Factor Investing with Smart Beta Indices
David Blitz *
First version: 29 April 2016
This version: 3 August 2016
Abstract
The added value of smart beta indices is known to be explained by exposures to
established factor premiums, but does that make these indices suitable for implementing a
factor investing strategy? This paper finds that the amount of factor exposure provided by
popular smart beta strategies differs considerably, as does their degree of focus on a
single target factor. It also provides insight into how ‘quality’ and ‘high dividend’ indices
relate to academic factors. Smart beta indices exhibit a performance that is in line with
the amount of factor exposure provided, but it seems that they do not unlock the full
potential offered by factor premiums. Altogether, these results imply that factor investing
with smart beta indices is not as straightforward as one might think.
Keywords: factor investing, smart beta
JEL Classification: G11, G12, G14
* David Blitz, Ph.D. is the Head of Quantitative Equity Research at Robeco Asset Management and can be
contacted at [email protected] and +31 (0)10 - 224 2079. The views expressed in this paper are solely
those of the author and not necessarily shared by Robeco or its subsidiaries.
Electroniccopy
Electronic copy available
available at:
at:https://ssrn.com/abstract=2771621
http://ssrn.com/abstract=2771621
Asset allocators increasingly consider factor premiums next to traditional asset class risk
premiums. Such an approach is known as “factor investing”, and typically focuses on
capturing factor premiums that have been extensively documented in the literature, such
as the value, momentum and low-volatility premiums. Studies advocating factor investing
include Ang, Goetzmann and Schaefer [2009], Bender et al. [2010], Blitz [2012, 2015],
and Ang [2014]. In theory, the benefits of factor investing are largest for long-short factor
portfolios, which capture pure factor premiums and have a low correlation with asset
class risk premiums; see, for example, Ilmanen and Kizer [2012]. In practice, however,
factor investing is typically implemented using long-only strategies, which are designed
to capture a factor premium on top of an asset class risk premium. A popular way to do so
is by replicating the performance of smart beta indices, which use mechanical rules to
deviate from the capitalization-weighted market index. These rules tend to result, either
explicitly or implicitly, in systematic tilts towards certain factors. Chow et al. [2011]
empirically show that the added value of popular smart beta indices can be attributed
entirely to exposures to established factor premiums.
This paper investigates how smart beta indices can be used to implement a factor
investing strategy. As a starting point we use the theoretical framework of Blitz [2012],
who argues to allocate strategically to value, momentum and low-volatility equity factor
portfolios, and Blitz [2015], who also finds some added value for the two new factors in
the Fama and French [2015] 5-factor model, profitability and investment. The factor
portfolios considered by Blitz [2012, 2015] are long-only and restricted to the large-cap
segment of the market, in order to ensure that they can be implemented on a large scale.
The portfolios are considered on a value-weighted as well as on an equal-weighted basis,
with the latter showing the best results. We use a returns-based style analysis approach to
investigate if, and to which extent, smart beta indices are able to capture the theoretical
returns of these factor portfolios.
The main finding is that factor investing with popular smart beta indices is not as
straightforward as one might think. Smart beta strategies typically seem to target one
particular academic factor, but it turns out that the amount of exposure they provide to
that factor can differ a lot. Many smart beta strategies do not offer maximum factor
exposure, but still contain a significant amount of market index exposure as well, or some
Electroniccopy
Electronic copy available
available at:
at:https://ssrn.com/abstract=2771621
http://ssrn.com/abstract=2771621
unexpected exposures to other factors. We also provide insight into how popular practical
concepts such as ‘quality’ and ‘high dividend’ relate to established factors. Finally, we
find that the performance of smart beta indices seems to be in line with the amount of
factor exposure they provide, so there does not appear to be significant slippage.
However, smart beta indices may not unlock the full potential of factor premiums,
because in most cases they seem to be exposed to value-weighted factor strategies, while
equally-weighted factor strategies are known to generate higher returns.
Altogether, these results imply that smart beta indices may be used to harvest
generic factor premiums, but also that it is crucial to properly understand the
characteristics of these indices in order to obtain the desired amount of exposure to each
particular factor, and the intended portfolio risk-return profile.
DATA AND METHODOLOGY
In the academic literature, factor portfolios are typically defined following the
methodology of Fama and French [1993]. Their approach consists of going long in the
30% stocks that are most attractive on a certain factor, and short in the 30% stocks that
are least attractive on a certain factor, in the large-cap segment of the market as well as in
the small-cap segment of the market (fifty-fifty). Blitz [2012] proposes a simplified factor
investing approach, which considers only the long leg (i.e., the 30% most attractive
stocks on a certain factor) in just the large-cap segment of the market. The motivation is
that such factor portfolios are much easier to implement in size, because they do not
involve short-selling or investing in illiquid stocks, and can easily replace large existing
active or passive equity allocations. Using factor portfolios defined in this way, the study
recommends allocating strategically to the value, momentum and low-volatility factors.
Since smart beta indices also tend to be long-only and focused on the liquid, large-cap
segment of the market, a natural question is how they relate to these factor portfolios.
Blitz [2015] additionally considers profitability and investment factor strategies,
inspired by Fama and French [2015], who propose a 5-factor model by augmenting their
classic 3-factor model with these two factors. A simple, equally-weighted combination of
value, momentum and low-volatility factor portfolios is found to result in a Sharpe ratio
of 0.49, versus only 0.32 for the capitalization-weighted market portfolio. The Sharpe
Electronic copy available at: https://ssrn.com/abstract=2771621
ratio can be improved further to 0.58 by using equally-weighted instead of value-
weighted factor portfolios. By adding profitability and investment factor portfolios to the
mix these Sharpe ratios are not improved further, but information ratios do go up, i.e.
market-relative instead of absolute risk-adjusted performance improves.
This paper follows the methodology of Blitz [2012, 2015] for defining the value,
momentum, low-volatility, profitability and investment factor strategies that will serve as
a theoretical reference point. The value, momentum, profitability and investment
portfolios are taken directly from the online data library of Kenneth French. The value
strategy selects stocks on their book-to-market ratio and the momentum strategy selects
stocks on their past 12-1 month return. Profitability is defined as annual revenues minus
cost of goods sold, interest expense, and selling, general, and administrative expenses
divided by book equity, and investment as the change in total assets (where a low score is
better than a high score). The low-volatility strategy is not provided by Kenneth French
and therefore self-constructed, following a similar methodology. Specifically, it selects
the 30% stocks with the lowest past 36-month total-return volatility in the large-cap
segment of the market, which is defined as all stocks in the CRSP database with a market
capitalization above the NYSE median.
The purpose of this paper is not to provide a comprehensive analysis of the vast
number of smart beta indices available to investors nowadays. It turns out that many
valuable insights can already be obtained by focusing on some of the most popular ones.
We obtain smart beta index data from Thomson Reuters Datastream, preferring indices
which offer a long data history. All indices are for the U.S. equity market. Unlike the
academic factor portfolios, the smart beta indices are not defined in a consistent manner,
but each use a different index construction methodology. For the value factor we consider
the classic Russell 1000 Value index, which is available since January 1979, and,
additionally, a fundamentally weighted index. Fundamentally weighted indices were
pioneered by Research Affiliates, but since data for the FTSE/RAFI index is only
available from the end of 1999 we take the more recently introduced MSCI Value
Weighted index, which uses a similar methodology, but offers data availability since
December 1976. For the momentum factor we consider the MSCI Momentum index,
which is available from January 1975 onwards. For the low-volatility factor we take the
Electronic copy available at: https://ssrn.com/abstract=2771621
S&P Low Volatility index, available since December 1990, instead of the MSCI
Minimum Volatility index, which has a longer live track record, but only history from the
end of 1998 onwards. Indices that specifically target the new profitability and investment
factors do not seem to be available, but instead we consider two indices that are quite
popular in practice: the MSCI Quality index, with data from December 1975 onwards,
and the MSCI High Dividend index, available since January 2001. For the market
portfolio we use the standard index of MSCI. The data frequency is monthly, and the
sample ends in December 2015. All portfolio and index returns are taken in excess of the
risk-free return and reported on an annualized basis.
We analyze the performance of the smart beta indices using a returns-based style
analysis approach in the spirit of Sharpe [1992]. Our approach is equivalent to regressing
the monthly returns of smart beta indices on the monthly returns of factor indices, with
the restriction that the coefficients estimated for the factor indices are non-negative. In
other words, we find a long-only, weighted combination of factor indices which provides
the best match with each smart beta index, in the sense that it minimizes the sum of
squared errors. For each smart beta index we use the maximum available history.
RESULTS
Exhibit 1 shows how value-weighted factor portfolios are able to explain the return of
smart beta indices, considering the maximum available data history for each index. Not
surprisingly, the Russell 1000 Value index loads most on the Value factor portfolio.
However, its exposure to the Value factor portfolio is only 36%. The remaining exposure
is attributed to the market portfolio (23%), the Investment factor (21%) and the Low
Volatility factor (20%). This implies that the Russell 1000 Value index provides highly
diffuse factor exposures, and that it is not very suitable for investors seeking pure and
sizable exposure to the (academic) value factor. The MSCI Value Weighted index, which
uses fundamental weighting, provides more pure value exposure, but again very little, as
60% of its exposure is attributed to the market portfolio. The high loading on the market
portfolio is not surprising, as the Value Weighted index invests in all index constituents,
which results in considerable overlap with the market index and therefore a relatively low
active share.
Electronic copy available at: https://ssrn.com/abstract=2771621
INSERT EXHIBIT 1 ABOUT HERE
Exhibit 1 also shows that the MSCI Momentum index loads heavily on the Momentum
factor portfolio, with a weight of 73%. The remaining weight goes to the market
portfolio. The S&P Low Volatility loads heavily on the Low Volatility factor portfolio
(71%), combined with a bit of value exposure. These figures suggest that these indices
are quite suitable for obtaining momentum and low-volatility factor exposure, although
they still seem to fall short of offering maximum exposure to these factors. The MSCI
Quality index turns out to load heavily on the Profitability factor portfolio (76%), which
implies that it may be used to obtain exposure to this new Fama-French factor. However,
it does not provide exposure to the other new Fama-French factor, Investment, as the
remaining weight is simply attributed to the market portfolio. For investors specifically
seeking exposure to the Investment factor, none of the smart beta indices considered here
appear to be very useful. Interestingly, the MSCI High Dividend index provides a huge
(83%) exposure to the Low Volatility factor portfolio, and next to that some value
exposure. Although the MSCI High Dividend index outperforms its replicating portfolio
of factor strategies with almost 1% per annum, the outperformance is statistically
insignificant, which implies that there is insufficient evidence for added value beyond
classic low-volatility and value factor exposure.
The performance of the other smart beta indices is also in line with their factor
replication portfolios. The one exception is the S&P Low Volatility index, which shows
an economically large (over 2% per annum) and statistically significant outperformance.
This outperformance disappears entirely though if we also include equally-weighted
factor portfolios in the analysis, as shown in Exhibit 2. In this case we find 84% exposure
to the equally-weighted Low Volatility factor portfolio, plus another 8% exposure to the
value-weighted Low Volatility factor portfolio, and a performance difference which is
close to zero. The finding that the S&P Low Volatility index bears much more
resemblance to an equally-weighted Low Volatility factor portfolio than to a value-
weighted one is not surprising, given that the weighting methodology used to construct
this index (inverse volatility) is much closer to equal weighting than to value weighting.
Electronic copy available at: https://ssrn.com/abstract=2771621
INSERT EXHIBIT 2 ABOUT HERE
From Exhibit 2 we can also infer that the other smart beta indices do not load much on
the equally-weighted factor portfolios, since most of the weight still goes to the value-
weighted factor portfolios. As equally-weighted factor portfolios show a better
performance than value-weighted ones, this suggests that smart beta indices may not
unlock the full potential offered by factor premiums. Perhaps the providers of smart beta
indices give more priority to maximizing the capacity of their indices, in which case
value weighting does make more sense than equal weighting.
REVERSE ANALYSIS
We continue by turning the analysis around, i.e. instead of trying to match the
performance of smart beta indices with a combination of theoretical factor portfolios, we
now use the same methodology to match the performance of theoretical factor portfolios
with a combination of smart beta indices. Specifically, we consider the diversified factor
portfolio recommended by Blitz [2012], which invests in the value, momentum and low-
volatility factors portfolios (with a weight of 1/3 each), and the alternative diversified
factor portfolio of Blitz [2015], which adds the profitability and investment factors to this
mix (with weights of 1/5 for each factor). We distinguish between 3x1/3 and 5x1/5
diversified factor portfolios that use either value-weighting or equal-weighting for the
underlying single factor portfolios.
We consider the sample period from December 1990 until December 2015, which
allows us to include all the smart beta indices considered before, except for the MSCI
High Dividend index which does not go back that far. The latter probably does not matter
much, because if we consider the shorter period over which that index is also available,
we find that it gets assigned a weight of zero in all instances, suggesting that it is
redundant in this particular analysis.
INSERT EXHIBIT 3 ABOUT HERE
Electronic copy available at: https://ssrn.com/abstract=2771621
The first two columns of Exhibit 3 show that the performance of the value-weighted
3x1/3 and 5x1/5 theoretical factor portfolios is matched well by their replicating
portfolios of smart beta indices. The absence of a smart beta index which specifically
targets the investment factor does not appear to be very problematic. Interestingly, the
weights assigned to the various smart beta indices are quite different from 3x1/3 and
5x1/5, although this is not really surprising given our observation that the smart beta
indices provide very different amounts of factor exposure. Most notably, about half of the
weight in the matching portfolio goes to the classic Russell 1000 Value index.
The last two columns of Exhibit 3 show that it is a lot more difficult to match the
performance of the equally-weighted 3x1/3 and 5x1/5 theoretical factor portfolios with
smart beta indices. The tracking errors of the replicating portfolios are twice as high
compared to the case with value-weighted factor portfolios. More importantly, with
equally-weighted factor portfolios the mean annual return of the replicating portfolios
falls short by around 2% per annum, and this shortfall is statistically significant. This
confirms our earlier result that smart beta indices are unable to capture the full potential
offered by factor premiums. If smart beta indices already fail to keep up with simple
equally-weighted, top 30% large-cap factor portfolios, they are likely to fall short even
more if we would consider factor portfolios that are specifically designed for maximum
return. For instance, the return of the theoretical factor portfolios might be boosted further
by increasing their concentration (e.g. take the top 20% or top 10% stocks, instead of the
top 30%) and by not limiting the universe to large-cap stocks (as factor premiums are
known to be stronger in the small-cap segment of the market).
SUMMARY
This paper shows that using smart beta indices to implement a factor investing strategy is
not as straightforward as one might think. Smart beta strategies typically seem to target a
specific factor, but the amount of exposure provided to that factor can differ a lot. Many
smart beta strategies do not offer maximum factor exposure, but still contain a significant
amount of market index exposure as well, or some exposure to other factors. We also
show that ‘quality’ predominantly provides exposure to one of the two new Fama-French
factors, namely profitability, while ‘high dividend’ seems to be redundant for investors
Electronic copy available at: https://ssrn.com/abstract=2771621
who already have exposure to the low-volatility and value factors. Finally, the
performance of smart beta indices seems to be in line with the amount of factor exposure
they provide, so there does not appear to be significant slippage. However, smart beta
indices may not unlock the full potential because in most cases they seem to be exposed
to value-weighted factor strategies, while equally-weighted factor strategies are known to
generate significantly higher returns. Altogether, these results imply that smart beta
indices may be used to harvest generic factor premiums, but also that it is crucial to
properly understand the characteristics of these indices in order to get the intended factor
exposures.
Electronic copy available at: https://ssrn.com/abstract=2771621
REFERENCES
Ang, A. “Asset Management: A Systematic Approach to Factor Investing.” Oxford
University Press, 2014.
Ang, A., Goetzmann, W.N., and Schaefer, S.M. “Evaluation of Active Management of
the Norwegian Government Pension Fund – Global”, 2009, available at
http://www.regjeringen.no
Bender, J., Briand, R., Nielsen, F., and Stefek, D. “Portfolio of Risk Premia: A New
Approach To Diversification.” Journal of Portfolio Management, 36 (2010), pp. 17-
25.
Blitz, D. “Strategic Allocation to Premiums in the Equity Market.” Journal of Index
Investing, 2 (2012), pp. 42-49.
–––––. “Factor Investing Revisited.” Journal of Index Investing, 6 (2015), pp. 7-17.
Chow, T., Hsu, J., Kalesnik, V., and Little, B. “A Survey of Alternative Equity Index
Strategies.” Financial Analysts’ Journal, 67 (2011), pp. 37-57.
Fama, E., and K. French. “Common Risk Factors in the Returns on Stocks and Bonds.”
Journal of Financial Economics, 33 (1993), pp. 3-56.
–––––. “A 5-Factor Asset Pricing Model.” Journal of Financial Economics, 116 (2015),
pp. 1-22.
Ilmanen, A., and J. Kizer. “The Death of Diversification Has Been Greatly Exaggerated.”
Journal of Portfolio Management, 38 (2012), pp. 15-27.
Sharpe, W. “Asset Allocation: Management Style and Performance Measurement.”
Journal of Portfolio Management, 18 (1992), pp. 7-19.
10
Electronic copy available at: https://ssrn.com/abstract=2771621
EXHIBIT 1 | Results using value-weighted factor portfolios
Russell MSCI MSCI S&P MSCI MSCI
1000 Value Value Weighted Momentum Low Volatility Quality High Dividend
Replicating portfolio weights
Market 23% 60% 25% - 23% -
Value 36% 29% - 12% - 20%
Momentum - - 73% - - -
Low Volatility 20% 8% 1% 71% 83%
Profitability 1% - - - 76% -
Investment 21% 6% - - - -
Total 102% 103% 99% 83% 99% 103%
Excess return target index 7.89% 7.55% 9.91% 8.34% 8.03% 6.08%
Excess return replicating portfolio 8.51% 7.63% 9.83% 6.58% 7.90% 5.14%
Return difference -0.62% -0.08% 0.08% 1.75% 0.14% 0.95%
Tracking error 2.53% 2.11% 4.85% 4.18% 2.98% 4.03%
T-statistic -1.49 -0.24 0.10 2.10 0.29 0.91
11
Electronic copy available at: https://ssrn.com/abstract=2771621
EXHIBIT 2 | Results using value-weighted and equally-weighted factor portfolios
Russell MSCI MSCI S&P MSCI MSCI
1000 Value Value Weighted Momentum Low Volatility Quality High Dividend
Replicating portfolio weights
Market 24% 54% 25% - 23% -
Value VW 27% 20% - - - 12%
Value EW 5% 6% - - - 8%
Momentum VW - - 73% - - -
Momentum EW - - - - - -
Low Volatility VW 12% 13% 1% 8% - 83%
Low Volatility EW 13% - - 84% - -
Profitability VW - - - - 76% -
Profitability EW 3% - - - - -
Investment VW 17% - - - - -
Investment EW 1% 9% - - - -
Total 103% 102% 99% 92% 99% 103%
Excess return target index 7.89% 7.55% 9.91% 8.34% 8.03% 6.08%
Excess return replicating portfolio 8.88% 7.85% 9.83% 8.31% 7.90% 5.55%
Return difference -0.99% -0.31% 0.08% 0.02% 0.14% 0.54%
Tracking error 2.39% 2.00% 4.85% 3.11% 2.98% 3.96%
T-statistic -2.52 -0.96 0.10 0.04 0.29 0.52
12
Electronic copy available at: https://ssrn.com/abstract=2771621
EXHIBIT 3 | Reverse analysis
VW 3x1/3 VW 5x1/5 EW 3x1/3 EW 5x1/5
Replicating portfolio weights
MSCI USA 6% 15% - -
Russell 1000 Value 51% 44% 60% 41%
MSCI Value Weighted - - - 35%
MSCI Momentum 25% 17% 24% 19%
S&P Low Volatility 12% 10% 15% 7%
MSCI Quality 3% 12% - -
Total 98% 98% 99% 102%
Excess return target factor strategy 8.67% 8.90% 10.70% 11.05%
Excess return replicating portfolio 8.81% 8.66% 8.93% 9.04%
Return difference -0.14% 0.24% 1.77% 2.01%
Tracking error 2.41% 1.72% 4.17% 4.22%
T-statistic -0.29 0.69 2.13 2.39
13
Electronic copy available at: https://ssrn.com/abstract=2771621
