VOLUME 44 NUMBER 7  www.iijpm.

com SUMMER 2018

The Conservative Formula:

Quantitative Investing Made Easy
David Blitz and Pim van Vliet
Robeco is an International asset manager offering an extensive range of active Investments,
from equities to bonds. Research lies at the heart of everything we do, with a ‘pioneering but
cautions’ approach that has been in our DNA since our foundation in Rotterdam in 1929. We
believe strongly in sustainability investing, quantitative techniques and constant innovation.
 The Conservative Formula:
Quantitative Investing
Made Easy
David Blitz and Pim van Vliet

T
David Blitz he early tests of the capital asset not straightforward. Today, a good starting
is the head of quantitative pricing model (CAPM) in the point for a limited set of persistent factors
equity research at Robeco
1970s showed that the empirical might be the ones identified by Fama and
Asset Management
in Rotterdam, risk–return relation is too f lat. French [2015]. These factors are rigorously
the Netherlands. This low-risk effect is now regarded as one of tested, complement each other, and require a
[email protected] the very first stock market anomalies. Many limited amount of trading. To benefit, inves-
other anomalies have been documented tors should tilt their portfolio toward small,
P im van Vliet since, such as the size, value, quality, and attractively priced, profitable firms with low
is the founder and
head of Conservative
momentum effects. However, the “publish or levels of investments. However, Fama and
Equities at Robeco perish” culture in academia may give a bias French did not create one integrated invest-
Asset Management toward many false positive results. Harvey, ment portfolio, but rather separate long–
in Rotterdam, the Liu, and Zhu [2016] articulated this con- short portfolios for each factor, with many
Netherlands. cern, documented the recent explosion in the overlapping positions. This is of little use for
[email protected]
number of factors, and suggested that thresh- investors looking for an easy and effective
olds for statistical significance be raised. Hou, factor-based investment strategy. One could
Xue, and Zhang [2017] took a fresh look at consider identifying stocks that score well on
the anomalies literature and found that many each of the Fama–French factors, but with
results cannot be replicated and are therefore such an approach the resulting number of
likely the result of factor fishing. Moreover, stocks will be very low. For example, the
the significance of many factors turns out to expected number of stocks with a top quin-
be critically dependent on methodological tile score on each of five independent fac-
assumptions such as frictionless rebalancing, tors is less than 1 in 3,000. Moreover, most
no leverage costs, and unlimited short-selling investors face constraints on leverage and
of micro caps. short selling. Therefore, an important prac-
Investors who want to prof it from tical question is how investors can obtain
academic insights therefore have to be very one liquid, easy to implement, long-only
careful. First, they should be highly critical portfolio that gives access to multiple fac-
as to which factors work and which do not tors simultaneously.
because many results are spurious. The These practical considerations are less
number of factors needs to be narrowed relevant for academics. If a simple investment
down to a subset that is persistent and robust. strategy is profitable but can be explained
Second, translating multiple factors into one by exposures to a wide range of theoretical
easily implementable investment strategy is long–short factors, then from an academic

Summer 2018 The Journal of Portfolio M anagement    1

point of view there is no relevant contribution to the We find that a conservative portfolio consisting
existing literature, even if the control factors require of 100 low-risk stocks with high NPY and positive
the use of accounting data, rely on micro caps, require price momentum returned 15.1% per year since 1929.
high-frequency trading, and assume full shorting oppor- The performance is persistent over time, with positive
tunities. For practitioners, however, efficient exposure returns in every decade. It outperforms a portfolio of
to multiple established factor premiums simultaneously speculative stocks with the opposite characteristics (high
is desirable—especially if this can be achieved in a risk, low NPY, and negative momentum) by 13.0% per
single liquid portfolio that requires infrequent rebal- year, with lower risk. For U.S. midcaps, Europe, Japan,
ancing, does not require expensive borrowing and/or and emerging markets (EM) we find similar results. The
shorting, and is not critically dependent on the inclusion alphas are stable across different economic regimes and
of microcaps. are robust for high levels of trading costs. This con-
It comes as no surprise therefore that books that servative investment strategy gives simultaneous posi-
present simple investment formulas are very popular tive exposures to the most profitable factor premiums
among investors. Value investing was popularized by and beats all Fama–French combinations of common
the investment book The Intelligent Investor by Graham investment strategies based on size, quality, value, and
[2005], who presented an intrinsic value formula in momentum.
the 1973 version of his book. In 1991, a strategy called The article is structured as follows. First we dis-
the Dogs of the Dow was proposed, which selects stocks cuss the data and global results. Then we analyze the
with the highest dividend yields from the Dow Jones results through time and directly compare the conser-
Industrial Average Index (see O’Higgins and Downes vative formula with single-sorted, double-sorted, and
[2000]). Greenblatt [2006] combines value (earnings triple-sorted Fama–French portfolios. We next apply a
to enterprise value) with a quality measure (return on battery of academic asset pricing tests to the formula and
capital) in his “Magic Formula,” which helps investors examine the robustness across economic regimes, size
identify the best 30 stocks. segments, and international markets. Finally, we ana-
More recently, Van Vliet and de Koning [2016] lyze the impact of trading costs. The Appendix further
explained the concept of low-risk investing to investors examines the impact of compounding.
and presented a new investment formula that selects 100
conservative stocks based on volatility, net payout yield DATA AND GLOBAL RESULTS
(NPY), and momentum. The aim of this article is to rig-
orously test this easy-to-implement Conservative Formula, We create a liquid universe of stocks by focusing on
which aims to give investors full and efficient exposure the largest 1,000 stocks at each point in time. Anomalies
to the strongest academic factor premiums. The for- or price patterns that are only strong among small-cap or
mula relies on simple price data and dividend data only, micro-cap stocks have little economic relevance, so we
which enables us to back test its persistency and robust- want to stay away from those. Restricting the universe
ness all the way back to 1929. The limited number of to large-caps is a prudent approach, and results are likely
simple factors also reduces the risk of p-hacking or factor to be more sustainable. We further reduce the risk of
fishing. The formula is applied to the largest 1,000 U.S. p-hacking by only using market data, so no accounting
stocks to ensure that the results are economically mean- or other data are used. Specifically, our formula only
ingful. Turnover is limited by rebalancing on a quarterly uses dividends, shares outstanding, and returns as input
basis. Although the formula is based on three simple data. For the United States we use the CRSP stock data-
investment criteria, we evaluate it considering the most base, which offers data going back to 1926.1 In addi-
advanced asset pricing models that have been proposed tion, we also use international data, unlike many studies
in the recent literature. For robustness, we also test the
conservative formula on U.S. midcaps (defined as the 1
second largest 1,000 stocks) and international equity  The series for the U.S. large stocks are also used in the book
High Return from Low Risk by van Vliet and de Koning [2016] (www.
markets. Finally, we also test for sensitivity to different paradoxinvesting.com). The U.S. midcap data and international
macroeconomic regimes and examine the impact of data and the robustness analyses in this study are not included in
trading costs. the book.

2    The Conservative Formula: Quantitative Investing M ade Easy Summer 2018
in the anomalies literature that only consider U.S. data 500 stocks with the highest volatility, those stocks with
and report results that do not necessarily carry over to the weakest combined scores on momentum and NPY.
other markets. We focus on three broad regions: Europe, This portfolio is used for comparing returns and for
Japan, and EM. Although the United States makes up testing the formula in a long–short context. Exhibit 1
about half of the global stock market, ignoring the other conceptually illustrates the selection mechanism of the
half would be a significant loss of information. There- conservative formula.
fore, we consider international samples consisting of Exhibit 2 shows the main findings for the U.S.
the 1,000 largest European, 1,000 largest Japanese, and and international equity markets. For all regions, the
1,000 largest EM stocks at each point in time. The data conservative formula portfolio exhibits much lower risk
are sourced from Factset and start in 1986 for Europe than the speculative portfolio, yet much higher returns.
and Japan and 1991 for EM. The European and Japa- The finding that ex post risk for the formula is con-
nese sample is derived from FTSE World Developed sistently low implies that historical volatility is a good
or S&P World Developed index constituents, and predictor of future risk, also on a portfolio level. The
the EM sample is derived from the S&P/IFC Global risk reduction varies from 50% for the United States
Emerging Markets index. Stocks that do not have at least and EM to about 35% in Europe and Japan. The return
36 months of return data available are not eligible for difference varies between 8.8% for Japan, 13.0% for the
inclusion in the various top 1,000 universes. We gather United States, 14.2% for Europe, and finally 17.2% for
monthly gross stock returns in local currencies, as well EM. This is a very large and internationally consistent
as in U.S. dollars, taking into account dividends, stock return spread and shows the global potential of factor-
splits, and other capital adjustments. Our stock return based investing. In other words, the conservative for-
data sources are Interactive Data Exshare, MSCI, and mula offers a consistently high return-to-risk ratio across
S&P/IFS, in that order. Monthly returns are truncated at all markets. These results are based on the full samples
500%. The free-f loat adjusted market capitalization data and are not yet related to alternative factor-based strat-
come from FTSE and S&P/IFC. Portfolio returns are egies, nor tested for significance. We will proceed by
in U.S. dollars. Stock-level volatility is calculated using taking a closer look at the U.S. stock market, for which
returns in local currency. Finally, as a robustness check, the richest data history exists. After this, we further
we also test the formula on the second largest 1,000 U.S. examine the results for the other markets.
stocks. This sample starts in 1970 because from that point
in time onward the CRSP database contains more than RESULTS OVER TIME
2,000 stocks in total.
At the end of each quarter the 1,000 largest A $100 investment in the 1929 U.S. equity market
stocks at that point in time are sorted into two groups, (CRSP value-weighted portfolio) would have grown to
each consisting of 500 stocks, based on their historical $246,000 dollars at the end of 2016. This translates into a
36-month stock return volatility (e.g., see Black [1993]; compounded annualized nominal return on investment
Blitz and Van Vliet [2007]; and Baker, Bradley, and of 9.3% per year. Note that actual returns to investors
Wurgler [2011]). Each stock is further ranked on its tend to be lower as a result of taxes, implementation
12 – 1 month price momentum ( Jegadeesh and Titman costs, and adverse market timing.2 The conservative for-
[1993]) and total NPY to shareholders (Boudoukh et al. mula portfolio exhibits a terminal wealth that is almost
[2007]). This shareholder yield consists of dividend yield a factor of 100 higher compared to the market, with a
and the net change in shares outstanding (calculated as compounded return on investment of 15.1% per year. By
the latest level divided by the 24-month average). The contrast, the speculative portfolio shows a meager com-
momentum and NPY ranks (1–500) are simply aver- pounded return of only 2.1% per year. An inspection of
aged and the top 100 of stocks are selected. To limit
turnover this procedure is repeated on a quarterly basis.
All factor scores are compared directly across sectors and 2
 Adverse market timing reduces average equity returns by
(for Europe and EM) countries, and the 100 stocks in 1.3% (Dichev [2007]), and implementation costs are often above
the final portfolio are equally weighted. We also create 1%. Furthermore, real returns are 2.8% lower due to inf lation over
an opposite speculative portfolio by selecting from the this sample period.

Summer 2018 The Journal of Portfolio M anagement    3

Exhibit 1
Visual Illustration of the Conservative Formula
+LJK13< /RZ13<
+LJK0RPHQWXP

&RQVHUYDWLYH

/RZ9RODWLOLW\

/RZ0RPHQWXP

+LJK0RPHQWXP

+LJK9RODWLOLW\

6SHFXODWLYH

/RZ0RPHQWXP
+LJK13< /RZ13<

Note: This exhibit visually illustrates the conservative formula.

Exhibit 3 shows that the dollar wealth development of COMPARISON WITH SINGLE-FACTOR
the conservative formula portfolio is robust over time. PORTFOLIOS
Exhibit 4 shows performance by decade. The
average return for conservative stocks is remarkably Results look stable through time, but how does
stable, with positive (>8%) returns during each indi- this compare to well-known alternative factor strate-
vidual decade. The absence of negative 10-year invest- gies such as size, value, and momentum? To this end
ment results is important because most investors have we directly compare the conservative formula to other
relatively short investment horizons (e.g., Benartzi and single-factor strategies using the same methodology:
Thaler [1995]). The return is also higher than the market equally weighting 100 top stocks from the universe
average in 8 out of 9 decades, with the exception being consisting of the largest 1,000 stocks. In the next sec-
the 1990s. The speculative portfolio shows consistently tion we compare the conservative formula with publicly
weak returns, falling short of the conservative portfolio available double-sorted and triple-sorted portfolios, and
during each decade. in the following section we control for factor exposures

4    The Conservative Formula: Quantitative Investing M ade Easy Summer 2018
Exhibit 2
Summary
8QLWHG6WDWHV 860LGFDSV
 ±  ±
&RQVHUYDWLYH
&RQVHUYDWLYH
 

 

  6SHFXODWLYH
6SHFXODWLYH

 
         

-DSDQ (XURSH
 ±  ±
&RQVHUYDWLYH
 

&RQVHUYDWLYH
 

 
6SHFXODWLYH
6SHFXODWLYH
 
         

(PHUJLQJ0DUNHWV
 ±


&RQVHUYDWLYH





6SHFXODWLYH



    

Notes: This exhibit shows the main results of this study. The total of largest 1,000 stocks are screened on historical three-year volatility. From the top 500
low-risk stocks the conservative formula selects the 100 stocks with the best combined 12 – 1 month momentum and NPY factor scores. The speculative
portfolio consists of stocks with completely opposite characteristics. Portfolios are equally weighted and rebalanced on a quarterly frequency. The exhibit shows
results for the United States, U.S. midcaps, Japan, Europe, and EM. The vertical axis denotes average compounded return, and the horizontal axis shows
the full sample volatility. Note that the EM vertical axis is adjusted.

using various specifications of different multifactor asset The simple return of small, value, and momentum strat-
pricing models. egies is a bit higher, but the compounded return shows a
Exhibit 5 shows that the conservative formula has different ranking. Interestingly, the return gap between
a return similar to that of strategies based on size, value, simple returns and compounded returns is only 0.4%
momentum, and NPY, but with markedly lower risk. for the formula. For most other factors this return gap

Summer 2018 The Journal of Portfolio M anagement    5

Exhibit 3
Development of Dollar Value over Time
9DOXHRI,QYHVWHGLQ

















    

&RQVHUYDWLYH 0DUNHW 6SHFXODWLYH

Notes: This exhibit shows the dollar development over time for the conservative and speculative portfolios. The active portfolios each consist of 100 stocks
and are equally weighted and rebalanced on a quarterly frequency. For comparison, we also show the U.S. stock market portfolio (CRSP value-weighted).
The results are gross of implementation costs and before taxes.

Exhibit 4
Performance across Decades
1HYHUD/RVW'HFDGH







V V V V V V V V V

±

&RQVHUYDWLYH 0DUNHW 6SHFXODWLYH

Notes: This exhibit shows the average 10-year return of the conservative and speculative portfolios. The active portfolios each consist of 100 stocks and
are equally weighted and rebalanced on a quarterly frequency. For comparison, we also show the U.S. stock market portfolio (CRSP value-weighted).
The results are gross of implementation costs and before taxes.

6    The Conservative Formula: Quantitative Investing M ade Easy Summer 2018
Exhibit 5
Conservative Formula versus Other Factors

± )RUPXOD 0DUNHW 6PDOO 9DOXH 0RPHQWXP /RZ9RO 13<

5HWXUQVLPSOH       
5HWXUQFRPSRXQGHG       
'LIIHUHQFH6LPSOH²       
&RPSRXQGHG
9RODWLOLW\       
'HULVNLQJIDFWRU       
6KDUSHUDWLRVLPSOH       
5HWXUQVDPHULVN       

Notes: This exhibit shows the portfolio with 100 conservative stocks ( formula) compared to other factors. These single-factor strategies equally weight the
top 100 stocks based on market capitalization of equity (small), book-to-market ratio (value), 12 – 1 month return (momentum), three-year volatility
(low vol), and NPY. Rebalancing is on a quarterly basis. Furthermore, the portfolios are scaled to the same volatility for a 1–1 comparison. The Sharpe
ratio is calculated as the simple average excess return (minus U.S. 30-day T-bill) divided by the volatility. All figures are annualized. These results are
gross of implementation costs and before taxes.

is above 1%, and in the case of small cap and value it COMPARISON WITH FAMA–FRENCH
even amounts to almost 4%. Asset pricing theory, such PORTFOLIOS
as the CAPM, does not specify the length of the invest-
ment horizon. Still, Exhibit 5 shows that the assumed Because a large portion of the return can be attrib-
investment horizon matters greatly. This horizon deter- uted to integrating multiple factors, we also compare
mines the average return and is for most investors longer the conservative formula with other multifactor strate-
than one month (simple return) but shorter than the gies (see Haugen and Baker [1996]). In particular, we
full sample (compounded return). Some of the return consider the commonly used portfolios of Fama and
drag is technical and driven by a higher volatility of French [1993, 2015]: double-sorted (2 × 3) portfolios
the underlying strategy. To better compare strategies based on size and book-to-market and based on size and
we de-risk all factors to the same volatility level as for momentum, and triple-sorted (2 × 4 × 4) portfolios on
the conservative formula (16.5%). For example, the size, book-to-market, profitability, and investments.4
de-risked market factor invests 88% in stocks and 12% Exhibit 6 shows that the conservative formula
in the risk-free asset to end up with the same risk as the has a higher Sharpe ratio compared to all other Fama–
formula. This approach converts Sharpe ratio differences French factor-combination strategies. The small-winner
into return differences.3 The conservative formula has (SW) and small-value (SV) strategies have the highest
the highest return per unit of risk compared to all other Sharpe ratios among the double-sorted portfolios, but
factors. The formula, which consists of three factors, still they all fall short of the conservative formula. The
also has a higher return compared to each of its three triple-sorted Fama–French portfolios offer more detailed
components: low volatility, momentum, and NPY. On insight. Because of limited data availability for prof-
an equal-risk basis, the return differences amount to itability and investments, the sample for triple-sorted
2%–3% per year. Because the methodology is exactly the portfolios starts in July 1963. Small-value stocks with
same—100 stocks with equal weighting and quarterly
rebalancing—this difference can only be attributed to 4
 The Fama–French portfolios include all stocks available in
the integration and diversification benefits of combining the CRSP universe, but the universe is divided based on the NYSE
multiple factors into one strategy. median. The Fama–French small-cap portfolio (below median)
grows from 227 stocks to 2,329 at the end of the sample. The
large-cap portfolio (above median) grows from 251 to 856 at the
3
 This de-risking explains most of the return drag resulting end of the sample. To further limit the impact of micro caps or
from compounding, but not all (see Exhibit A1). small caps driving results, the Fama–French portfolio returns are
value weighted.

Summer 2018 The Journal of Portfolio M anagement    7

Exhibit 6
Conservative Formula versus Fama–French Portfolios
6KDUSH5DWLRV±YHUVXV6L]H%W0 6KDUSH5DWLRV±YHUVXV6L]H0RP
 
 
 
 
 
 
 
 
 
)RUPXOD 6* 61 69 %* %1 %9 )RUPXOD 6/ 61 6: %/ %1 %:

6KDUSH5DWLRV±YHUVXV6L]H%W03URILWDELOLW\










)RUPXOD
























/RZ3URI

+LJK3URI
/RZ3URI

+LJK3URI
/RZ3URI

+LJK3URI
/RZ3URI

+LJK3URI
/RZ3URI

+LJK3URI
/RZ3URI

+LJK3URI
/RZ3URI

+LJK3URI
/RZ3URI

+LJK3URI
6PDOO*URZWK 6PDOO%0 6PDOO%0 6PDOO9DOXH %LJ*URZWK %LJ%0 %LJ%0 %LJ9DOXH

6KDUSH5DWLRV±YHUVXV6L]H%W0,QYHVWPHQWV










)RUPXOD
/RZ,QY


+LJK,QY
/RZ,QY


+LJK,QY
/RZ,QY


+LJK,QY
/RZ,QY


+LJK,QY
/RZ,QY


+LJK,QY
/RZ,QY


+LJK,QY
/RZ,QY


+LJK,QY
/RZ,QY


+LJK,QY

6PDOO*URZWK 6PDOO%0 6PDOO%0 6PDOO9DOXH %LJ*URZWK %LJ%0 %LJ%0 %LJ9DOXH

Notes: This exhibit shows the Sharpe ratios of the conservative formula with double-sorted and triple-sorted portfolios. The double-sorted portfolios are based
on size (ME) and book-to-market ratio (BM) and size (ME) and momentum (12 – 1 month return). Profitability is defined as one-year operational prof-
itability, and investment is defined as one-year change in total assets. The sample starts in January 1929 for the double-sorted portfolios and in July 1963
for the triple-sorted portfolios. Fama–French portfolios are value weighted and include all stocks in the CRSP database, including small and micro caps.
The average number of included stocks is 2,569 stocks over the 1929–2016 period, of which 668 classify as above median NYSE. All figures are
annualized. These results are gross of implementation costs and before taxes.

8    The Conservative Formula: Quantitative Investing M ade Easy Summer 2018
high profitability are among the best performing port- in the formula. In general, larger stocks tend to have
folios, with a Sharpe ratio of around 0.75. This comes lower volatility, and smaller stocks tend to have higher
close to the Sharpe ratio of 0.84 of the conservative volatility. Over the long run, value stocks are more vola-
formula over this sample period, but it still falls a bit tile and have higher betas than growth stocks (see Ang
short. Moreover, the Fama–French portfolio consists of and Chen [2007]), and the formula loads negatively on
only 26 stocks with an average market capitalization of the HML factor over the full sample period. Although
$350 million as of December 2016. This specific port- NPY is a price-based measure that tilts the portfolio to
folio may look attractive on paper, but it represents less value (book-to-price), this is countered by the fact that
than 0.05% of the total U.S. stock market, which means low volatility is strongly negatively exposed to HML in
that large investors have limited opportunity to invest the earlier part of the sample (1930s) and to momentum
in this portfolio. By contrast, the conservative portfolio most of the time. Therefore the formula exhibits a nega-
consists of 100 stocks with an average market capital- tive exposure to value over the full sample period. When
ization of $35 billion as of December 2016. This makes momentum (MOM) is added, the 4-factor alpha drops
up almost 20% of the total U.S. market. The lowest to 8.8%, which is expected because momentum is one
Sharpe ratios are earned by small growth firms with low of the three building blocks of the formula. The alpha
profitability. We also observe that the positive relation remains highly significant, however.
between profitability and return is not consistent in the Panel B of Exhibit 7 shows that over the more
large-cap segment. Finally, small value stocks with low recent (1963–2016) period the CAPM alpha of the for-
levels of investment earn high Sharpe ratios. Within the mula remains similar, at 14.0%. During this period value
small-cap segment the relation between investment and stocks are less risky and volatile. At 13.0%, the 3-factor
risk-adjusted return is not linear. In the large-cap seg- alpha is not much lower because the impact of a posi-
ment this relationship is more persistent, and the large tive loading on the HML value factor is largely offset
value portfolio with low levels of investment earns a by a negative loading on the SMB size factor. Fama and
high Sharpe ratio. However, none of these triple-sorted French [2015] proposed to extend their classic 3-factor
portfolios is able to match the Sharpe ratio of the con- model to a 5-factor model by augmenting it with prof-
servative formula. itability (RMW) and investment (CMA) factors, for
which data are available from 1963. Panel B shows that
CONTROL FOR FACTOR EXPOSURES the conservative strategy has positive loadings on both
these new factors, which reduces the alpha to 8.8%.
The return spread between the conservative port- When we additionally control for the momentum factor,
folio and speculative portfolio is 13.0% per year. How we get another positive loading, and the 6-factor alpha is
much of this return spread can be attributed to exposure further reduced to 3.3%. In sum, the CMS strategy gives
to the classic Fama–French long–short factors or expo- positive and significant exposure to the value (HML),
sure to newer factors? To answer this question we apply momentum (MOM), profitability (RMW), and invest-
time-series spanning tests to the conservative minus ment (CMA) factors at the same time. These are factor
speculative (CMS) portfolio. Some factors are available tilts most investors would like to see in their portfolio
since 1929, whereas other factors only come available at because all these factors are known to be associated with
a later date. Exhibit 7 shows various regression outcomes higher returns. After controlling for all these factors,
using the longest available data histories. a small but still statistically significant positive alpha
Panel A of Exhibit 7 shows that the CMS portfolio remains. Unreported results show that this is mainly
has a full-sample CAPM alpha of 13.9%. This goes along driven by the inclusion of NPY in the formula.
with a large negative exposure to the market factor, In Panel C of Exhibit 7 we subject the CMS port-
which is not surprising given that the formula is long folio to the recently proposed q-factor model of Hou,
low-volatility stocks and shorts high-volatility stocks. Xue, and Zhang [2015], who argued that this model is
The 3-factor alpha is somewhat higher, mainly caused superior to the Fama–French models.5 We find large
by a negative size (SMB) exposure. The positive expo- positive loadings on the investment (IA) and profitability
sure to large caps (and negative exposure to small caps)
is caused by the inclusion of the low-volatility factor 5
 We thank Prof. Zhang for kindly providing the data.

Summer 2018 The Journal of Portfolio M anagement    9

Exhibit 7
Conservative Formula Dissected Using Multifactor Models
3DQHO$)DPDDQG)UHQFK>@&DUKDUW>@
± $OSKD 0NW5) 60% +0/ 020
&RHIILFLHQW  ±
WYDOXH  ±
&RHIILFLHQW  ± ± ±
WYDOXH  ± ± ±
&RHIILFLHQW  ± ± ± 
WYDOXH  ± ± ± 

3DQHO%)DPDDQG)UHQFK>@
± $OSKD 0NW5) 60% +0/ 50: &0$ 020
&RHIILFLHQW  ±
WYDOXH  ±
&RHIILFLHQW  ± ± 
WYDOXH  ± ± 
&RHIILFLHQW  ± ±   
WYDOXH  ± ±   
&RHIILFLHQW  ± ±    
WYDOXH  ± ±    

3DQHO&+RX;XHDQG=KDQJ>@
± $OSKD 0NW5) 0( ,$ 52(
&RHIILFLHQW  ± ±  
WYDOXH  ± ±  

3DQHO'$45
± $OSKD 0NW5) 60% +0/ 40- %$% 020
&RHIILFLHQW  ± ±   
WYDOXH  ± ±   
&RHIILFLHQW  ± ±    
WYDOXH  ± ±    

Notes: This exhibit shows the conservative minus speculative (CMS) portfolio controlled for multiple factor models. All factors are from the online data
library of Kenneth French. The sample for the three-factor and four-factor model starts in January 1926, whereas the five-factor and six-factor models start
in July 1963; both end at December 2016.

(ROE) factors in the q-factor model, but after control- Exhibit 7 we observe large positive loadings on both
ling for these exposures an alpha of 4.1% per annum the QMJ and BAB factors, implying that the formula
remains. Thus, the formula is also attractive for investors also provides exposures toward these factor premiums.
looking for efficient exposure (and some more) to the We also observe that only the combination of the classic
factors in the q-factor model. value and momentum factors of Fama–French and the
In the final spanning test we augment the Fama– two new AQR factors is powerful enough to render the
French three-factor model and the Carhart [1997] four- alpha of the CMS portfolio insignificant, although even
factor model with the quality (QMJ) factor of Asness, in this case it does remain positive. It is good to keep in
Frazzini, and Pedersen [2013] and the betting-against- mind here, however, that the QMJ factor consists of over
beta (BAB) factor of Frazzini and Pedersen [2014]. 20 underlying variables, which are aggregated into one
We call this the AQR model, as the authors of these combined quality score. The formula, on the other hand,
two studies are employed at that firm. In Panel D of only uses three variables, which do not even require

10    The Conservative Formula: Quantitative Investing M ade Easy Summer 2018
Exhibit 8
U.S. Midcap and International Results
3DQHO$)DFWRU%HWDV&06
0NW5) 60% +0/ 020 50: &0$
86/DUJHFDSV ± ±    
860LGFDSV ± ±    
(XURSH ± ±    
-DSDQ ± ±    
(0 ± ±    

3DQHO%)DFWRU3UHPLXPV
0NW5) 60% +0/ 020 50: &0$
86/DUJHFDSV      
860LGFDSV      
(XURSH  ±    
-DSDQ      
(0      

3DQHO&$OSKDV&06
3UHPLXP &$30 )0 )0 )0 )0
86/DUJHFDSV      
860LGFDSV      
(XURSH      
-DSDQ      
(0      

Notes: This exhibit shows the U.S. small- and midcap and international results for the conservative minus speculative (CMS) portfolio. The U.S. large-
cap portfolio consists of the largest 1,000 stocks, and the U.S. midcap portfolio consists of the second largest 1,000 stocks. For each international market the
largest 1,000 stocks are used to construct the two portfolios. The U.S. sample starts in January 1970, the European and Japanese samples start in January
1986, and the EM sample starts in January 1993. For Europe and Japan regional Fama–French control factors are used and backfilled prior to November
1990 with U.S. long–short factors. For EM, global long–short factors are used. Panel A shows the factor betas, Panel B shows the factor premiums, and
Panel C shows the CMS premium and the factor-corrected alphas of the four regional strategies. The 3-factor model (3FM) consists of SMB and HML.
The 4-factor model (4FM) adds momentum to the 3FM. The 5-factor model (5FM) consists of SMB, HML, RMW, and CMA. The 6-factor model
(6FM) adds momentum to the 5FM.
Asterisks are used to indicate significance at 5% (*) or 1% (**) level.

accounting data. Thus, the formula is able to match or CRSP exceeds 2,000 from 1970 onward, when AMEX
outperform the most advanced asset pricing factors with stocks were added to the database, so for this analysis
much lighter data input requirements. we consider the 1970–2016 sample period. For com-
parison purposes we also show the alphas of the con-
U.S. MIDCAP AND INTERNATIONAL servative formula for the largest 1,000 stocks over this
RESULTS period. Exhibit 8 shows that the midcap conservative
minus speculative portfolio exhibits factor betas that are
The largest 1,000 stocks represent about 90% of very similar to those of the large-cap CMS portfolio.
total U.S. market capitalization at the end of 2016. The Exhibit 8 shows positive factor tilts toward the four fac-
market capitalizations for the next 1,000 stocks range tors with the highest premiums: HML, MOM, RMW,
from $3.0 billion to $0.5 billion. Although this group and CMA. The midcap CMS premium is 13.0%, which
of stocks represents less than 10% of total market capi- is about 3% higher than the CMS premium in the large-
talization, it provides an interesting robustness test for cap segment. The multifactor alphas are all statistically
the conservative formula. The total number of stocks in significant, ranging between 6% and 18% per annum.

Summer 2018 The Journal of Portfolio M anagement    11

Exhibit 9
Conservative Formula Alpha across Economic Regimes
$OSKDVDFURVVUHJLPHV











±

([SDQVLRQ

5HFHVVLRQ

/RZ

+LJK

)DOOLQJ

5LVLQJ

/RZ

+LJK

)DOOLQJ

5LVLQJ
$OO 1%(5 %RQG<LHOG< &UHGLW6SUHDG

Notes: This exhibit shows the conservative formula long-only CAPM alpha (6.2%) and the CAPM alphas across different regimes for the period
1929–2016. In total, 10 economic regimes are defined: NBER expansion and recession, low/high 10-year bond yields, falling/rising 10-year bond yields,
low/high credit spreads (defined as Baa-Aaa), and falling/rising credits spreads.

The pattern in alphas is very similar compared to the Exhibit 10

results for the largest 1,000 U.S. stocks. Although After Trading Net Returns (%)
returns are higher, expected implementation costs are
also higher in this particular market segment, so after- 86 (XURSH -DSDQ (0
cost return differences are likely to be smaller. *URVV5HWXUQ    
We also test for global robustness of the conserva- 7XUQRYHU4XDUWHUO\    
tive formula by considering the European, Japanese, and (VWLPDWHG7UDGLQJ    
emerging stock markets. To calculate multifactor alphas &RVWV²/RZ
(VWLPDWHG7UDGLQJ    
we use local factors provided in the Kenneth French data
&RVWV²+LJK
library for Europe and Japan and global factors for EM. 7UDGLQJ&RVWVKLJK    
As shown in Exhibit 2, the conservative portfolio consis- DVRI5HWXUQ
tently outperforms the speculative portfolio in all regions. 1HW5HWXUQ    
Exhibit 8 shows the betas of the four regional CMS port- KLJKWUDGLQJFRVWV
folios to the six different factors. For all regions we find a
negative market beta exposure and a negative size (SMB) Notes: This exhibit shows the impact of trading costs on the net returns of
exposure, which is consistent with the U.S. results. We the conservative formula for the four regional strategies. The quarterly turn-
over is shown, which is calculated as single-counted. For transaction costs,
also find positive factor tilts to profitability (RMW), a low level of 10 bps per dollar traded and a high level of 30 bps per dollar
momentum (MOM), investment (CMA), and value traded are assumed. For EM, these figures are 20 and 60 bps, respectively.
(HML) for all regions, which is again consistent with Total trading costs include both market impact and broker costs. The net
return assumes the conservative high estimate of transaction costs.
the U.S. results. This is a desirable feature because each
of these four factors carries a positive premium in each
region. Panel B shows that the value premium is stable at low in Japan (2%). The low momentum premium in
around 4%, whereas the new Fama–French profitability Japan is a known phenomenon (see, e.g., Griffin, Ji, and
and investment factors have premiums of around 3%. Martin [2003]; Asness [2011]). Panel C shows the return
Momentum carries the highest premium of 7%, but this spread and the single-factor and multifactor alphas for
varies between very high in Europe (10%) and quite the regional strategies. The CAPM alphas are all posi-

12    The Conservative Formula: Quantitative Investing M ade Easy Summer 2018
tive, with the highest alpha in EM and the lowest alpha (instead of monthly) rebalancing frequency. This reduces
in Japan. Most of the high returns can be explained by a turnover, which is the most important driver of transac-
positive exposure to the four factors that give the highest tion costs (Novy-Marx and Velikov [2016]). Also, the
premiums: value, momentum, profitability, and invest- focus is on the largest 1,000 U.S. stocks. Nevertheless,
ment. When the momentum factor is added to the Fama– implementation costs could still be significant and nega-
French three-factor and five-factor models, we see the tively affect net results. Exhibit 10 shows for each of the
largest part of the alpha being explained. Interestingly, regional strategies the amount of (single-counted) turn-
the momentum factor carries the highest premium, but over needed to implement the conservative formula. The
it is also most difficult to implement in practice because turnover levels are all very similar, at around 30% per
it requires very high levels of turnover. That is one of quarter, which translates into an average stock holding
the reasons why Fama and French prefer to not include period of slightly less than one year. The turnover can
this factor in their multifactor asset pricing models. The easily be brought down with more sophisticated trading
conservative formula offers positive and significant expo- rules—for instance, rules that do not immediately sell a
sure to the momentum factor, but because momentum stock if its rank drops to number 101 out of 1,000, but
is integrated with other factors, it is able to achieve this stay invested in such positions until they drop beyond a
with a relatively modest level of turnover. This lower certain sell threshold. We will refrain from such sophis-
turnover is a desirable feature for any practically imple- tications here and work with the raw turnover levels to
mentable investment strategy. keep matters simple. Frazzini, Israel, and Moskowitz
[2012] estimated transaction costs to be less than 20 bps
MACRO RISK AND TRADING COSTS for U.S. and international large-cap and midcap stocks.
We show results for a low assumed transaction cost level
It could be that systematically investing in stable of 10 bps (20 bps for EM) and a high assumed transac-
stocks with high net payouts that are in a positive trend tion cost level of 30 bps (60 bps for EM). For the most
gives exposure to some form of macroeconomic risk. For conservative estimate applied to EM we find that net
example, one might be concerned that these stocks are returns go down up to 1.5% per year, which is not much
sensitive to the economic cycle and underperform when compared to the raw return of the formula. In all cases,
the economy is in recession. Or they might be sensitive trading costs amount to less than 10% of the total return,
to interest rate changes or credit cycles. so more than 90% of the returns remain for investors.
Exhibit 9 shows that the conservative-formula long
portfolio (100 stocks) has an average CAPM alpha of CONCLUSION
6.2%, that this alpha is spread evenly across expansions
and recessions, and that it is stable across high and low We propose a conservative formula that is designed
levels of interest rates. The alpha is sensitive to changes to make quantitative investing easy for investors. Using
in interest rates, dropping to 3.3% when rates are rising. three simple investment criteria, many different factor
This rate sensitivity is a known feature of low-volatility premiums can be captured with a single investment
stocks (see, e.g., Baker and Wurgler [2012]; De Franco, strategy consisting of 100 liquid stocks. The conservative
Monnier, and Rulik [2017]). Compared to a generic formula uses past returns and NPY only, which means
low-volatility approach, the formula does mitigate the that no accounting or other data sources are needed.
sensitivity to interest rate changes by also including other Despite its simplicity, the strategy gives simultaneous
factors, in particular momentum. In any case, the alpha positive exposure to well-known factors such as low beta,
remains significant, with all t-statistics being above 3.5. value, quality, and momentum. The returns are consis-
There is no clear relation with the credit cycle, as neither tent over time and across international stock markets and
credit spread levels nor changes seem to have a large are present in U.S. small- and midcap stocks. In sum, this
impact on results. simple formula can be used by active investors as a way
Finally, we estimate the impact of transaction costs to profit directly from half a century of academic insights
on results. The formula is designed to suffer limited by applying one simple investment formula.
impact from transaction costs by assuming a quarterly

Summer 2018 The Journal of Portfolio M anagement    13

A pp e n d i x
Exhibit A1
Compounding Matters, Even When Strategies Have the Same Volatility

± )RUPXOD 0DUNHW 6PDOO 9DOXH :LQQHU /RZ9RO 13<

/HYHUDJH)DFWRU       
6LPSOH5HWXUQ       
&RPSRXQGHG5HWXUQ       
7RWDO5HWXUQ'UDJ ± ± ± ± ± ± ±
&DXVHGE\9RODWLOLW\ ± ± ± ± ± ± ±
&DXVHGE\&RUUHODWLRQ       

Notes: This exhibit disentangles the difference between compounded returns and simple returns by splitting the return drag in volatility and correlation
components. When all strategies are brought to the same risk level, differences between simple and compounded returns could remain but are solely caused
by non-zero time-series correlation. Without time-series correlation, we expect a return difference of 0.5 × variance (0.5 × 16.5%^2) of 1.4% per annum.
The conservative formula has a return drag of only –0.4%, which is caused by a favorable (negative) autocorrelation pattern. By contrast, small and value
strategies have positive autocorrelation, which increases the return by about 2.5%.

ACKNOWLEDGMENTS Black, F. 1993. “Beta and Return: Announcements of the

‘Death of Beta’ Seem Premature.” The Journal of Portfolio Man-
The authors would like to thank Guido Baltussen and agement 20 (1): 8–18.
Jan de Koning for valuable discussions and Milan Vidojevic
Blitz, D. C., and P. Van Vliet. 2007. “The Volatility Effect.”
for programming assistance.
The Journal of Portfolio Management 34 (1): 102–113.

REFERENCES Boudoukh, J., R. Michaely, M. Richardson, and M.

R. Roberts. 2007. “On the Importance of Measuring Payout
Ang, A., and J. Chen. 2007. “CAPM over the Long Run: Yield: Implications for Empirical Asset Pricing.” The Journal
1926–2001.” Journal of Empirical Finance 14 (1): 1–40. of Finance 62 (2): 877–915.

Asness, C. 2011. “Momentum in Japan: The Exception That Carhart, M. M. 1997. “On Persistence in Mutual Fund Per-
Proves the Rule.” The Journal of Portfolio Management 37 (4): formance.” The Journal of Finance 52 (1): 57–82.
67–75.
De Franco, C., B. Monnier, and K. Rulik. 2017. “Interest
Asness, C., A. Frazzini, and L. H. Pedersen. “Quality Minus Rate Exposure of Volatility Portfolios.” The Journal of Index
Junk.” Working paper, 2013. Investing 8 (2): 53–67.

Baker, M., B. Bradley, and J. Wurgler. 2011. “Benchmarks Dichev, I. D. 2007. “What Are Stock Investors’ Actual His-
as Limits to Arbitrage: Understanding the Low-Volatility torical Returns? Evidence from Dollar-Weighted Returns.”
Anomaly.” Financial Analysts Journal 67 (1): 40–54. American Economic Review 97 (1): 386–401.

Baker, M., and J. Wurgler. 2012. “Comovement and Predict- Fama, E. F., and K. R. French. 1993. “Common Risk Fac-
ability Relationships between Bonds and the Cross-Section tors in the Returns on Stocks and Bonds.” Journal of Financial
of Stocks.” The Review of Asset Pricing Studies 2 (1): 57–87. Economics 33 (1): 3–56.

Benartzi, S., and R. H. Thaler. 1995. “Myopic Loss Aver- ——. 2015. “A Five-Factor Asset Pricing Model.” Journal of
sion and the Equity Premium Puzzle.” The Quarterly Journal Financial Economics 116 (1): 1–22.
of Economics 110 (1): 73–92.

14    The Conservative Formula: Quantitative Investing M ade Easy Summer 2018
Frazzini, A., R. Israel, and T. J. Moskowitz. “Trading Costs ——. “Replicating Anomalies.” Working paper, 2017.
of Asset Pricing Anomalies.” Working paper, 2012.
Jegadeesh, N., and S. Titman. 1993. “Returns to Buying
Frazzini, A., and L. H. Pedersen. 2014. “Betting against Winners and Selling Losers: Implications for Stock Market
Beta.” Journal of Financial Economics 111 (1): 1–25. Efficiency.” The Journal of Finance 48 (1): 65–91.

Graham, B. The Intelligent Investor, revised ed. New York: Novy-Marx, R., and M. Velikov. 2016. “A Taxonomy of
Harper Collins, 2005. Anomalies and Their Trading Costs.” The Review of Financial
Studies 29 (1): 104–147.
Greenblatt, J. The Little Book That Beats the Market. Hoboken,
NJ: John Wiley & Sons, 2006. O’Higgins, M. B., and J. Downes. Beating the Dow—A High-
Return, Low-Risk Method for Investing in the Dow Jones Industrial
Griffin, J. M., X. Ji, and J. S. Martin. 2003. “Momentum Stocks with as Little as $5,000, revised and updated. New York:
Investing and Business Cycle Risk: Evidence from Pole to Harper Business, 2000.
Pole.” The Journal of Finance 58 (6): 2515–2547.
Van Vliet, P., and J. de Koning. High Returns from Low Risk: A
Harvey, C. R., Y. Liu, and H. Zhu. 2016. “... And the Cross- Remarkable Stock Market Paradox. Hoboken, NJ: John Wiley &
Section of Expected Returns.” The Review of Financial Studies Sons, 2016.
29 (1): 5–68.

Haugen, R. A., and N. L. Baker. 1996. “Commonality in the To order reprints of this article, please contact David Rowe at
Determinants of Expected Stock Returns.” Journal of Financial d.rowe@ pageantmedia.com or 646-891-2157.
Economics 41 (3): 401–439.

Hou, K., C. Xue, and L. Zhang. 2015. “Digesting Anomalies:

An Investment Approach.” Review of Financial Studies 28 (3):
650–705.

Summer 2018 The Journal of Portfolio M anagement    15