced stocks never revert to their
and market price v - e buys v/(v - e) of the true true values. If reversion occurs, it offers an addi-
value. If the MVI investors spend v dollars on each tional benefit (see Appendix A).
stock, they make the same total investment as the To be sure, the correct integral is not as simply
related to the standard error of stock prices as our
cap-weighted investors and get
crude approximation is. But in the event, small
pricing errors will be much more frequent than
large pricing errors. The reader can get some sense
of how bad our approximation is by imagining that,
worth of true value, or instead of being sample averages, the numbers in
For the MVI investor with equal positions in the before dividing by the effective reversion time.
two stocks, the average return is For the whole portfolio, the return is
1 (e _ e e) e 1 1
2 P2P1) 2P P2 Pi) yT I( jf (d (var v) (A4
again assuming a mean of zero.
e (P -P2 )e ( 2e) (A3)
2 ( PI P2 ) 2 tPI P2)
e 2
PI P2 )
References
Arnott, Robert D., Jason Hsu, and Philip Moore. 2005. Fama, Eugene, and Kenneth French. 1973. "The Cross-Section
"Fundamental Indexation." Financial Analysts Journal, vol. 61, of Expected Returns." Journal of Finance, vol. 47, no. 2
no. 2 (March/April):83-99. (June):427-465.