SSRN Id4000418
SSRN Id4000418
SSRN Id4000418
India
January 2022
ABSTRACT
We show that monthly rebalanced, equal-weighted, long-only winner portfolios, drawn from the
top 200 stocks in India, built using systematic rules that underpin popular factors of momentum, low
volatility and quality deliver alpha for the period under study. The market exposure is significant across
all the style strategies we looked at. Therefore, correlations between the strategies are significantly
higher than those observed for academic factor returns. We include alternate calculation methodologies
for some factors and find that not all implementations of factor strategies are the same. Not all
strategies have high turnover. Indeed, strategies like low volatility and quality show fairly low turnover.
Factor exposure persistence over time varies across strategies and persistence should be considered
when implementing factor-style strategies. We also find that size and sectoral preferences of factors
are dynamic and could reduce perceived diversification benefits. Finally, we show that alpha for
momentum, low volatility and quality strategies survives real-world implementation costs. While
winner portfolios using momentum, low volatility, and quality rank higher than the broad S&P 200
Index over the period under study, there is not one factor-style that is a consistent winner.
‡
The authors thank Pim van Vliet for valuable feedback on an earlier draft of this paper.
1
Electronic copy available at: https://ssrn.com/abstract=4000418
1 Introduction
Recently there has been considerable excitement around factor investing1 in India. Domestic asset managers
are offering funds tracking these ‘strategy indices’ to retail investors2 . Internationally, factor investing has been
more established with a longer track record. The field has seen a veritable explosion of academic literature
on factor models originating with the work of Fama (1972). There is abundant academic evidence for the
existence of various factor premiums in equity markets, such as the value, momentum, low-risk, and quality
premiums (Fama and French (1988, 2011, 2012); Jegadeesh and Titman (1993); Asness et al. (2013); Ang et al.
(2008); Frazzini and Pedersen (2014); Asness et al. (2018)). The evidence that these factors exist in multiple
emerging markets is also considerable (de Groot et al. (2012b); Cakici et al. (2013, 2016); Hanauer and Linhart
(2015)). Standard academic factor portfolios, following Fama and French (1988), take hypothetical long po-
sitions in stocks with attractive characteristics and simultaneously take hypothetical short positions in stocks
with unattractive characteristics. Factor returns are computed from these market-neutral portfolios. While
theoretically elegant, this long-short approach is practically not feasible in most markets. Practical expressions
of factor investing are not market-neutral long-short portfolios. Rather they are usually long-only portfolios
with significant market exposure. Most practical factor-based portfolios are more of a ‘factor-tilt’ effort than a
factor replicating investment with significant divergence in returns between the two.
Factor models are ‘empirical asset pricing models’ and rely on historical data to show possible out-performance
relative to a benchmark. Rational asset pricing models interpret these premia as the compensation to bear sys-
tematic risks not captured by the market factor (Fama and French, 1993). Behavioural theories view such
premia arising from arbitrage or behavioral biases (De Bondt and Thaler, 1985; Lakonishok et al., 1994; Daniel
and Titman, 1997; Barberis et al., 1998). Since the 1980s, factors like size, value, and momentum were shown
to deliver returns that could not be explained by capital asset pricing models. Hundreds of ‘factors’ allegedly
offering a unique source of return have emerged since then. With cheap, powerful computing power, increas-
ingly accessible machine learning algorithms, easy-to-code high-level computer languages, associated statistical
libraries, and access to data sources, brute force is used to find correlations between millions of variables in
the hunt for ‘factors’. This has resulted in a large number of false positives3 . While the more obvious and
egregious false positives are easily dealt with, it is very difficult to learn whether a factor has a true economic
basis. “There are three kinds of lies: lies, damned lies, and statistics”. Unfortunately for factor investing, in its
empirical nature lies its Achilles heel.
1
Also called styles, strategy, smart-beta or alternative beta investing, especially in marketing brochures and the
popular financial press.
2
for example, UTI Nifty200 Momentum 30 Index Fund from UTI Mutual Fund tracking the Nifty200 Momentum
Index. https://utimf.com/nfo/uti-nifty-200-momentum-30-index-fund/
Kotak Nifty Alpha 50 ETF from Kotak Mutual Fund tracking the Nifty Alpha 50 Index. https://www.kotakmf.com/
Products/nfo/etf-funds/Kotak-Nifty-Alpha-50-ETF/Dir-G
3
Arnott et al. (2018) highlight an example using the first three letters of a firm’s ticker symbol. They find a strategy
that buys stocks listed on the New York Stock Exchange with the letter ‘S’ as the third letter and shorts stocks with the
letter ‘U’ as the third letter has spectacular performance backtests and meets all the common statistical tests. However,
this is an instance of ‘correlation without causation. See Harvey and Liu (2017) for more examples of false ‘discoveries’.
Jan 2022 2
Electronic copy available at: https://ssrn.com/abstract=4000418
Internationally there is a vigorous and heated debate around taming the ‘factor zoo’4 (see Harvey et al.
(2015); Mclean and Pontiff (2016); de Prado (2018); Blitz and Hanauer (2020); Bartram et al. (2020); Hou et al.
(2018) amongst others). This debate, like the hunt for factors, is empirically based. In India, however, there is
little formal research on factor investing. Ansari and Khan (2012); Joshipura and Joshipura (2016); Agarwalla
et al. (2017); Raju (2019) are a few examples of research into factors in India. The increasing interest in factor
investing requires more detailed India-specific empirical analysis to separate evidence from myth. There are
significant implementation issues in India for strategies that short individual stock. So any examination of
factor strategies in the Indian context necessarily entails examining the long-only legs. We examine empirical
evidence to answer questions such as: Do long-only factor strategies outperform the broader market? How much
factor/market exposure do these strategies have? What is the turnover experienced? Do such strategies persist
over time? What is their size/sectoral composition? Do they survive in real-world implementations?
We differ from traditional academic studies that examine the contribution of the long and short legs of
portfolios from academic long/short factor datasets. Practitioners in India rightly point out that such academic
portfolios include a number of small and micro-caps. The theoretical returns consequently shrink significantly
in implementation. Unfortunately, there is little research in the performance of long-only factor portfolios that
are implementable and where costs of implementation are factored in. This paper addresses this gap. We build
decile portfolios from the most liquid and traded universe of stocks in the Indian equity market. The top decile
portfolio is our factor-strategy portfolio. This approach makes our findings more aligned to real-world imple-
mentations of factor strategies and, therefore, of direct interest to practitioners. We use statistical techniques to
examine potential alpha, factor exposure for each implementation, turnover, persistence and correlations, size
and sectoral bias and finally whether the alpha survives implementation costs. Wherever possible we use indus-
try definitions so that the results are meaningful for practitioners. The broad and holistic approach provides a
unique evidence-based insight on long-only strategies in the Indian context.
Using time-series data from December 2006 for the S&P BSE 200 constituents, we show that some long-only
top decile portfolios deliver alpha in India. Factor-tilt strategies have varying exposure to underlying factors.
Not all implementations of a factor are equal. In fact, in some implementations, the desired factor exposure is
completely absent. Market exposure is significant across all the strategies we looked at, and therefore correla-
tions between the strategies are significantly higher than those indicated in academic research. Turnovers vary
significantly between strategies. Not all turnovers are high. Factor exposure persistence varies across strategies
and should be an important consideration when choosing strategies. Finally, we show that for some strategies
alphas do survive real-world implementation after costs. Our analysis shows that factor strategies are not the
panacea or silver bullets for investors. Like any investment, factor strategies have risks such as volatility, po-
tentially long periods of under performance, and changing factor exposure.
This paper is one of the first to empirically explore several possible factor strategies using a consistent
4
a term coined by Cochrane (2011)
Jan 2022 3
Electronic copy available at: https://ssrn.com/abstract=4000418
methodology within the Indian context. By building and examining portfolios implementable in the real-world,
the paper contributes pragmatic and real-world insight to the growing debate on the attractiveness and viabil-
ity of factor strategies in India using an evidence based approach. The framework developed could serve as a
process to evaluate factor-tilt strategies by various industry participants including regulators, index providers,
asset managers, wealth advisers and investors. More importantly, we hope this paper will encourage further
research into factor investing within the Indian context and allow a similar empirically-driven robust debate
around factor investing in India as ongoing in developed equity markets.
The rest of the paper is organized as follows: Section 2 covers the literature, in Section 3, we discuss our
methodology and data; in Section 4, we present our results; and we conclude in Section 5.
2 Literature Review
Amongst the ever-growing factor zoo, we take a conservative approach in our choice of factors to investigate
in this study. In the main, we focus on the long-established factors that have been exhaustively tested across
markets and periods. There is a wide range of academic research on factors, some of which we have referenced
earlier. The size, value, and momentum factors laid out by Fama and French (1993, 2011) and Jegadeesh and
Titman (1993) are among the most established factors. In the Indian market, Agarwalla et al. (2013, 2017)
show strong evidence of the existence of these factors in their exploration and construction of the Fama-French
Four-Factor model. Agarwalla et al. (2014) showed the existence of the ‘betting-against-beta’ factor in India5 .
Joshipura and Joshipura (2016); Ali et al. (2021) look at the low-volatility premium in India.
Long-only factor strategy returns are not the same as theoretical factor returns (Arnott et al., 2017). While
marketing brochures tend to gloss over the technical nuances that differentiate the two implementations, any
serious consideration of factor-tilt performance must account for the differences between the academic factors
and real-world implementation. Huij et al. (2014) compare long-only and long-short implementations of factors
and conclude that “investors should carefully consider the pros and cons of long-only and long-short approaches
when implementing factor investing”. Blitz et al. (2020) found that most of the value-add from long-short factor
strategies comes from the long legs. Raju and Chandrasekaran (2019) analysed a large-cap long-only momentum
factor-tilt portfolio in India and concluded that such a strategy does show partial exposure to the momentum
factor. Raju (2019) looked at the quality factor tilt in India, adapting the work done by Asness et al. (2018),
again in the large-cap space.
In addition to size, value, momentum, low-risk, and quality, we include a couple of factor-tilt indices that
popular index providers have created (high beta and growth). Factor portfolios can be constructed in many
different ways. For example, a value strategy can be based on just the market-to-book ratio or a combination of
value parameters such as price-to-earnings or dividend yield. Or a momentum strategy constructed with a look-
5
Following the approach outlined by Frazzini and Pedersen (2014).
Jan 2022 4
Electronic copy available at: https://ssrn.com/abstract=4000418
back of 12 months with or without adjusting for volatility. The specific strategy execution method6 can result
in a significant difference to the outcome of the factor portfolio. Israel et al. (2017) argue that “skillful targeting
and capturing of style premia may constitute a form of alpha on its own — one we refer to as ‘craftsmanship
alpha’ ”. The construction complexity increases with multi-factor portfolios. One fundamental decision that
portfolio managers managing multi-factor portfolios need to deal with is whether combining individual factor
portfolios is equivalent to building a bottom-up multi-factor portfolio. Bender and Wang (2016) analyse this
question and find that interaction between individual factors impacts portfolio performance significantly. They
conclude ‘both intuition and empirical evidence favour bottom-up multi-factor portfolio construction’. For this
paper, we focus on single-factor portfolios but look at alternate construction approaches for some of the factors
to highlight that portfolios claiming to track the same factor are not necessarily the same in their factor exposure.
We use the CAPM (one-factor) and the multi-factor market models in analysing factors and factor tilts.
Internationally, the excellent Ken French Data Library7 with factor returns updated monthly and going back
till 1926 for the US Markets, 1990 for Developed Markets, and 1989 for Emerging Markets serves as a reference
dataset for much of the research. Agarwalla et al. (2013), at the Indian Institute of Management, Ahmedabad
(IIMA), created and maintain a four-factor Fama-French Data Library for Indian Market8 with data from 1993
onwards. Recently, the dataset was updated9 with changes in data sources and refinements in methodology
and run till March 202110 With the growing interest in factor investing in India, the importance of having an
independent factor-return dataset for India that is freely and publicly available cannot be overstated. Such
datasets serve as a vital control mechanism to ensure the fair implementation and marketing of factor strategies
through enabling transparent and independent research.
3.1 Methodology
We look at the size, value, momentum, low-risk, quality, growth and beta factors for this study. Our universe
is the S&P BSE 200 constituents which we have collated from December 2006 through to October 202111 . The
total number of firms in the universe is 40512 . The S&P BSE 200 is rebalanced semi-annually. The constituents
are rolled over every month-end between two rebalancing periods
6
choice of the parameter(s), weights for these parameters, transformations of the underlying parameter, the portfolio
formation look-back period, the portfolio rebalancing frequency, the number of instruments in a portfolio, and weighting
scheme of portfolio amongst many others.
7
https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html
8
https://faculty.iima.ac.in/~iffm/Indian-Fama-French-Momentum/
9
We have also run our analysis using the legacy dataset which stopped Dec 2019. We do not present this analysis in
the paper. However, our findings and conclusions remain the same indicating the robustness of our approach.
10
The dataset uses CMIE’s Prowess DX https://prowessdx.cmie.com/ which provides three data releases every year
in March, September and December. The authors plan to have three releases every year.
11
The index consists of the top 200 companies by float-adjusted market-cap listed on the BSE Ltd. It is
widely used as a benchmarking tool and covers more than 85% of the market cap of all the companies listed
on the BSE Ltd. https://www.spglobal.com/spdji/en/idsenhancedfactsheet/file.pdf?calcFrequency=M&force_
download=true&hostIdentifier=48190c8c-42c4-46af-8d1a-0cd5db894797&indexId=1852729
12
inclusive of all firms that were part of the index at any point during our observation period.
Jan 2022 5
Electronic copy available at: https://ssrn.com/abstract=4000418
The choice of the top 200 stocks by free-float market capitalization minimises the inclusion of liquidity-
constrained stocks in the analysis. Without adequate liquidity in the stocks, any factor strategy will face
serious implementation issues which erode theoretical returns13 . de Groot et al. (2012a) attribute the high
transaction costs incurred in many factor strategies implemented in the US stock market to excessive trading
in small caps. Our choice of S&P BSE200 minimises this effect thereby significantly reducing the possible
trading costs. Impact costs, arising from thin liquidity is a second factor that contributes to shrinkage of re-
turns between theory and practice. Getting data on impact costs for illiquid instruments is almost impossible.
Like Schrödinger’s cat, one can either measure liquidity or the cost, but not both across a range of liquidity
simultaneously. By choice of universe, we avoid the hidden, but real, costs of the lack of liquidity which shrink
theoretical returns very quickly in the real world and estimate with some confidence the potential reduction
between the theoretical and real-world returns. As most real-world strategies will use some optimisation tech-
niques to reduce excessive turnover and impact costs, our estimates for the unoptimised strategies are likely on
the higher end of realised costs. Our estimates of shrinkage, consequently, are conservative, but reasonable.
From the S&P BSE 200 constituents, we create equally-weighted decile portfolios for each factor monthly.
Each decile has 20 stocks. We do not follow the standard academic process of creating double-sorted portfolios
on Size and Value. This is done for two pragmatic reasons: a. the current factor-tilt indices do not follow the
double-sort methodology and, b. among wealth advisers and DIY investors minimising administrative proce-
dures is preferred. On any month, from the 200 firms, adopting the double sorting approach potentially will
leave some of the double-sorted portfolios with too few firms. Our approach follows real-world implementation.
The weighting scheme of a portfolio is an important driver of performance (see Bender and Wang (2015)).
The capitalisation-weighted (CW) “market” portfolio has a central role in asset pricing Sharpe (1966) especially
in the real-world investable portfolios. Most indices, and therefore index-tracking funds/ETFs, are CW. Aca-
demic research has used equal-weighted (EW) portfolios as a norm. Plyakha et al. (2014) show “with monthly
rebalancing, an equal-weighted portfolio outperforms a value-weighted portfolio in terms of total mean return,
four-factor alpha, and Sharpe ratio”. Some index providers offer “signal-weighted-capitalisation-scaled” (SWCT)
portfolios. Here the weights are determined by the strength of the signal and then scaled appropriately by the
capitalisation of the firm. Academic research on SWCT portfolios is limited. We adopt the EW approach in our
analysis for two reasons: first, it is aligned to general academic research methodology, and second, it remains
an easily understood, intuitive method14 .
Under a market-neutral factor model, the top decile stocks would be the “longs” and the bottom decile,
the “shorts”. In our analysis, the top decile is the “winner” portfolio and the bottom decile, the “loser”. For
portfolios in general and winner portfolios in particular, we compute various performance metrics divided into
13
see Arnott et al. (2017) for a detailed overview on how factor returns shrink in the real world
14
One area of future research is to examine other weighting schemes for factor-tilt strategies.
Jan 2022 6
Electronic copy available at: https://ssrn.com/abstract=4000418
two groups. First, the core performance measures. We compute returns for each decile on a close-to-close basis
using month-end returns. To measure the risk-return trade-off we use the raw Sharpe ratio. The raw Sharpe
ratio is measured as the mean return divided by the volatility. The raw Sharpe provides a starting baseline
to compare the performance of portfolios in general. For all deciles, we compute mean returns, volatility, and
raw Sharpe ratios. Second, we look at the excess return decompositions for each decile against the one- and
four-factor (FF4)15 models. While we run the one-factor regression across the entire period under observation
(178 months from January 2007 to October 2021), due to the limitation of the IIMA Fama French Library, we
run the FF4 from January 2007 to March 2021, a total of 172 months. The FF4 analysis provides a robust
empirical framework to determine if indeed long-only portfolios have factor exposures. A priori, by using the
Database for the Indian Market as the control, each factor portfolio should demonstrate statistically significant
exposures to the underlying factors.
The persistence of factors is an important operational consideration in the rebalancing frequency of a sys-
tematic factor-based portfolio. We trace the change in the subsequent portfolio deciles of all the constituents
in the winner portfolio over time. At t= 0, all the constituents of the top decile will have a score of 10 by
construction. We look at how the constituents decile scores decay over time t = t+1, t+2, t+3, ..., t+12, t+24,
and t+36 months. As the factor strength dissipates, the portfolio score16 will drop below 10. These trajectories
of scores evidence the persistence of the factors over time17 . We us the results from the FF4 analysis and the
trajectories to ask what are reasonable hold periods for systematic long-only factor strategy portfolios.
Factors expose investors to different risks. Inherently, academic factors have a low correlation with each
other. However, our long-only factor-tilt portfolios have market exposure. Hence, to explore the diversification
benefits between them, we compute correlations of returns. Since our universe is constrained to the top 200
stocks by market capitalization, there is a chance that the same stocks appear in many factor portfolios, re-
ducing diversification benefits. To extend the returns-based analysis, we break down the winner portfolios into
their size and sector categories to measure differences in size and sectoral exposure in each strategy.
The S&P BSE 200 constituents, by definition, are skewed in size. To determine ‘Big’ and ‘Small’ size
categories, we could either split the S&P BSE 200 into the top 100 and the bottom 100 following the Secu-
rities and Exchange Board of India’s categorisation of firms as the universe for ‘Large Cap’ funds18 or follow
Edwards and Cavalli-Sforza (1965) suggestion that the best split of observations into two clusters is one which
minimizes the within-group sum of squares or maximizes the between-group sum of squares. Following the
latter, we checked for various split-points from the 40th to the 70th percentile in steps of 5 percentile points
and found that the within-group sum of squares and the between-group sum of squares were best configured
15
the standard Fama French Four Factors - Market, Size, Value, and Momentum. (Fama and French, 1993; Carhart,
1997)
16
an equal-weighted average of scores of each of the constituents of original winner portfolio.
17
Another approach is to look at the probabilities of persistence over time for each constituent of the winner portfolios.
The results will be similar to the approach we follow.
18
https://www.amfiindia.com/Themes/Theme1/downloads/1507291273374.pdf
Jan 2022 7
Electronic copy available at: https://ssrn.com/abstract=4000418
Table 1: Market Capitalisation of Firms
at 60%. A firm’s market capitalization is its average market capitalisation at the end of each month between
October of the prior year and September of the calculation year. All stocks are classified as ‘Big’ (B) and
‘Small’ (S) based on the average market capitalisation in September of every year19 . Table 1 summarises the
breakpoints, and Table 2 the median number of firms classified as ‘Big’ and ‘Small’ across our observation period.
Big Small
Size 90 111
This table shows the median number of firms across the entire observation period classified
as ‘Big’ or ‘Small’. The sum of the two median values might not add to 200.
Pct Share
Financials 20.7
Basic Materials 13.6
Consumer Cyclicals 13.1
Industrials 11.6
Healthcare 9.1
Consumer Non-Cyclicals 8.6
Technology 8.6
Energy 6.6
Utilities 6.1
Real Estate 2.0
This table shows the median share of each sector calculated by the TRBC Economic Sector of each constituent
firm within the Index at the end of each month across the entire observation period. The share is calculated
on an equal weight basis rather than the traditional market cap basis. Due to rounding, the total may not
be exactly 100%.
The Refinitiv Business Classification (TRBC) schema20 is used to categorise all firms into 10 economic
19
For the period from December 2006 through to end September 2007, we used the December 2006 market caps to
derive the breakpoint and the starting categorisation.
20
Covering over 250,000 securities in 130 countries, TRBC is a global, comprehensive, industry classification sys-
tem owned and operated by Refinitiv. See https://www.refinitiv.com/content/dam/marketing/en_us/documents/
fact-sheets/trbc-business-classification-fact-sheet.pdf
Jan 2022 8
Electronic copy available at: https://ssrn.com/abstract=4000418
sectors: Basic Materials, Consumer Cyclicals, Consumer Non-Cyclicals, Energy, Financials, Healthcare, Indus-
trials, Real Estate, Technology, Utilities. For all the portfolios formed, we compute the sectoral allocation and
compare these between the portfolios and the S&P BSE 200 index constituents. As we use equal-weighted
weights for the portfolios, the S&P BSE 200 sector comparison is also done on an equal weight basis. Table
3 shows the median share of each sector computed on the number of firms classified by their sector for the
constituents of the index every month.
Table 4 outlines the factor strategies we have chosen and their definitions. Size, value, momentum and
low volatility are referenced primarily from academic research. For high beta, growth and quality, we adapt
popular ‘strategy-index’ approaches by leading index providers. There are more sophisticated ways of defining
each factor. We deliberately keep the definitions as simple as possible for this study. To be clear, we are not
attempting to replicate any earlier research or index methodology
We have deliberately chosen alternate methodologies for some factors (value, momentum, low volatility) to
explore the differences in outcomes arising from the design choice. Additionally, for some strategies we use a
multi-parameter approach21 . In all cases, we calculate the underlying factor exposure using a consistent calcu-
lation method. For instance, we apply a 3-month lag for accounting metrics from the month first appearing on
Refinitiv consistently. So, if EPS was reported on Refinitiv at the end of March 2018, it will only be included
effective June 2018.
21
Using multiple parameters and optimising within the chosen parameter is one popular method used by managers for
factor strategies. It is prone to ‘p-hacking’ or ‘over-fitting’.
Jan 2022 9
Electronic copy available at: https://ssrn.com/abstract=4000418
We use z-scores to normalise the variables before we rank firms and create decile portfolios:
xi − µ
z= (1)
σ
where xi is the value for firm i, µ is the mean for the variable across all constituent firms for the month, and σ
is the standard deviation of the variable across all constituent firms for the month. We further transform the
z-score to deal with negative values as follows :
(1 + z) if z ≥ 0
z= (2)
1
if z < 0
1−z
All stocks in the universe for the month are ranked on z-scores and the ranks are used to create the decile
portfolios.
1/t ! 1/t !
Pm,i BSE200m
αi = 1+ −1 − 1+ −1 (3)
Pm−t,i BSE200m−t
where i is factor i, Pm,i is the NAV for the portfolio for factor i as at end of month m, Pm−t,i is the NAV for
the same portfolio t years prior (t = 3, 5) and BSE200m and BSE200m−t is the value of the S&P BSE 200 at
the end of months m and m − t respectively.
The one factor (CAPM) model uses the S&P BSE 200 as the market and is described by:
ri − rf = α + β × (RBSE200 − rf ) + ϵ (4)
where ri is the return for factor portfolio i, rf is the risk-free rate, RBSE200 is the return for the S&P BSE 200
index. All periods are monthly periods. As the winner portfolios are not market neutral, the size of market β
will inform the contribution of the market excess returns to explain the factor portfolio excess returns.
The four-factor Fama French model uses the factors from the Data Library for Indian Market and is described
by :
where ri is the return for factor portfolio i, rfIIM A is the risk-free rate in the Data Library, M KT IIM A is
the market factor from the library, SM B is the size factor (Small-minus-Big), HM L is the value factor (High-
minus-Low), and W M L is the momentum factor (Winner-minus-Loser). All periods are, again, monthly. We are
interested in the coefficients of all the factors as well as the α. A portfolio needs to show statistically significant
Jan 2022 10
Electronic copy available at: https://ssrn.com/abstract=4000418
relevant βs to demonstrate factor exposure. Hunstad and Dekhayser (2015) proposed a factor efficiency ratio
(FER). We calculate a modified FER for factor i, fi :
fi
F ERi = P (6)
k̸=i |fk |
where |fk | is the absolute exposure from the kth undesired factor. The higher the FER, the more efficient the
fund is at gaining desired factor exposure. As we expect the long-only portfolios to show exposure to MKT,
we also compute FER excluding the MKT factor. Both metrics will inform the factor exposure. As the Data
Library for Indian Market only consists of SMB, HML and WML, we compute the FER for size, value and
momentum winner portfolios.
Turnover of factor strategies is a well-researched topic. We do not seek to optimise the implementation
to reduce turnover. Consequently, the turnover rates of portfolios in our study are probably higher than any
real-world strategies. The Securities and Exchange Board of India’s definition of turnover22 is used to compute
turnover:
Pt Pt
min ( n=1 Salest , n=1 P urchasest )
T urnover = Pt
AuMt
(7)
n=1
N
Pt Pt
where n=1 Salest is the total value of sales over period n = 1, ..t and n=1 P urchasest is the total value
Pt
AuMt
of purchase the same period, and n=1
N is the average Assets under Management for the period. We
express turnover as a percentage. Each portfolio has two drivers of total sales/purchases for a period. First, the
sales/purchases related to the rebalancing required to maintain the equal weight. Second, the sales/purchases
related to the changes in constituents. We can use equation 7 to calculate the turnover related to rebalancing
and that related to new constituents.
where T urnovertotal is the turnover using the aggregate of all purchases and sales for the period, T urnoverrebalancing
is the turnover related to the purchases and sales required for rebalancing for the same period, T urnovernew is
the turnover related to the purchases and sales required from changing constituents for the same period. The
min method may create a small difference in the sum of the two sub-turnovers and the total turnover for a period
due to the different rates of return within the individual constituents leading to small differences, especially in
the rebalancing related turnover.
While we ignore costs for most part, we estimate the shrinkage arising from implementation23 . We use
turnover and size splits of portfolios to estimate shrinkage of returns due to implementation. Brokerage costs
have declined rapidly in recent years in India due to increased competition, adoption of technology-enabled
22
the lower of sales or purchase divided by the average Assets under Management for the period.
23
We ignore taxes entirely.
Jan 2022 11
Electronic copy available at: https://ssrn.com/abstract=4000418
market infrastructure (including payments, clearing, custodial, and reporting services), and rise of technology-
enabled low/no-cost brokerages amongst other reasons. We estimate these to be in the region of 10-30 basis
points per leg and use a 20 bps cost per leg of a trade. We estimate impact costs based on the size of the firm,
using the Nifty Impact cost data24 to build a simple model of impact cost. For firms classified ‘Big’, we estimate
a 5bps impact cost, and for all other firms, we estimate a 30 bps impact cost. Brokerage costs are the cost of
2 legs of the average number of constituents changed every month in the winner factor portfolio. Impact costs
are calculated as the sum of the 2 legs of impact spreads for ‘Big’ and ‘Small’ firms multiplied by the average
size splits for each winner portfolio over the period.
By being conservative in our estimate of costs (both brokerage and impact) and implementing an EW
monthly rebalanced portfolio methodology, which will likely have higher turnover than optimised implementa-
tion, our estimate for costs will be at the higher end of the possible range of implementation costs. If positive
alpha exists under these conditions, we are reasonably confident that real-world implementations with more so-
phisticated implementation choices to optimise turnover and impact costs can generate higher expected alphas
over the long term.
3.2 Data
All our data is from Refinitiv and Datastream. The S&P BSE 200 constituents are from Datastream. The
constituent dataset is available from December 2006. Constituents are updated monthly from 2011 and at 6
monthly intervals between 2006 and 2011. There is no survivorship bias in the constituent list as we cover all
firms including those that were de-listed or amalgamated/merged subsequently. We get daily closing prices of all
the firms in our universe25 . Market capitalization for all firms are as of the month-end26 . Price-to-Book (PB),
Price-to-Earnings (PE), and Dividend Yields (DY) for all firms is as of every month end. The values are lagged
by 3 months to avoid forward-looking bias. The price for the PB, PE, DY are adjusted using the firm’s closing
price for the previous month divided by the firm’s closing price from 3 months ago to reflect the adjustment in
the PB, PE, and DY values. As a result, we ensure a consistent 3 month lag in the accounting values reported
by the firm, while adjusting the price for the measurement period. Annual Revenue from Business Activity,
Return on Capital Employed (ROCE), Return on Equity (ROE), Earnings per Share (EPS), and Debt Equity
Ratio (DE) are annual from financial statement data on Refinitiv. These are also lagged by 3 months for our
calculations. For both the annual and monthly fundamental accounting variables, we recognize that different
firms are on different reporting cycles. Every month we take the most current lagged variables available. Across
24
The impact costs for the Nifty 50 stocks for a portfolio of |50 lacs is, on average, 2bps (range from 1-10bps). For
a |25 lacs portfolio size in the Next 50 is, on average, 8bps (range 2bps to 210 bps) (Data for 2021 from Nifty Indices
https://www.niftyindices.com/reports/monthly-reports
25
We have prices for all except for one stock. This stock entered the S&P BSE 200 in 2021 and would not be included
in any factor portfolio during our observation.
26
We have 4 firms for which we do not have market-cap data for some months. Using secondary research, we have
classified these as ‘Small’ firms. Unfortunately, these firms are excluded from the size factor calculations for the relevant
months. Therefore, there is very little survivorship bias in our accounting dataset.
Jan 2022 12
Electronic copy available at: https://ssrn.com/abstract=4000418
the year, every firm will have reported their results27 and therefore would be accounted for in the portfolios.
We use survivorship-bias adjusted monthly data from Fama French and Momentum Factors: Data Library
for Indian Market(Agarwalla et al., 2013) with data till March 2021 as our FF4 dataset. The risk-free rate is
computed using the 91-days T-bill rate sourced from the Reserve Bank of India’s weekly auction data available
at Refinitiv28 . The implied yields are converted to daily and monthly rates. Our method is the same as followed
by Agarwalla et al. (2013).
Figure 1 shows the development of the wealth index (Dec 2006=100) for the winner portfolios across the various
factor styes. Over the observation period, many, but not all, of the winner portfolios outperform the S&P BSE
200 (shown in grey). This chart is similar to those found in the marketing materials of factor tracker strategies.
Unfortunately, it does not tell the whole story and, as we show, actually says very little.
Figure 1: Wealth Index of Winner Long-only Factor Portfolios: Jan 2007 to Oct 2021
We compute rolling alphas for the winner portfolios for each of our factor strategies. Our preferred metric is
mean rolling alphas to minimise the ‘point-in-time’ bias. Alphas are computed using equation 3 as rolling 3- and
5-years means using monthly returns. Table 5 summarises the alphas. Seemingly, other than size, one flavour
of value, the high beta factor and growth, all the other winner portfolios generate significant positive mean 3-
and 5-year alphas relative to the S&P BSE200 Index.
27
There are cases where firms have changed their reporting cycle. This is a matter of routine, and we, therefore, treat
the accounting variables as they get reported. While Refinitiv provides standardised data, in the case of other sources,
the data may need to be first normalised.
28
The data is also available at http://dbie.rbi.org.in/DBIE/dbie.rbi?site=statistics, under Financial Market»
Government Securities Market.
Jan 2022 13
Electronic copy available at: https://ssrn.com/abstract=4000418
Table 5: Mean Annualised 3 and 5 year rolling Alpha Winner Long-only Factor Portfolios: Jan
2007 to Oct 2021
3 yr 5 yr
SIZE -3.7 -3.7
VAL_ac -9.6 -10.4
VAL_mp 0.4 1.6
MOM_ac 6.5 7.6
MOM_vol 9.4 10.3
LOVOL 7.7 7.4
BAB 8.5 8.1
BETA -13.9 -13.7
GROW_mp -0.8 -0.0
QUAL 7.1 7.6
This table shows the annualised rolling alpha computed for each winner portfolio over the returns
of the S&P BSE 200 Index.
Figure 2 shows the range of the rolling returns. On average 3- years alphas have a wider 5-year alphas:
the longer the period of following any systematic, the realised returns are converge closer to average returns.
Amongst all the factors, low volatility winner portfolios have the narrowest range of alpha over the 3- and 5-
year periods. The range over 5-years for momentum is also relatively narrow compared to other factors. In
general, the range of alpha is wide. Depending upon the entry timing, an investor could experience a negative
or positive 3-year alpha. Across all strategies, for our observation period, an unfortunate investor entering at
the worst time would experience negative alphas over 3-years. Over 5-years, only MOM_vol and BAB winner
portfolios would have outperformed the broader market for the same hapless investor. However, the maximum
3- and 5-year alphas are positive across all strategies except for 5-year BETA. Empirically, entry timing does
play a role in realised alpha across strategies even over “long” periods29 . Factor strategies can under perform,
even over “long” periods.
Figure 2: Rolling 3 and 5 years Annualised Alphas Ranges for Winner Long-only Factor Portfolios: Jan 2007 to Oct 2021
Source: Authors calculations, Refinitiv. All alphas are excluding any costs.
Rolling alphas offer a better picture of the potential attractiveness of factor style portfolios. Different style
29
In India, 3 years is often seen as an eternity amongst many investors and advisers.
Jan 2022 14
Electronic copy available at: https://ssrn.com/abstract=4000418
have different inherent volatility. Shorter holding periods are more volatile than longer holding periods. Differ-
ent styles have different risk and we turn next to risk-adjusted returns.
Table 6: Gross Annualised Returns and Risk Across Deciles for different Long-only Factor Portfolios: Jan 2007
to Oct 2021
D_1 D_2 D_3 D_4 D_5 D_6 D_7 D_8 D_9 D_10
Mean 13.63 9.67 13.62 17.99 20.40 16.49 15.24 14.88 20.34 18.06
SIZE std 22.34 25.84 25.74 27.91 27.66 27.25 29.97 33.13 34.95 41.89
Raw Sharpe 0.61 0.37 0.53 0.64 0.74 0.61 0.51 0.45 0.58 0.43
Mean 16.47 18.72 16.92 17.03 15.61 13.26 17.34 15.68 16.19 13.62
VAL_ac std 22.65 23.42 23.50 26.53 28.12 30.36 29.71 36.10 37.52 42.74
Raw Sharpe 0.73 0.80 0.72 0.64 0.56 0.44 0.58 0.43 0.43 0.32
Mean 12.38 9.02 16.59 16.68 19.97 16.05 16.11 20.17 20.10 13.06
VAL_mp std 35.15 30.49 28.57 29.34 27.05 29.39 29.20 30.34 28.26 29.18
Raw Sharpe 0.35 0.30 0.58 0.57 0.74 0.55 0.55 0.66 0.71 0.45
Mean 13.71 13.67 15.56 11.10 14.95 14.49 18.43 19.16 16.76 22.35
MOM_ac std 45.08 37.34 34.84 31.93 27.53 26.94 23.59 24.12 24.48 26.99
Raw Sharpe 0.30 0.37 0.45 0.35 0.54 0.54 0.78 0.79 0.68 0.83
Mean 10.77 10.40 14.44 11.58 14.72 17.25 16.19 18.84 21.18 24.99
MOM_vol std 38.25 35.37 33.10 32.82 31.30 29.43 26.05 24.94 24.80 25.71
Raw Sharpe 0.28 0.29 0.44 0.35 0.47 0.59 0.62 0.76 0.85 0.97
Mean 7.83 18.12 19.79 18.21 14.90 14.60 16.78 13.68 17.05 19.28
LOVOL std 47.14 39.21 36.34 34.22 28.91 26.96 24.39 23.53 21.48 15.26
Raw Sharpe 0.17 0.46 0.54 0.53 0.52 0.54 0.69 0.58 0.79 1.26
Mean 14.44 9.82 13.67 19.43 18.35 17.28 14.65 15.82 18.85 18.88
BAB std 48.40 37.74 34.30 31.91 30.00 28.84 24.82 23.90 21.66 16.02
Raw Sharpe 0.30 0.26 0.40 0.61 0.61 0.60 0.59 0.66 0.87 1.18
Mean 20.49 18.90 17.32 17.26 17.48 16.62 13.56 11.51 15.35 11.58
BETA std 15.73 22.59 24.81 26.83 27.70 28.94 29.74 35.24 39.20 49.66
Raw Sharpe 1.30 0.84 0.70 0.64 0.63 0.57 0.46 0.33 0.39 0.23
Mean 15.25 16.09 14.28 12.17 10.74 15.17 21.58 18.80 18.38 17.70
GROW_mp std 40.55 34.56 33.48 27.62 27.89 24.67 25.50 25.73 26.94 31.69
Raw Sharpe 0.38 0.47 0.43 0.44 0.38 0.62 0.85 0.73 0.68 0.56
Mean 10.96 16.70 16.44 9.88 16.32 16.25 18.96 17.35 20.99 17.33
QUAL std 36.76 33.11 32.73 31.57 28.01 26.44 26.08 23.90 25.53 24.14
Raw Sharpe 0.30 0.50 0.50 0.31 0.58 0.61 0.73 0.73 0.82 0.72
This table shows the annualised mean return for each decile (D_1 being the lowest or worst and D_10 being the
highest or best ranked firms) across each of the factors in Table 4, the annualised standard deviation of the returns
and raw Sharpe Ratio.
Ideally, if factors affect performance, returns should increase monotonically across the deciles. Table 6 shows
annualised gross returns, annualised standard deviation and raw Sharpe Ratio for decile portfolios across all
styles. The S&P BSE 200 had an annualised mean return of 13.89%, an annualised standard deviation of 23.25
and a raw Sharpe of 0.6. Momentum and low-volatility show moderate to strong monotonic increases in raw
Sharpe across deciles. High Beta shows a generally reducing raw sharpe from the loser decile to the winner
decile. Value’s raw Sharpe does not have a clear trend and varies between the two variants30 . Size, reflective
of the relative growth of returns in large-cap firms during the current decade, shows a downward trend in the
raw Sharpe ratio across the deciles. Growth and quality show increasing raw Sharpe ratios across deciles, but
the path is not linear. The lowest decile portfolio for all factors has a positive return during our observation
period. With the caveats of the universe, the S&P BSE 200, the construction method, and the observation
period empirically on average, shorting a naive loser portfolio is a losing proposition.
30
VAL_ac shows a weakening raw Sharpe to a lesser degree. VAL_mp has a more complex raw Sharpe trajectory.
Jan 2022 15
Electronic copy available at: https://ssrn.com/abstract=4000418
Across factors, risk generally improves over deciles. The higher deciles for multi-parameter value, momen-
tum, low-volatility, growth, and quality show lower standard deviations in the highest decile compared with
the lower deciles. Size, as expected, shows higher risk in D_10 (small cap) compared with D_1 (large cap).
As the decile portfolios have only 20 stocks, one would expect the standard deviation of returns of top decile
portfolios to be higher than the much broader S&P BSE 200 Index. Interestingly, the standard deviation of
returns for the low volatility winner portfolios is significantly lower than the S&P BSE 200 index. For a 20-stock
portfolio to have significantly lower volatility while delivering higher expected returns should be of interest to
asset allocators and researchers. For the winner portfolios of momentum, the standard deviations of returns are
higher than that of the S&P BSE 200 Index, but not by very much.
Across the style winner portfolios, size, value and, more marginally, growth under-perform the index in raw
Sharpe terms. Winner portfolios for other styles have superior risk adjusted returns during our observation
period.
A small digression on how we present decile portfolio for the styles. We have adopted the convention that
winner portfolios across styles are portfolios of firms showing the highest z-scores for the characteristic. Thus,
the winner portfolios for size would be the smallest firms; the ‘cheapest’ firms for value; the highest momentum
firms for momentum; the lowest volatility for low volatility; the highest beta for beta; the firms showing the
highest growth for growth and the firms with the highest z-scores across our quality characteristics for quality.
Some researchers present findings such that low-volatility and high beta styles are ordered in the same manner
- where the highest decile shows the highest volatility and the highest beta. For readers more used to the other
method, the ‘inverted’ Beta factor is another version of low volatility31 .
Taking a more formal analytical approach, we regress the excess returns of each of the decile portfolios against
the S&P BSE 200 excess returns. Table 7 summarises this analysis. First, we look at α. The higher decile
portfolios for momentum and low volatility have a statistically significant positive intercept (α). The annualised
α for momentum is between 7% and 10% and approx 7% for low-volatility. Between the two variations in mo-
mentum, MOM_vol shows a stronger statistical α than MOM_ac with the top 3 deciles of MOM_vol showing
statistically significant and positive α. The top two decile portfolios of both variations of low volatility32 also
have statistically significant and positive α. For increasing deciles for quality and growth, α generally increases.
However, the statistical significance is less clear. The value portfolios do not exhibit any statistically significant
trend. Loser portfolios across all styles do have statistical evidence of negative α. Once again, going short loser
portfolios is not statistically likely to generate positive expected returns for our universe and the observation
31
Such readers will need to read the BETA row from right to left instead of left-to-right for the other factors for the
low-volatility interpretation.
32
as well as the lowest two decile portfolios of BETA.
Jan 2022 16
Electronic copy available at: https://ssrn.com/abstract=4000418
Table 7: One Factor Regression Summary for Long-only Factor Portfolios: Jan 2007 to Oct 2021
D_1 D_2 D_3 D_4 D_5 D_6 D_7 D_8 D_9 D_10
Intercept 0.01 -0.36** -0.05 0.23 0.41* 0.14 -0.00 -0.10 0.27 -0.02
(0.09) (0.14) (0.17) (0.22) (0.23) (0.23) (0.26) (0.28) (0.34) (0.44)
MKTRF 0.94*** 1.08*** 1.06*** 1.12*** 1.10*** 1.08*** 1.18*** 1.31*** 1.34*** 1.58***
SIZE
(0.01) (0.02) (0.03) (0.03) (0.03) (0.03) (0.04) (0.04) (0.05) (0.07)
R-squared 0.97 0.94 0.91 0.87 0.85 0.85 0.84 0.85 0.80 0.77
Adj. R-squared 0.97 0.94 0.91 0.87 0.85 0.85 0.84 0.85 0.80 0.77
Intercept 0.27 0.39** 0.25 0.18 0.05 -0.17 0.15 -0.10 -0.07 -0.35
(0.23) (0.19) (0.19) (0.19) (0.22) (0.23) (0.24) (0.33) (0.38) (0.47)
MKTRF 0.86*** 0.93*** 0.94*** 1.08*** 1.13*** 1.22*** 1.18*** 1.41*** 1.42*** 1.58***
VAL_ac
(0.03) (0.03) (0.03) (0.03) (0.03) (0.03) (0.04) (0.05) (0.06) (0.07)
R-squared 0.78 0.86 0.87 0.89 0.87 0.88 0.86 0.83 0.78 0.74
Adj. R-squared 0.78 0.86 0.87 0.89 0.87 0.88 0.86 0.83 0.78 0.74
Intercept -0.31 -0.48* 0.12 0.10 0.39* 0.06 0.07 0.33 0.37* -0.14
(0.35) (0.26) (0.23) (0.23) (0.22) (0.24) (0.23) (0.24) (0.22) (0.27)
MKTRF 1.34*** 1.20*** 1.14*** 1.18*** 1.08*** 1.17*** 1.17*** 1.22*** 1.14*** 1.13***
VAL_mp
(0.05) (0.04) (0.03) (0.03) (0.03) (0.04) (0.03) (0.04) (0.03) (0.04)
R-squared 0.79 0.84 0.86 0.87 0.86 0.86 0.87 0.87 0.87 0.82
Adj. R-squared 0.79 0.84 0.86 0.87 0.86 0.86 0.87 0.87 0.87 0.81
Intercept -0.38 -0.26 -0.08 -0.35 0.02 -0.02 0.35** 0.41** 0.24 0.61**
(0.52) (0.37) (0.32) (0.29) (0.22) (0.18) (0.17) (0.20) (0.23) (0.30)
MKTRF 1.65*** 1.43*** 1.36*** 1.25*** 1.11*** 1.11*** 0.96*** 0.96*** 0.95*** 0.99***
MOM_ac
(0.08) (0.05) (0.05) (0.04) (0.03) (0.03) (0.03) (0.03) (0.03) (0.04)
R-squared 0.72 0.79 0.83 0.83 0.87 0.91 0.89 0.85 0.82 0.74
Adj. R-squared 0.72 0.79 0.83 0.83 0.87 0.91 0.89 0.85 0.82 0.73
Intercept -0.48 -0.46 -0.12 -0.35 -0.08 0.15 0.14 0.35** 0.53** 0.83***
(0.41) (0.35) (0.31) (0.25) (0.26) (0.25) (0.20) (0.17) (0.21) (0.31)
MKTRF 1.43*** 1.35*** 1.28*** 1.32*** 1.24*** 1.17*** 1.05*** 1.02*** 0.98*** 0.93***
MOM_vol
(0.06) (0.05) (0.05) (0.04) (0.04) (0.04) (0.03) (0.03) (0.03) (0.05)
R-squared 0.76 0.79 0.81 0.88 0.86 0.85 0.88 0.90 0.85 0.70
Adj. R-squared 0.76 0.79 0.81 0.88 0.86 0.85 0.88 0.90 0.85 0.70
Intercept -0.92** 0.00 0.16 0.12 -0.02 -0.00 0.22 0.02 0.31* 0.64***
(0.46) (0.35) (0.27) (0.30) (0.22) (0.20) (0.18) (0.18) (0.17) (0.17)
MKTRF 1.81*** 1.54*** 1.47*** 1.34*** 1.16*** 1.09*** 0.99*** 0.95*** 0.86*** 0.57***
LOVOL
(0.07) (0.05) (0.04) (0.05) (0.03) (0.03) (0.03) (0.03) (0.03) (0.03)
R-squared 0.80 0.84 0.89 0.83 0.87 0.89 0.88 0.87 0.87 0.74
Adj. R-squared 0.80 0.84 0.89 0.83 0.87 0.89 0.88 0.87 0.87 0.74
Intercept -0.43 -0.59* -0.21 0.24 0.21 0.16 0.06 0.18 0.46** 0.61***
(0.50) (0.32) (0.30) (0.25) (0.25) (0.24) (0.21) (0.21) (0.23) (0.20)
MKTRF 1.82*** 1.50*** 1.35*** 1.28*** 1.20*** 1.15*** 0.99*** 0.94*** 0.82*** 0.56***
BAB
(0.07) (0.05) (0.04) (0.04) (0.04) (0.04) (0.03) (0.03) (0.03) (0.03)
R-squared 0.77 0.85 0.84 0.87 0.86 0.86 0.86 0.83 0.77 0.66
Adj. R-squared 0.77 0.85 0.84 0.87 0.86 0.86 0.86 0.83 0.77 0.66
Intercept 0.74*** 0.44** 0.26 0.21 0.20 0.11 -0.14 -0.40 -0.19 -0.70
(0.21) (0.22) (0.21) (0.22) (0.23) (0.22) (0.22) (0.30) (0.35) (0.46)
MKTRF 0.54*** 0.87*** 0.98*** 1.07*** 1.10*** 1.16*** 1.20*** 1.40*** 1.53*** 1.93***
BETA
(0.03) (0.03) (0.03) (0.03) (0.03) (0.03) (0.03) (0.04) (0.05) (0.07)
R-squared 0.63 0.81 0.85 0.86 0.86 0.87 0.88 0.85 0.83 0.82
Adj. R-squared 0.63 0.80 0.85 0.86 0.86 0.87 0.88 0.85 0.83 0.82
Intercept -0.20 -0.03 -0.14 -0.19 -0.30 0.09 0.53*** 0.34* 0.29 0.15
(0.42) (0.33) (0.31) (0.21) (0.22) (0.17) (0.18) (0.20) (0.23) (0.29)
MKTRF 1.53*** 1.34*** 1.30*** 1.11*** 1.12*** 1.00*** 1.04*** 1.03*** 1.07*** 1.23***
GROW_mp
(0.06) (0.05) (0.05) (0.03) (0.03) (0.03) (0.03) (0.03) (0.03) (0.04)
R-squared 0.78 0.81 0.82 0.88 0.87 0.89 0.89 0.87 0.85 0.82
Adj. R-squared 0.78 0.81 0.81 0.88 0.87 0.89 0.89 0.87 0.85 0.82
Intercept -0.47 0.04 0.01 -0.45* 0.11 0.14 0.35 0.28 0.50** 0.28
(0.32) (0.29) (0.27) (0.24) (0.23) (0.21) (0.22) (0.20) (0.20) (0.22)
MKTRF 1.44*** 1.30*** 1.30*** 1.27*** 1.12*** 1.06*** 1.04*** 0.95*** 1.03*** 0.95***
QUAL
(0.05) (0.04) (0.04) (0.04) (0.03) (0.03) (0.03) (0.03) (0.03) (0.03)
R-squared 0.84 0.83 0.86 0.88 0.86 0.86 0.85 0.86 0.88 0.83
Adj. R-squared 0.84 0.83 0.86 0.87 0.86 0.86 0.85 0.86 0.88 0.83
This table shows the summary of regression of the decile portfolio monthly excess returns against the monthly excess returns of the S&P BSE
200 Index. We show the Intercept and the MKTRF (excess returns of the Index). Standard errors in parentheses under each coefficient.
* p<.1, ** p<.05, ***p<.01
period.
Second, except for size and VAL_ac, in general, higher decile portfolios for momentum, low volatility,
growth, and quality have lower betas than the lower decile portfolios. As noted above, low volatility strategies
have significantly lower betas than the market - making these potentially attractive portfolios for certain types
Jan 2022 17
Electronic copy available at: https://ssrn.com/abstract=4000418
of investment strategies. The higher decile portfolios for momentum have betas below 1, but not by much. The
increase in beta for size is aligned to evidence from other markets - small-cap stocks have a higher beta. In the
Indian context, for the period under observation and the universe selected for the study, this additional risk
taken due to firm size is not compensated by higher returns. In the case of value, we have a mixed picture -
one method having an increase in beta as value deciles increase, and another demonstrating reduced beta. In
both variations, beta is more than 1.
Finally, we look at r-square - how much of the excess returns of the decile portfolios are explained by the
market. A priori, we would expect that both high deciles and low deciles to have lower r-squares than the
central deciles to show the strength of the factor performance. Generally, except size, all other factors show
evidence of lower r-squares at the extremes, and the r-squares of the top decile portfolios are at the lower end
of the range of r-squares across deciles. R-squares range between 0.66 and 0.83 for the top decile portfolios and
0.63 and 0.97 for the bottom decile portfolios33 . The low volatility factor has the lowest r-squares (0.66 for
BAB and 0.74 for LOVOL) followed by momentum (0.70 for MOM_vol and 0.74 for MOM_ac). For size, the
top-heavy nature of the S&P BSE 200, and hence the cap-weighted market indices, is seen in the high r-square
for the loser portfolio for the category (0.97).
The one-factor return decomposition shows statistically significant and positive α for low volatility and
momentum winner portfolios. For others, while there is no robust statistical α, there are encouraging signs
such as low r-squares. Low volatility shows significantly lower beta than the S&P BSE 200 Index. Momentum,
growth, and quality show marginally lower beta than the index. Surprisingly value portfolios, as defined, show
significantly higher beta than the index.
4.4 Long-only Systematic Decile Portfolios: Fama French Four Factor Analysis
The Fama-French four-factor return decomposition shows the factor tilts of each of our long-only style portfo-
lios. We use the Database for the Indian Market to decompose returns into four factors. We use the monthly
survivorship-bias adjusted dataset of MKT (market), SMB (small-minus-big or size), HML (high-minus-low
or value) and WML (winner-minus-loser or momentum) as the exogenous variables for our regression. The
Database for the Indian Market uses a much larger universe than the S&P BSE 200. Table 8 summarises the
results. In general, the FF4 decomposition explains more of the excess returns as seen by the larger r-squares
than the one-factor decomposition (from mean r-square across all portfolios of 0.84 for one factor to 0.88 for
FF4). The mean r-squares for the winner portfolios of each strategy are also higher (from 0.76 for one factor to
0.82 for FF4). The highest decile low volatility portfolios continue to show significantly lower r-squares (0.70
for BAB and 0.75 for LOVOL) compared to the r-squares of other winner portfolios.
If the FF4 decomposition shows a statistically significant exposure to the underlying factor, we would also
33
The lowest BETA decile shows the lowest r-square. This is equivalent to the top deciles of the low volatility portfolios.
Jan 2022 18
Electronic copy available at: https://ssrn.com/abstract=4000418
expect α in the FF4 analysis to be zero. Under this circumstance, the winner portfolios returns are well explained
by the underlying factor. If there is a residual α, the FF4 does not fully explain the result and we would need
to look for other exogenous variables. From our factor strategies, we would expect size, value and momentum
to have no residual α while low volatility, beta, growth, and quality to have lower r-square and probably some α.
Mean MKT β across all decile portfolios reduces from 1.17 for one factor to 1.01 for FF4. The additional
factors provide the increased explanatory power. A priori, each of our factor strategies should show statistically
robust exposure to the relevant factor. As the portfolios are long-only, the specific factor βs will likely be less
than 1. Our criteria to check for factor exposure for size, value and momentum strategies:
• The relevant factor exposure to be statistically significant for the winner and loser portfolios: a positive
exposure for the winner portfolios and a negative exposure for the loser portfolios. At a minimum, the
winner portfolios should have statistically significant factor β even if the loser portfolios do not show any
exposure.
• There should be a progression from significant and negative underlying factor exposure at the loser
portfolio to a significant and positive exposure for the winner portfolio. Towards the middle deciles, the
factor exposure should dissipate. This trend of factor exposure is used in the analysis of persistence of
the factor exposure.
• The relevant factor β is more than 0.20. This is an arbitrary cut-off point. We choose it as it implies that
below this, less than a fifth of the change in 1% of returns is explained by the factor.
• α=0
The size portfolios show clear exposures to SMB under the first two criteria. The highest decile (the smallest
market-cap firms in the S&P BSE 200 Index) has a positive and significant coefficient (0.41***) and the lowest
decile (the largest caps in the 200 Index) show a negative and significant coefficient (-0.13***) to the SMB
factor. The SMB exposure switches from positive to negative/zero between deciles 4 and 5. By decile 5, the
exposure to SMB is less than 20% - and could serve as a point where we consider factor effect has effectively
dissipated. The two extreme deciles also have opposite exposures to HML - the smallest caps have a positive and
significant coefficient (0.22***). During our observation period and our universe, the small cap decile portfolio
is more value biased. The higher decile portfolios all show negative and significant exposure to momentum
(WML). As our size portfolios are split only on market-cap, this result implies that, for the observation period
and our universe, the firms with the smallest market-caps in the S&P BSE 200 Index would be detractors in
a momentum portfolio. For our third test, the winner portfolio has a significant and positive α. This result is
contrary to the one-factor and the 3- and 5-year analysis.
Jan 2022 19
Electronic copy available at: https://ssrn.com/abstract=4000418
Table 8: Four Factor Regression Summary for Long-only Factor Portfolios: Jan 2007 to Mar 2021
D_1 D_2 D_3 D_4 D_5 D_6 D_7 D_8 D_9 D_10
Intercept -0.02 -0.33** 0.06 0.42* 0.57** 0.22 0.30 0.46* 0.88*** 0.75**
(0.13) (0.16) (0.18) (0.24) (0.22) (0.23) (0.26) (0.24) (0.27) (0.33)
MKTRF 0.93*** 1.04*** 0.95*** 1.00*** 1.00*** 0.90*** 0.98*** 1.10*** 1.06*** 1.20***
(0.02) (0.03) (0.03) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.05)
SMB -0.13*** -0.07* 0.05 -0.04 0.15*** 0.27*** 0.30*** 0.32*** 0.34*** 0.41***
(0.03) (0.04) (0.04) (0.06) (0.05) (0.05) (0.06) (0.06) (0.06) (0.08)
SIZE HML -0.01 -0.02 0.06 0.12** 0.11** 0.18*** 0.09* 0.02 0.08 0.22***
(0.03) (0.03) (0.04) (0.05) (0.05) (0.05) (0.05) (0.05) (0.06) (0.07)
WML -0.05** -0.10*** -0.16*** -0.19*** -0.11*** -0.12*** -0.22*** -0.31*** -0.44*** -0.55***
(0.02) (0.03) (0.03) (0.04) (0.04) (0.04) (0.05) (0.04) (0.05) (0.06)
R-squared 0.94 0.93 0.91 0.87 0.89 0.88 0.87 0.91 0.90 0.89
Adj. R-squared 0.94 0.93 0.91 0.87 0.88 0.88 0.87 0.91 0.89 0.89
Intercept 0.21 0.48** 0.28 0.38** 0.45** 0.32 0.64*** 0.33 0.40 -0.05
(0.22) (0.20) (0.20) (0.18) (0.21) (0.21) (0.21) (0.28) (0.35) (0.38)
MKTRF 0.91*** 0.89*** 0.89*** 1.01*** 1.00*** 1.06*** 0.99*** 1.13*** 1.08*** 1.19***
(0.04) (0.03) (0.03) (0.03) (0.03) (0.03) (0.03) (0.05) (0.06) (0.06)
SMB 0.15*** 0.23*** 0.20*** 0.16*** 0.15*** 0.12** 0.11** 0.23*** 0.16* 0.12
(0.05) (0.05) (0.05) (0.04) (0.05) (0.05) (0.05) (0.07) (0.08) (0.09)
VAL_ac HML -0.16*** -0.12*** -0.02 -0.05 -0.07 -0.04 0.06 0.24*** 0.34*** 0.62***
(0.05) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.06) (0.07) (0.08)
WML 0.08* -0.02 -0.02 -0.12*** -0.26*** -0.33*** -0.35*** -0.36*** -0.44*** -0.43***
(0.04) (0.04) (0.04) (0.03) (0.04) (0.04) (0.04) (0.05) (0.06) (0.07)
R-squared 0.83 0.87 0.88 0.92 0.90 0.91 0.92 0.89 0.85 0.86
Adj. R-squared 0.83 0.87 0.88 0.92 0.90 0.91 0.91 0.89 0.85 0.85
Intercept -0.03 -0.51* 0.25 0.49** 0.59** 0.44* 0.43* 0.69*** 0.71*** 0.24
(0.30) (0.27) (0.23) (0.23) (0.24) (0.23) (0.23) (0.22) (0.22) (0.26)
MKTRF 1.05*** 1.05*** 0.97*** 0.99*** 0.94*** 1.03*** 1.00*** 1.07*** 1.02*** 1.01***
(0.05) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04)
SMB 0.21*** 0.02 0.11** 0.19*** 0.15*** 0.19*** 0.25*** 0.21*** 0.11** 0.19***
(0.07) (0.06) (0.05) (0.05) (0.06) (0.05) (0.05) (0.05) (0.05) (0.06)
VAL_mp HML 0.30*** 0.24*** 0.20*** 0.03 0.11** 0.00 0.05 0.00 -0.07 -0.04
(0.06) (0.06) (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) (0.05)
WML -0.35*** -0.15*** -0.17*** -0.28*** -0.15*** -0.24*** -0.20*** -0.23*** -0.24*** -0.22***
(0.05) (0.05) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.05)
R-squared 0.87 0.86 0.89 0.90 0.87 0.90 0.90 0.91 0.90 0.86
Adj. R-squared 0.87 0.85 0.88 0.89 0.86 0.89 0.89 0.90 0.89 0.86
Intercept 0.69* 0.71*** 0.67*** 0.30 0.32 0.14 0.40** 0.23 -0.17 0.01
(0.36) (0.27) (0.26) (0.26) (0.21) (0.20) (0.19) (0.22) (0.22) (0.28)
MKTRF 1.16*** 1.08*** 1.10*** 1.00*** 0.97*** 1.00*** 0.88*** 0.90*** 0.96*** 1.09***
(0.06) (0.04) (0.04) (0.04) (0.03) (0.03) (0.03) (0.04) (0.04) (0.05)
SMB 0.22** 0.07 0.11* 0.23*** 0.12** 0.14*** 0.20*** 0.18*** 0.20*** 0.15**
(0.08) (0.06) (0.06) (0.06) (0.05) (0.05) (0.04) (0.05) (0.05) (0.07)
MOM_ac HML 0.29*** 0.04 0.05 0.04 -0.01 0.04 0.05 0.09** 0.15*** 0.09
(0.08) (0.06) (0.05) (0.05) (0.04) (0.04) (0.04) (0.05) (0.05) (0.06)
WML -0.82*** -0.71*** -0.51*** -0.41*** -0.24*** -0.12*** -0.03 0.06 0.21*** 0.35***
(0.06) (0.05) (0.05) (0.05) (0.04) (0.04) (0.03) (0.04) (0.04) (0.05)
R-squared 0.89 0.91 0.91 0.89 0.90 0.91 0.89 0.86 0.86 0.81
Adj. R-squared 0.89 0.91 0.90 0.88 0.90 0.90 0.89 0.86 0.86 0.80
Intercept 0.44 0.37 0.45 0.07 0.43* 0.44* 0.24 0.27 0.40* 0.18
(0.27) (0.26) (0.28) (0.23) (0.24) (0.23) (0.22) (0.20) (0.21) (0.26)
MKTRF 1.05*** 1.02*** 1.01*** 1.13*** 1.02*** 1.00*** 0.93*** 0.95*** 0.97*** 1.06***
(0.04) (0.04) (0.05) (0.04) (0.04) (0.04) (0.04) (0.03) (0.03) (0.04)
SMB 0.14** 0.13** 0.17** 0.10* 0.21*** 0.23*** 0.17*** 0.14*** 0.23*** 0.10
(0.07) (0.06) (0.07) (0.06) (0.06) (0.05) (0.05) (0.05) (0.05) (0.06)
MOM_vol HML 0.19*** 0.06 0.16*** 0.07 0.00 0.05 0.08* 0.09** 0.03 0.08
(0.06) (0.05) (0.06) (0.05) (0.05) (0.05) (0.05) (0.04) (0.04) (0.06)
WML -0.69*** -0.63*** -0.41*** -0.34*** -0.37*** -0.22*** -0.11*** 0.02 0.11*** 0.39***
(0.05) (0.05) (0.05) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.05)
R-squared 0.91 0.91 0.88 0.91 0.90 0.89 0.88 0.88 0.88 0.81
Adj. R-squared 0.91 0.91 0.87 0.91 0.90 0.89 0.88 0.88 0.87 0.81
Intercept -0.20 0.38 0.31 0.53* 0.32 0.35* 0.43** 0.13 0.50*** 0.56***
(0.38) (0.31) (0.24) (0.30) (0.21) (0.20) (0.18) (0.18) (0.17) (0.18)
MKTRF 1.39*** 1.28*** 1.27*** 1.08*** 0.97*** 0.98*** 0.86*** 0.90*** 0.85*** 0.56***
(0.06) (0.05) (0.04) (0.05) (0.03) (0.03) (0.03) (0.03) (0.03) (0.03)
SMB 0.17* 0.20*** 0.16*** 0.26*** 0.17*** 0.15*** 0.19*** 0.18*** 0.05 0.07*
(0.09) (0.07) (0.06) (0.07) (0.05) (0.05) (0.04) (0.04) (0.04) (0.04)
LOVOL HML 0.31*** 0.24*** 0.21*** 0.16** 0.10** -0.06 0.04 -0.06 -0.16*** 0.03
(0.08) (0.07) (0.05) (0.06) (0.05) (0.04) (0.04) (0.04) (0.04) (0.04)
WML -0.63*** -0.33*** -0.23*** -0.32*** -0.26*** -0.21*** -0.16*** -0.06* -0.09*** 0.05
(0.07) (0.06) (0.04) (0.05) (0.04) (0.04) (0.03) (0.03) (0.03) (0.03)
R-squared 0.88 0.89 0.92 0.87 0.90 0.90 0.90 0.90 0.89 0.75
Adj. R-squared 0.88 0.88 0.92 0.86 0.90 0.90 0.90 0.90 0.89 0.75
Continued on next page
Jan 2022 20
Electronic copy available at: https://ssrn.com/abstract=4000418
Table 8: Four Factor Regression Summary for Long-only Factor Portfolios: Jan 2007 to Mar 2021
D_1 D_2 D_3 D_4 D_5 D_6 D_7 D_8 D_9 D_10
Intercept 0.25 -0.16 -0.02 0.51** 0.39 0.48** 0.29 0.47** 0.68*** 0.54***
(0.44) (0.29) (0.29) (0.24) (0.26) (0.23) (0.19) (0.20) (0.22) (0.21)
MKTRF 1.39*** 1.23*** 1.11*** 1.07*** 1.02*** 1.05*** 0.88*** 0.86*** 0.77*** 0.57***
(0.07) (0.05) (0.05) (0.04) (0.04) (0.04) (0.03) (0.03) (0.04) (0.03)
SMB 0.02 0.09 0.11 0.23*** 0.18*** 0.19*** 0.26*** 0.19*** 0.16*** 0.14***
(0.10) (0.07) (0.07) (0.06) (0.06) (0.05) (0.04) (0.05) (0.05) (0.05)
BAB HML 0.46*** 0.21*** 0.29*** 0.16*** 0.18*** -0.08* -0.05 -0.12*** -0.08* -0.03
(0.09) (0.06) (0.06) (0.05) (0.05) (0.05) (0.04) (0.04) (0.05) (0.04)
WML -0.64*** -0.40*** -0.25*** -0.23*** -0.16*** -0.20*** -0.15*** -0.16*** -0.10*** 0.07*
(0.08) (0.05) (0.05) (0.04) (0.05) (0.04) (0.03) (0.04) (0.04) (0.04)
R-squared 0.86 0.90 0.87 0.90 0.87 0.89 0.90 0.87 0.82 0.70
Adj. R-squared 0.85 0.89 0.87 0.90 0.87 0.89 0.90 0.87 0.82 0.70
Intercept 0.71*** 0.56*** 0.46** 0.57*** 0.42* 0.27 0.11 -0.09 0.38 -0.10
(0.21) (0.20) (0.20) (0.20) (0.23) (0.23) (0.21) (0.28) (0.32) (0.43)
MKTRF 0.48*** 0.87*** 0.89*** 0.96*** 0.98*** 1.04*** 1.00*** 1.15*** 1.23*** 1.54***
(0.03) (0.03) (0.03) (0.03) (0.04) (0.04) (0.03) (0.05) (0.05) (0.07)
SMB 0.21*** 0.13*** 0.25*** 0.19*** 0.19*** 0.12** 0.14*** 0.21*** 0.13* 0.04
(0.05) (0.05) (0.05) (0.05) (0.05) (0.06) (0.05) (0.07) (0.08) (0.10)
BETA HML 0.09** -0.13*** -0.06 -0.08* 0.01 0.09* 0.19*** 0.20*** 0.17** 0.35***
(0.04) (0.04) (0.04) (0.04) (0.05) (0.05) (0.04) (0.06) (0.07) (0.09)
WML 0.04 -0.05 -0.12*** -0.21*** -0.16*** -0.14*** -0.24*** -0.30*** -0.48*** -0.57***
(0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.05) (0.06) (0.08)
R-squared 0.68 0.86 0.88 0.91 0.88 0.88 0.91 0.89 0.88 0.87
Adj. R-squared 0.68 0.85 0.88 0.90 0.88 0.88 0.91 0.88 0.88 0.86
Intercept 0.49* 0.86*** 0.60** 0.21 0.06 0.09 0.59*** 0.33 0.12 0.04
(0.29) (0.25) (0.27) (0.20) (0.23) (0.19) (0.20) (0.21) (0.23) (0.29)
MKTRF 1.14*** 1.03*** 1.06*** 0.96*** 0.97*** 0.93*** 0.94*** 0.92*** 1.04*** 1.12***
(0.05) (0.04) (0.04) (0.03) (0.04) (0.03) (0.03) (0.03) (0.04) (0.05)
SMB 0.15** 0.14** 0.05 0.14*** 0.21*** 0.11** 0.16*** 0.20*** 0.18*** 0.34***
(0.07) (0.06) (0.06) (0.05) (0.05) (0.04) (0.05) (0.05) (0.05) (0.07)
GROW_mp HML 0.35*** 0.00 -0.03 -0.03 -0.01 0.07* 0.09** 0.11** 0.08* 0.19***
(0.06) (0.05) (0.06) (0.04) (0.05) (0.04) (0.04) (0.04) (0.05) (0.06)
WML -0.61*** -0.60*** -0.52*** -0.28*** -0.21*** -0.06 -0.07* -0.05 0.07* 0.04
(0.05) (0.05) (0.05) (0.04) (0.04) (0.03) (0.04) (0.04) (0.04) (0.05)
R-squared 0.91 0.91 0.89 0.91 0.88 0.90 0.89 0.88 0.87 0.85
Adj. R-squared 0.91 0.90 0.89 0.91 0.88 0.90 0.89 0.88 0.87 0.85
Intercept -0.27 0.12 0.36 -0.12 0.48** 0.46** 0.69*** 0.67*** 0.66*** 0.48**
(0.30) (0.27) (0.23) (0.22) (0.22) (0.22) (0.22) (0.20) (0.22) (0.24)
MKTRF 1.17*** 1.10*** 1.12*** 1.11*** 0.95*** 0.93*** 0.96*** 0.83*** 0.97*** 0.87***
(0.05) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.03) (0.04) (0.04)
SMB 0.14** 0.06 0.13** 0.12** 0.19*** 0.19*** 0.14*** 0.15*** 0.16*** 0.19***
(0.07) (0.06) (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) (0.05) (0.06)
QUAL HML 0.35*** 0.30*** 0.08* 0.11** 0.02 -0.05 -0.16*** -0.09** -0.07 -0.01
(0.06) (0.06) (0.05) (0.05) (0.05) (0.05) (0.05) (0.04) (0.05) (0.05)
WML -0.29*** -0.22*** -0.31*** -0.24*** -0.26*** -0.19*** -0.20*** -0.23*** -0.06 -0.07
(0.06) (0.05) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.04)
R-squared 0.88 0.89 0.91 0.91 0.89 0.88 0.88 0.88 0.87 0.83
Adj. R-squared 0.88 0.88 0.91 0.91 0.89 0.88 0.88 0.88 0.86 0.83
The value factor methodologies have two rather contrasting results using FF4. The winner portfolio with
the implementation using Price-to-Book (VAL_ac) shows a significant and positive (0.62***) exposure to HML
while the loser portfolio has significant and negative coefficients (-0.16***). For VAL_ac, the HML factor falls
below 20% below decile 8. The multi-parameter implementation of value (VAL_mf) has the reverse exposure
to HML: the loser portfolio has a significant and positive (0.30***) β to HML, while the winner portfolio has no
statistically significant β. VAL_mf does not show exposure to the academic value factor. Both variants seem
to be ‘Small’ cap tilted: HML β is generally positive across all deciles. Both flavours have negative exposure
to WML in the higher decile portfolios. This result is consistent with other global research on the relationship
between value and momentum (Asness et al., 2013). R-square is lower at the extremes than the middle deciles -
Jan 2022 21
Electronic copy available at: https://ssrn.com/abstract=4000418
indicating some relationship between our chosen parameters and relative performance. We cannot reject the null
hypothesis that α = 0 for the winner portfolios for both variants. VAL_ac meets all our criteria for exposure
to HML while VAL_mf does not. As we have said before, not all factor implementations are equal.
The momentum factor implementations show a difference of degree, but both meet our criteria for factor
exposure. For both, the exposure to WML changes from significant and negative to significant and positive
between the lowest deciles (-0.82*** for MOM_ac and -0.69*** for MOM_vol respectively) and the highest
deciles (0.35*** for MOM_ac and 0.39*** for MOM_vol respectively). WML β falls below 20% below decile
10 for MOM_vol and decile 9 for MOM_ac. Momentum shows short persistence. As we shall see, this has op-
erational implications. Both variations skew towards the smaller market-cap firms in the S&P BSE200 in their
respective loser portfolios. HML exposures do not exhibit any statistically robust relationship across deciles.
The statistically significant α seen in the CAPM regression for the highest decile portfolio disappears with the
FF4 model. The introduction of WML has reduced the α - a robust result showing that the winner portfolios
do have momentum.
The rest of our factor strategies do not have a direct academic factors in the IIMA Dataset to examine
factor exposure. For these other styles we see how much the FF4 decomposition can explain returns and spe-
cific factor exposures. We have two flavours of low volatility: realised volatility (LOVOL), longer-term beta
(BAB). The firms in the portfolios are not controlled for size or value. The BAB variation tilts towards smaller
market-cap firms in the S&P BSE 200 in the winner portfolios (0.14***). The LOVOL variant is less clear in
its relationship with SMB. The winner portfolio has a much less significant SMB β (0.07*). For both variants,
loser portfolios have a significant and positive bias towards value companies. With WML both variations show
increasing coefficients from loser to winner portfolios. The winner portfolios of both variants have similar α and
MKT β with the r-square for LOVOL (0.75) being higher than that of the r-square for BAB (0.70). The FF4
model does not fully account for the returns for the low volatility winner portfolio. To examine persistence, we
look at r-squares of the FF4 model. The relatively low r-squares of the winner portfolio increases to about 0.90
by decile 8 and stays elevated for the lower deciles. Both variants show statistically significant αs even in decile 9.
High beta winner portfolio is biased towards value firms (0.35***) and a negative WML β (-0.57***) with
no statistically significant α and a reasonably high r-square (0.87). The loser portfolio for BETA is nothing
other than the low volatility variant and it is, not surprisingly, similar to the winner portfolios of LOVOL and
BAB with a bias towards small-cap, a modest r-square and a significant and positive α. As described, the high
beta strategy in itself is unlikely to appeal, but the results reiterate the attractiveness of the low volatility style.
The winner portfolios of the multi-parameter growth factor is biased towards small cap (0.34***) and value
(0.19***) with no statistically significant α and a reasonably high r-square (0.85). Across the deciles, no clear
pattern of the SMB and HML is apparent, though r-square drops from the loser decile to the winner decile.
The WML parameter is more evident is the lower deciles than the upper deciles. α is statistically significant for
the loser (and deciles 2 and 3). This is likely a case of ‘correlation but not causation’. Adopting the r-square
Jan 2022 22
Electronic copy available at: https://ssrn.com/abstract=4000418
metric for persistence, by decile 6, the r-square is close to 90%.
Finally, quality. Other than low volatility, quality is the only other factor winner portfolio with a positive
and significant alpha (0.48**). Deciles 5 upward show statistically significant positive α. Other than a bias to
small-cap firms, and a negative exposure to WML, there is little information to be gleaned from the FF4 re-
gressions. The factor style is also one of the most persistent : both in terms of r-squares and statistical evidence
of α. It is only in deciles 4 and 5 that there seems to be a turn for the worse. Quality firms command higher
valuations (negative HML β than firms in the lower deciles: the strength is modest but statistically significant.
In terms of persistence, r-square is 88% from decile 8 and α significance drops to 5% from decile 7. We take
decile 8 as a cut-off.
For size, value and momentum, Table 9 summarises the FER. Except for the multi-parameter value variant,
all the other winner portfolios show a positive exposure to the underlying factor. Size and VAL_ac have
significantly higher exposures to MKT, followed by momentum. Stripping away MKT, the momentum variants
demonstrate strong exposure to the WML factor, with MOM_vol showing the largest FER and is the most
“factor-efficient” portfolio. VAL_ac and size also show the factor exposure. VAL_mp, on the other hand, takes
on more non-target factor exposure, despite being classified as value. Once again, not all factor implementations
are the same.
Table 9: Factor exposures, total absolute non-target exposures (with and without MKT) and FER (with and
without MKT): Jan 2007 to Oct 2021
MKTRF SMB HML WML Non Target FER Non Target ex MKT FER ex MKT
SIZE 1.20 0.41 0.22 -0.55 1.97 0.21 0.77 0.53
VAL_ac 1.19 0.12 0.62 -0.43 1.74 0.36 0.55 1.13
VAL_mp 1.01 0.19 -0.04 -0.22 1.50 -0.03 0.49 -0.08
MOM_ac 1.09 0.15 0.09 0.35 1.33 0.26 0.24 1.46
MOM_vol 1.06 0.10 0.08 0.39 1.24 0.31 0.18 2.17
This table shows the summary of factor exposure (the β from the FF4 regression) for the winner portfolio. FER is
calculated including and excluding MKT. Long-only portfolios inherently have MKT exposure.
• Long-only factor winner portfolios for size, value, momentum show clear evidence of factor exposure.
• Quality and low volatility winner portfolios show positive and statistically significant α in our sample.
• Not all factor styles are equal. Two variants that are similarly categorised can lead to diverging outcomes.
Having multiple parameters is no guarantee of increased exposure to factors. There will be differences in
outcomes based upon the factor measurement and strategy implementation34
• It is critical to ensure the factor exposure is tested empirically. This requires having credible and inde-
pendent datasets which are easily accessible to allow for comprehensive and robust testing.
34
look-back periods, portfolio holding periods, portfolio weight strategies amongst a host of operational parameters
which are essential in translation a theoretical strategy into the real world.
Jan 2022 23
Electronic copy available at: https://ssrn.com/abstract=4000418
Agarwalla et al. (2013) demonstrated evidence for traditional academic factors in the Indian equity market.
Systematic factor portfolios using the largest and most liquid stocks in India also exhibit statistically robust
evidence of factor exposure. The significant β for size, value and momentum of our winner portfolios using over
10 years of data demonstrates this. Moving beyond size, value and momentum, there are other anomalies which
the FF4 is unable to fully explain, which should encourage researchers and managers hunting for the anomalies
to exploit.
After showing that long-only factor portfolios demonstrate robust evidence of factor exposure, we evaluate
whether these winner portfolios show persistence of factor exposure. Specifically, do the firms selected in the
top decile remain in the top decile over time? We construct decile portfolios on a monthly basis. Variables
like price change monthly, while some of the accounting variables are refreshed annually. This will be reflected
in factor scores changes for each firm at varying rates over time. By construction, at t = t+0 the score of
all the constituents in the winner portfolio is 10. Over time, as the constituents scores change and they move
deciles, the average score of the winner portfolio changes. We compute the changes of score over time for all
the winner portfolios during our observation period. From our FF4 analysis, for each factor style we identity a
decile portfolio below which the factor style dissipates. From t=t+1, the original winner portfolio diverges from
the most current winner portfolio as the scores of the individual firms change. As a result, the performance of
the original portfolio moves down from a decile 10 portfolio. When the portfolio score reaches the cut-off, it no
longer can be expected to deliver what the most current winner portfolio delivers.
Table 10: Winner Portfolio Factor Persistence Over Time: Jan 2007 to Oct 2021
SIZE VAL_ac VAL_mp MOM_ac MOM_vol LOVOL BAB BETA GROW_mp QUAL
Time
t 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0 10.0
t+1 9.3 9.6 9.7 9.5 9.5 9.8 9.8 9.7 9.4 9.8
t+2 8.6 9.3 9.3 9.0 9.0 9.6 9.6 9.5 8.8 9.6
t+3 8.0 8.9 9.0 8.5 8.5 9.4 9.4 9.2 8.3 9.4
t+4 7.4 8.6 8.7 8.0 8.1 9.2 9.2 9.0 7.7 9.2
t+5 6.7 8.3 8.4 7.5 7.6 9.0 8.9 8.7 7.2 9.0
t+6 6.1 7.9 8.1 7.0 7.2 8.8 8.7 8.5 6.6 8.8
t+7 5.9 7.9 8.0 6.7 6.9 8.7 8.7 8.3 6.3 8.7
t+8 5.6 7.8 7.9 6.3 6.6 8.6 8.6 8.2 6.0 8.6
t+9 5.4 7.7 7.7 5.9 6.4 8.5 8.6 8.1 5.6 8.5
t+10 5.2 7.6 7.6 5.6 6.1 8.4 8.5 8.0 5.3 8.4
t+11 5.0 7.5 7.5 5.2 5.9 8.3 8.4 7.8 5.0 8.4
t+12 4.9 7.5 7.4 5.1 5.6 8.2 8.4 7.7 4.8 8.3
t+24 3.1 5.9 5.9 5.0 5.3 7.6 7.7 6.3 4.5 7.3
t+36 2.5 5.0 5.5 4.5 4.9 7.5 7.5 5.5 4.1 6.9
This table shows mean score over time for the constituents in D_10 portfolio for each factor. We compute the decile of
each firm in the D_10 portfolio over t = t+1,t+2, t+3..t+12, t+24 and t+36 months and calculate the average across the
cohort. For firms that are no longer part of the S&P BSE 200 index constituents, we assign a score of 0.
Table 10 summarises the persistence of the winner portfolios across the different factor styles. Table 10
should be along with Table 8. GROW_mp dissipates most quickly in our analysis. This information needs
to looked along with factor exposures dissipation analysis from FF4. We had decile 6 as the factor-dissipation
cut-off and the table shows this is reached by t+8. Momentum portfolios also dissipate relatively quickly (also
Jan 2022 24
Electronic copy available at: https://ssrn.com/abstract=4000418
see Jegadeesh and Titman (1993); Raju and Chandrasekaran (2019)). The cut-off from the FF4 analysis of
Decile 9 is reached by t+3. The cut-off for size (decile 5 reached at t+11), VAL_ac (decile 8 reached at t+6),
low volatility (decile 8 reached at t+12) and quality (decile 8 reached at t+12). We have seen that momentum,
low volatility and quality are attractive from a risk-adjusted-return perspective. Of these, the last two are
persistent for longer periods.
All factors dissipate over time. The analysis of persistence and connecting it to the performance of decile
portfolios using tools such as the FF4 regression shows a framework using empirical evidence to review opera-
tional matters like holding periods of portfolios before the factor-style dissipate completely.
Figure 3: Correlation of excess returns of winner portfolios and S&P BSE 200: Jan 2007 to Oct 2021
Source: Authors calculations, Refinitiv. All returns are excess to risk free returns.
Academic long-short factors have low correlation between each other (see Agarwalla et al. (2013) for evi-
dence in India). We have already seen that MKT β is significant in both the one factor and FF4 regressions.
Therefore, we would expect long-only factor-strategy portfolios to have significantly higher correlations between
each other. Figure 3 summarises the correlations between the excess returns of our winner portfolios. This
result should be an important consideration for portfolio construction using long-only factor portfolios.
Styles like low volatility, momentum, quality and growth show lower correlations with other styles. Amongst
flavours of factors, momentum has the higher correlation between the two versions, followed by low volatility
and value has the weakest correlation between the two versions in this study. Once again, the FF4 regressions
in Table 8 shine more light on the underlying reasons why. The relatively high correlations between the two
momentum variations, for example, also brings up an important point. Marketing materials of factor-strategy
funds tend to gloss over the underlying market exposure. Two variations under the same factor style could
offer very little in the way of diversification. investors should examine whether factor-style tracking funds herd
together. The analysis of correlations gives us a quick and useful guide into how factors could be combined
Jan 2022 25
Electronic copy available at: https://ssrn.com/abstract=4000418
together to form multi-factor portfolios.
To examine the size bias of the winner portfolios, we plot the difference in the median sizes of firms in the
monthly winner portfolios for each style and median size of firms in the S&P BSE 200 (Table 2). Figure 4 shows
the results. The SIZE winner portfolio only has ‘small’ firms and shows a negative difference equal to the median
exposure of the S&P BSE 200 constituents to Big. VAL_ac, BETA and GROW_mp portfolios have a bias
towards Small firms. All the other portfolios are over-weight the Big firms. LOVOL is the most overweight Big35
Figure 4: Summarised Size Deviation of Factor Winner Portfolios From S&P BSE 200 : Jan 2007 to Oct 2021
The medians, like any average, reduce the underlying changes. Figure 5 shows the winner portfolios every
month broken down by the size breakpoints determined annually. For each portfolio, there is variation around
a broad trend described by the median. LOVOL has the bias towards Big, but also has a significant amount of
variation across the months. Both versions of momentum also show similar exposure to size. The value variations
are very different in their size exposure with VAL_ac taking a much larger exposure to Small over the period. As
we saw in the correlation analysis, the variations of value have two very different characteristics in size exposure.
The variation to size highlights, again, the importance of empirically examining the portfolio characteristics
across various dimensions to fully understand the nature of risks that empirical based methodologies carry with
them.
35
In Table 8 LOVOL does not have a negative SMB β. This difference arises due to the two very different classifications
between the Data Library for the Indian Market and our approach.
Jan 2022 26
Electronic copy available at: https://ssrn.com/abstract=4000418
Figure 5: Factor Winner Portfolios by Size: Jan 2007 to Oct 2021
For sectoral allocation of winner portfolios, we follow a similar procedure used for Size exposure. For each
winner portfolio we compute the median sector allocation of all constituent firms across the period and compare
it to the median sector allocations of the constituents of the S&P BSE 200 index for the same period. Figure
6 shows the differences for each factor winner portfolio. All the portfolios are underweight Basic Materials.
Energy and Real Estate are almost universally under-weighted. While there is no sector which is over-weighted
by all factor winners, Financial and Healthcare, Technology and Utilities are overweight in a majority of the
strategies we have considered. In particular, the overweight to Financial and Healthcare is significant. The me-
dian sectoral weights give insight into the sectoral ‘preference’ of factors. Quality and low volatility have over
weights to Consumer Non-Cyclicals. Value has a significant over weight to financials36 . Momentum finds favour
in Financials, Healthcare, Technology and Utilities. Quality overweights Financials, Consumer Non-Cyclicals
and Financials.
Figure 7 shows the changes in sectoral allocation across the winner portfolios over time. Like we have
seen for Size, the allocations are dynamic. This dynamic allocation provides opportunities for asset managers
and investors to use factors in constructing portfolios with specific objectives in mind. Extending the returns-
based analysis, we note size and sectoral preferences by the various long-only factor winner portfolios we have
analysed. Adviser and investors looking to add factor exposures to their portfolios are well advised to adopt a
similar approach to optimise the diversification benefits of portfolios. The S&P BSE 200 constituents form the
mainstay of many existing portfolios and adding exposure from factor strategy portfolios constructed from the
same constituents might add less diversification than naive expectations.
36
The naive Price-to-Book metric is likely causing this sectoral weighting. A different value metric will likely have a
different sector exposure.
Jan 2022 27
Electronic copy available at: https://ssrn.com/abstract=4000418
Figure 6: Summarised Sector Deviation of Factor Winner Portfolios From S&P BSE 200 : Jan 2007 to Oct
2021
To examine if a particular set of factors is a consistent winner we look at an annual calendar returns horserace.
Figure 8 shows the outcome of such a horserace computed on a calendar year basis with the winner factor
portfolios. These results are aligned to similar factor and factor tilt portfolio horserace results in various other
markets. Momentum37 , low volatility38 and quality would rank order higher than the broad S&P BSE 200
Index over the various years. But this naive rank ordering ignores the very real volatility in returns across all
the winner portfolios. For example, globally, value, in particular, has underperformed the broad market for
37
MOM_vol ranks ahead of MOM_ac
38
BAB ranks ahead of LOVOL
Jan 2022 28
Electronic copy available at: https://ssrn.com/abstract=4000418
Figure 8: Horserace heatmap: 2007 to 2021
almost two decades but that does not imply that the factor is no longer relevant39 .
The takeaway from the horserace is that while winner momentum, low volatility and quality winner factor
portfolios demonstrate out-performance across our observation period, no single factor in our sample has per-
formed better than all other winner portfolios all the time. There is an ongoing debate on factor rotation in
other markets. In the Indian context, factor rotation is a possible avenue for future empirical research.
Having dealt with the research objectives without accounting for implementation costs, we finally turn to esti-
mating shrinkage of returns on implementation. Implementation costs are largely brokerage and impact costs.
The former is a function of turnover and the latter to the trading in ‘Small’ size firms. First, we turn to turnover.
High turnover is seen as one of the primary drivers of shrinkage of theoretical returns. As we have noted, our
approach40 , by definition, will likely have higher turnover than most actual implementations (see for instance
Ross et al. (2017)).
The number of constituent changes every month is a useful starting point for the turnover. Factor styles
with low persistence will have higher turnover of constituents. Table 11 summarises the annual turnover in
number of constituents and percentage across all the factor styles we consider. Momentum with low persistence
has the most number of constituent changes - almost a third of the constituents change every month. On
the other hand, quality, a persistent factor, shows the fewest changes in constituents. BAB also shows lower
39
See https://www.aqr.com/Insights/Perspectives/A-Gut-Punch for an excellent self-reflection by AQR’s Cliff As-
ness on the challenges of Value.
40
Monthly-rebalanced, EW with no optimisation or buffers to reduce turnover.
Jan 2022 29
Electronic copy available at: https://ssrn.com/abstract=4000418
Table 11: Constituents Turnover in Long Only Winner Factor Portfolios: 2007 to 2021
SIZE VAL_ac VAL_mp MOM_ac MOM_vol LOVOL BAB BETA GROW_mp QUAL
Num Pct Num Pct Num Pct Num Pct Num Pct Num Pct Num Pct Num Pct Num Pct Num Pct
2007-12-31 4.3 21.5 2.3 11.5 2.2 11.0 5.3 26.5 6.3 31.5 3.1 15.5 1.7 8.5 2.7 13.5 5.0 25.0 1.4 7.0
2008-12-31 3.3 16.5 3.2 16.0 2.5 12.5 5.7 28.5 5.9 29.5 2.1 10.5 2.0 10.0 2.4 12.0 5.0 25.0 1.2 6.0
2009-12-31 1.6 8.0 2.5 12.5 1.7 8.5 7.0 35.0 6.2 31.0 1.4 7.0 0.9 4.5 1.5 7.5 6.1 30.5 1.1 5.5
2010-12-31 3.2 16.0 2.2 11.0 3.2 16.0 6.7 33.5 7.1 35.5 2.3 11.5 1.2 6.0 2.8 14.0 6.6 33.0 1.1 5.5
2011-12-31 4.8 24.0 3.8 19.0 2.6 13.0 6.1 30.5 6.5 32.5 3.2 16.0 2.9 14.5 3.7 18.5 4.9 24.5 2.2 11.0
2012-12-31 4.8 24.0 4.4 22.0 3.3 16.5 7.2 36.0 6.9 34.5 2.4 12.0 2.6 13.0 2.4 12.0 6.1 30.5 1.9 9.5
2013-12-31 3.6 18.0 3.4 17.0 2.1 10.5 6.2 31.0 6.6 33.0 3.1 15.5 2.4 12.0 2.6 13.0 5.6 28.0 1.2 6.0
2014-12-31 2.5 12.5 1.2 6.0 1.8 9.0 5.5 27.5 6.4 32.0 2.9 14.5 1.9 9.5 1.2 6.0 4.9 24.5 0.8 4.0
2015-12-31 2.8 14.0 2.0 10.0 3.1 15.5 5.2 26.0 6.7 33.5 2.5 12.5 1.5 7.5 2.2 11.0 5.1 25.5 0.4 2.0
2016-12-31 2.5 12.5 2.0 10.0 2.9 14.5 5.6 28.0 6.8 34.0 1.8 9.0 1.2 6.0 2.5 12.5 4.8 24.0 0.8 4.0
2017-12-31 2.2 11.0 1.4 7.0 2.9 14.5 6.4 32.0 5.6 28.0 2.2 11.0 1.8 9.0 2.9 14.5 5.2 26.0 1.2 6.0
2018-12-31 3.8 19.0 1.8 9.0 2.0 10.0 4.8 24.0 6.8 34.0 2.1 10.5 1.3 6.5 2.0 10.0 3.5 17.5 0.8 4.0
2019-12-31 2.9 14.5 1.7 8.5 2.1 10.5 6.0 30.0 7.6 38.0 2.2 11.0 1.7 8.5 1.9 9.5 5.3 26.5 1.0 5.0
2020-12-31 3.5 17.5 2.5 12.5 1.5 7.5 5.5 27.5 6.0 30.0 3.2 16.0 1.8 9.0 2.2 11.0 5.4 27.0 0.7 3.5
2021-10-31 2.8 14.0 1.7 8.5 2.3 11.5 6.0 30.0 6.8 34.0 1.8 9.0 1.2 6.0 1.8 9.0 3.7 18.5 0.8 4.0
This table shows the mean number of monthly constituent changes in winner portfolios as well as the proportion of the total number of constituents as of December every
year for the prior twelve months expressed as a percentage. For 2021, the calculation ends on October 2021.
constituent changes every month. Using this table, it is quite trivial to estimate the brokerage costs associated
with the new constituents for winner long-only factor portfolios. For instance, assuming a 20 bps brokerage
commission for each leg, for MOM_vol with around 6.5 changes a month, the annual brokerage drag would
be about 1.6%41 , while for the quality portfolio with, on average 1.1 constituent changes per month, it would
be about 30 basis points. Table 12 summarises the annual turnover computed on a calendar year basis using
the definition turnover outlined in Section 3. We show turnoverrebalancing , turnovernew and turnovertotal from
equation 8.
SIZE VAL_ac VAL_mp MOM_ac MOM_vol LOVOL BAB BETA GROW_mp QUAL
Reb New Total Reb New Total Reb New Total Reb New Total Reb New Total Reb New Total Reb New Total Reb New Total Reb New Total Reb New Total
2007-12-31 50 300 340 40 150 190 40 150 190 40 310 360 30 360 400 30 200 230 40 110 140 40 160 210 40 320 380 40 90 130
2008-12-31 50 170 230 40 180 230 50 140 190 40 320 360 40 350 400 40 130 160 50 120 170 50 170 220 50 280 330 50 60 120
2009-12-31 50 90 140 40 150 190 50 100 160 30 430 460 30 380 410 30 80 120 30 50 90 40 80 120 40 380 420 50 70 120
2010-12-31 30 200 230 30 140 170 30 200 220 20 400 420 20 420 440 20 140 160 20 70 100 30 170 200 30 400 420 30 60 100
2011-12-31 40 280 320 40 210 250 30 150 190 30 360 390 20 390 410 20 190 220 20 180 200 40 220 260 30 290 320 30 130 160
2012-12-31 40 290 330 40 260 300 30 210 240 30 440 470 20 420 440 20 150 170 20 160 180 40 150 190 30 370 400 30 120 160
2013-12-31 50 210 250 40 200 240 40 120 160 30 380 410 30 390 430 20 190 210 20 140 170 40 160 200 30 340 380 40 70 110
2014-12-31 40 150 200 40 70 110 40 100 140 30 330 360 30 380 410 30 180 210 30 120 150 40 70 110 40 290 330 40 50 90
2015-12-31 40 160 210 40 120 160 30 180 220 30 310 340 20 400 420 20 150 170 20 90 120 40 130 170 40 300 340 30 20 60
2016-12-31 40 150 190 30 120 160 30 170 210 30 340 360 20 400 430 20 110 130 20 70 90 40 150 190 30 300 330 30 50 80
2017-12-31 30 140 180 40 90 130 30 170 200 20 380 410 20 330 360 20 130 150 30 110 130 40 180 220 30 310 340 30 70 100
2018-12-31 40 210 250 40 100 140 30 120 150 30 280 320 30 410 430 20 120 150 30 80 110 40 120 160 40 210 250 30 50 80
2019-12-31 50 170 220 50 90 150 40 120 160 20 360 390 20 450 470 20 130 160 30 100 130 50 110 160 30 320 350 40 60 100
2020-12-31 40 210 260 40 150 190 40 90 130 30 330 370 30 370 400 30 190 220 30 110 140 40 110 150 30 320 360 40 40 80
2021-10-31 40 170 210 40 100 130 30 140 170 40 350 400 40 390 440 30 110 140 30 70 100 40 120 160 50 210 270 30 50 80
This table shows the calculated turnoverrebalancing , turnovernew and turnovertotal of the equal-weighted winner factor portfolios as of December every year for the prior twelve months expressed as a percentage. For 2021, the calculation ends on
October 2021.The numbers are rounded to nearest 10.
As expected, momentum winner portfolios show the highest turnovertotal - almost turning over once every
quarter. The average 6.5 constituent changes per month for MOM_vol in Table 11 translates into a close to
400% turnover. MOM_vol, which uses the higher frequency 6-month measure as one of its components, has
a higher turnover than MOM_ac, which uses a longer frequency 12-month measure. turnovernew drives a
majority of turnovertotal - not just for momentum, but for all factors. Size, too, has a high turnovertotal driven
by the change in market-cap amongst the smallest firms by market-cap in the S&P BSE 200. QUAL, LOVOL
and BAB have relatively lower turnovertotal .
The turnover arising from rebalancing of the equal weighted portfolio is a much smaller contributor to
turnovertotal . Most of the turnover costs arise from changes to constituents in our approach. Other weighting
schemes will have markedly different turnover rates. Many factor-strategies use longer holding periods and
buffers as a means to control turnover. In addition, their weighting schemes are signal-weighted-capital-scaled.
While these measures reduce turnover, they will obviously would impact factor-exposure as well as persistence.
41
6.5 trades ÷ 20 stock portfolio × 2 legs × 12 months × 20bps
Jan 2022 30
Electronic copy available at: https://ssrn.com/abstract=4000418
The framework we have developed in this paper can be applied to evaluate these implementation decisions as
well.
The second driver of costs is impact costs : the slippage in traded price due to the demand and supply of
stocks during trading hours. we can add the impact costs, arising from liquidity based on size. For instance,
for the size winner portfolio, has an average change in constituents of 3.2 stocks per month. All of these are is
small cap firms. Assuming an average impact cost of 30 bps per trade, we would compute impact costs to be
approximately 1.2%42 .
To compute the total costs, we add the costs of brokerage and impact for the monthly rebalancing of port-
folios back to equal weights. We follow a similar process used in the computation of these costs for changes to
constituents except we only compute single-leg costs and multiply these by the average rebalancing turnover for
the style.
Table 13 shows the shrinkage of the average 3-year alphas using our estimates of brokerage and impact
costs. While the alpha shrinks due to the costs, winner portfolios from momentum, low volatility, and quality,
as defined in this paper, all continue to show positive mean 3-year rolling alphas in excess of the S&P BSE 200
returns. Brokerage costs could add upto 2.0% and impact costs upto 2.4% for our 20 stock portfolio with no
optimisation to reduce costs and turnover. For high turnover strategies like momentum and beta, the shrinkage
is between 3 and 4% per annum. For ‘Small’-cap-heavy strategies43 , the impact cost shrinks the alpha between
1.2-2.4% pa. Low turnover strategies44 do not suffer significantly due to implementation costs.
Table 13: Estimated Alpha Shrinkage of Winner Factor Portfolios including transaction costs
Ex-costs 3 yr Alpha Brokerage Impact Total Costs ex Tax Inc Costs 3 yr Alpha
SIZE -3.7 1.6 2.4 4.0 -7.7
VAL_ac -9.6 1.4 1.8 3.3 -12.9
VAL_mp 0.4 1.3 1.2 2.5 -2.1
MOM_ac 6.5 1.9 1.8 3.7 2.7
MOM_vol 9.4 2.0 1.7 3.7 5.6
LOVOL 7.7 1.1 0.7 1.8 5.9
BAB 8.5 1.0 0.8 1.9 6.7
BETA -13.9 1.4 1.6 3.0 -17.0
GROW_mp -0.8 1.9 2.1 4.0 -4.7
QUAL 7.1 1.1 0.9 2.0 5.2
This table shows the mean rolling 3 year alphas shown in Table 5, estimated brokerage and impact costs, and
alpha after deducting these costs. We assume brokerage costs of 20 bps for each leg. For impact costs, we
estimate 5 bps for firms categorised as ’Big’ and 30 bps for firms categorised as ‘Small’. Brokerage costs are
estimated from mean constituent turnover in Table 11. Impact cost estimates use the average size exposure of
the long-only winner factor portfolios in Figure 5.
Our results are similar to many studies : momentum strategies having high turnover show a non-trivial
shrinkage of theoretical alphas, while low volatility and quality with lower turnover show a fairly small shrinkage.
In the Indian context, while some long only factor strategies as we have constructed them show negative alpha,
42
3.2 trades per month ÷ 20 stock portfolio × 2 legs × 12 months ×30 bps.
43
Size, momentum, value
44
Quality, low volatility
Jan 2022 31
Electronic copy available at: https://ssrn.com/abstract=4000418
momentum, low volatility, and quality show healthy alphas to the broader market nett of key transaction costs.
In these cases, going ‘long’ factors is not ‘short’ change.
5 Conclusion
Building factor-style equal-weighted winner portfolios from the S&P BSE 200, the most liquid segment of the
Indian equity market, we show evidence of factor exposure for size, value and momentum factors. While aca-
demic factors for low volatility and quality are not available in India, we show that winner portfolios for these
two ‘factors’ show statistically significant α against both the one-factor and FF4 models. Factor-style imple-
mentations of momentum, low-volatility and quality show robust evidence of α relative to the S&P BSE 200.
Winner portfolios of size, value, high beta and to a lesser extent growth did not show out-performance for the
observation period and implementation we chose. Factor exposures persist over time. Different strategies have
different persistence trajectories. As expected, there is significant exposure to the market which makes the
correlations between long-only factor strategies higher than the theoretical long-short factor indicates.
The study sheds new light by estimating post-implementation-cost alphas as previous studies have not in-
cluded costs. Despite our conservative estimates of costs, alpha remains positive after accounting for brokerage
and impact costs for momentum, low volatility, and quality implementations. We are confident that real-world
implementations, with more sophisticated trading algorithms, will generate positive expected alphas over the
long term. We include alternative calculation methodologies for some factors to show that not all implementa-
tions of factor strategies are the same. The study also shows size and sector biases in the strategies that are
important considerations for diversification benefits when evaluating factor strategy implementations. Factors
are not a panacea or a silver bullet for investors. Factors work over long periods. Importantly, they do not work
all the time. Investors with shorter time horizons tend to abandon strategies out of favour and switch to those
in favour. Using an annual calendar returns horserace, we show that while there is no one consistent winner,
momentum, low volatility, and quality rank higher than the broad S&P BSE 200 index over the period under
consideration.
As factor models are empirical asset pricing models, the availability of cheap and powerful computing power
and access to data sources may result in false positives. Therefore, a blend of economic intuition, behavioural
finance insights, and quantitative robustness is imperative to ensure that implemented models do not end up in
the “factor zoo” graveyard. The framework outlined in this study can be used to evaluate factor-strategy im-
plementations and make important operational decisions to translate theoretical factor models to the real world.
A logical extension of our study would be to explore factor rotation, multi-factor portfolio combinations
and tactical management of single-factor portfolios by looking at factor behaviour over business and economic
cycles. This study can be extended to broaden the universe from the top 200 firms by market capitalisation.
The small-cap firms tend to show high theoretical alpha but come with higher implementation costs.
Jan 2022 32
Electronic copy available at: https://ssrn.com/abstract=4000418
References
Agarwalla, Sobhesh K.; Jacob, Joshy, and Varma, Jayanth R. Four factor model in Indian equities mar-
ket. WP No 2013 09 05, 2013. URL http://www.iimahd.ernet.in/~jrvarma/Indian-Fama-French-Momentum/
four-factors-India-90s-onwards-IIM-WP-Version.pdf.
Agarwalla, Sobhesh K.; Jacob, Joshy; Varma, Jayanth R., and Vasudevan, E. Betting against Beta in the Indian Market. (W.P.
No. 2014-07-01), July 2014. URL https://web.iima.ac.in/assets/snippets/workingpaperpdf/5848695332014-07-01.pdf.
Agarwalla, Sobhesh K.; Jacob, Joshy, and Varma, Jayanth R. Size, Value, and Momentum in Indian Equities. Vikalpa, 42(4):
211–219, 2017. doi: 10.1177/0256090917733848. URL https://doi.org/10.1177/0256090917733848.
Ali, Asgar; Badhani, K. N., and Kumar, Ashish. Does the low-risk anomaly exist in the Indian equity market? A
test using alternative risk measures. Journal of Economic Studies, ahead-of-print(ahead-of-print), 2021/12/03 2021. doi:
10.1108/JES-07-2021-0374. URL https://doi.org/10.1108/JES-07-2021-0374.
Ang, Andrew; Hodrick, Robert J.; Xing, Yuhang, and Zhang, Xiaoyan. The Cross-Section of Volatility and Expected Returns. The
Journal of Finance, 61(1):259–299, 2006. doi: https://doi.org/10.1111/j.1540-6261.2006.00836.x. URL https://onlinelibrary.
wiley.com/doi/abs/10.1111/j.1540-6261.2006.00836.x.
Ang, Andrew; Hodrick, Robert; Xing, Yuhang, and Zhang, Xiaoyan. High Idiosyncratic Volatility and Low Returns: International
and Further U.S. Evidence. National Bureau of Economic Research, Working Paper, 2008. doi: 10.3386/w13739.
Ansari, Valeed Ahmad and Khan, Soha. Momentum anomaly: Evidence from India. Managerial Finance, 38(2):206–223, Jan 2012.
Arnott, Robert D.; Kalesnik, V., and Wu, Lillian J. The Incredible Shrinking Factor Return. Capital Markets: Asset Pricing &
Valuation eJournal, 2017. URL https://ssrn.com/abstract=3040964.
Arnott, Robert D.; Harvey, Campbell R., and Markowitz, Harry. A Backtesting Protocol in the Era of Machine Learning. Available
at SSRN, November 2018. URL https://ssrn.com/abstract=3275654.
Asness, Clifford S.; Moskowitz, Tobias J., and Pedersen, Lasse Heje. Value and Momentum Everywhere. The Journal of Finance,
68(3):929–985, Jun 2013.
Asness, Clifford S.; Frazzini, Andrea, and Pedersen, Lasse Heje. Quality Minus Junk. Review of Accounting Studies, 24(1):34–112,
May 2018. doi: 10.1007/s11142-018-9470-2.
Barberis, Nicholas; Shleifer, Andrei, and Vishny, Robert W. A model of investor sentiment. Journal of Financial Economics, 49(3):
307 – 343, Sep 1998. ISSN 0304-405X. doi: https://doi.org/10.1016/S0304-405X(98)00027-0. URL http://www.sciencedirect.
com/science/article/pii/S0304405X98000270.
Bartram, Söhnke M.; Lohre, Harald; Pope, Peter F., and Pallasena Ranganathan, Ananthalakshmi. Navigating the Factor Zoo
Around the World: An Institutional Investor Perspective. SSRN Electronic Journal, June 2020. URL https://ssrn.com/
abstract=3510989.
Bender, Jennifer C. and Wang, Taie. Multi-Factor Portfolio Construction for Passively Managed Factor Portfolios. May 2015. URL
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3080348.
Bender, Jennifer C. and Wang, Taie. Can the Whole Be More Than the Sum of the Parts? Bottom-Up versus Top-Down Multifactor
Portfolio Construction. The Journal of Portfolio Management, 42(5):39–50, 2016. ISSN 0095-4918. doi: 10.3905/jpm.2016.42.
5.039. URL https://jpm.pm-research.com/content/42/5/39.
Blitz, David and Hanauer, Matthias. How many factors are there? Or how to navigate the ‘factor zoo’, March 2020. URL
https://www.robeco.com/docm/docu-202003-how-to-navigate-the-factor-zoo-us.pdf.
Blitz, David; Baltussen, Guido, and van Vliet, Pim. When Equity Factors Drop Their Shorts. Financial Analysts Journal, 76(4):
73–99, 2020. doi: 10.1080/0015198X.2020.1779560. URL https://doi.org/10.1080/0015198X.2020.1779560.
Cakici, Nusret; Fabozzi, Frank J., and Tan, Sinan. Size, value, and momentum in emerging market stock returns. Emerging
Markets Review, 16:46–65, 2013. ISSN 1566-0141. doi: https://doi.org/10.1016/j.ememar.2013.03.001. URL https://www.
sciencedirect.com/science/article/pii/S1566014113000198.
Cakici, Nusret; Tang, Yi, and Yan, An. Do the size, value, and momentum factors drive stock returns in emerging markets? Journal
of International Money and Finance, 69:179 – 204, Dec 2016. ISSN 0261-5606. doi: https://doi.org/10.1016/j.jimonfin.2016.06.
001. URL http://www.sciencedirect.com/science/article/pii/S026156061630047X. SI - Tribute Jim Lothian.
Carhart, Mark M. On Persistence in Mutual Fund Performance. The Journal of Finance, 52(1):57–82, 1997. doi: 10.1111/j.
1540-6261.1997.tb03808.x. URL http://onlinelibrary.wiley.com/doi/abs/10.1111/j.1540-6261.1997.tb03808.x.
Cochrane, John H. Presidential Address: Discount Rates. Journal of Finance, 66(4):1047–1108, August 2011.
Daniel, Kent and Titman, Sheridan. Evidence on the Characteristics of Cross Sectional Variation in Stock Returns. The Journal
of Finance, 52(1):1–33, 1997. doi: https://doi.org/10.1111/j.1540-6261.1997.tb03806.x. URL https://onlinelibrary.wiley.
com/doi/abs/10.1111/j.1540-6261.1997.tb03806.x.
Jan 2022 33
Electronic copy available at: https://ssrn.com/abstract=4000418
De Bondt, Werner F. M. and Thaler, Richard H. Does the Stock Market Overreact? The Journal of Finance, 40(3):793–805, Dec
1985. doi: 10.1111/j.1540-6261.1985.tb05004.x. URL http://onlinelibrary.wiley.com/doi/abs/10.1111/j.1540-6261.1985.
tb05004.x.
de Groot, Wilma; Huij, Joop, and Zhou, Weili. Another Look at Trading Costs and Short-Term Reversal Profits. Journal of
Banking and Finance, 36(2), November 2012a. URL https://ssrn.com/abstract=1963131.
de Groot, Wilma; Pang, Juan, and Swinkels, Laurens. The cross-section of stock returns in frontier emerging markets. Journal
of Empirical Finance, 19(5):796–818, 2012b. ISSN 0927-5398. doi: https://doi.org/10.1016/j.jempfin.2012.08.007. URL https:
//www.sciencedirect.com/science/article/pii/S0927539812000643.
de Prado, Marcos López. The 10 Reasons Most Machine Learning Funds Fail. The Journal of Portfolio Management, 44(6):120,
06 2018. doi: 10.3905/jpm.2018.44.6.120. URL http://jpm.pm-research.com/content/44/6/120.abstract.
Edwards, A. W. F. and Cavalli-Sforza, L. L. A Method for Cluster Analysis. Biometrics, 21(2):362–375, 1965. ISSN 0006341X,
15410420. URL http://www.jstor.org/stable/2528096.
Fama, Eugene F. Components of Investment Performance. The Journal of Finance, 27(3):551–567, 1972. ISSN 00221082, 15406261.
URL http://www.jstor.org/stable/2978261.
Fama, Eugene F. and French, Kenneth R. Dividend Yields and Expected Stock Returns. Journal of Financial Economics, 1988.
Fama, Eugene F. and French, Kenneth R. Common risk factors in the returns on stocks and bonds. Journal of Financial Economics,
33(1):3–56, 1993. doi: 10.1016/0304-405x(93)90023-5.
Fama, Eugene F. and French, Kenneth R. Size, Value, and Momentum in International Stock Returns. SSRN Electronic Journal,
Jun 2011. doi: 10.2139/ssrn.1720139. URL http://ssrn.com/abstract=1720139.
Fama, Eugene F. and French, Kenneth R. Size, value, and momentum in international stock returns. Journal of Financial
Economics, 105(3):457–472, May 2012. doi: 10.1016/j.jfineco.2012.05.011. URL http://www.sciencedirect.com/science/
article/pii/S0304405X12000931.
Frazzini, Andrea and Pedersen, Lasse Heje. Betting against beta. Journal of Financial Economics, 111(1):1–25, 2014. doi:
10.1016/j.jfineco.2013.10.005.
Hanauer, Matthias X. and Linhart, Martin. Size, Value, and Momentum in Emerging Market Stock Returns: Integrated or
Segmented Pricing? Asia-Pacific Journal of Financial Studies, 44(2):175–214, 2015. doi: https://doi.org/10.1111/ajfs.12086.
URL https://onlinelibrary.wiley.com/doi/abs/10.1111/ajfs.12086.
Harvey, Campbell R. and Liu, Yan. False (and Missed) Discoveries in Financial Economics. November 2017. URL https:
//ssrn.com/abstract=3073799.
Harvey, Campbell R.; Liu, Yan, and Zhu, Heqing. . . . and the Cross-Section of Expected Returns. The Review of Financial Studies,
29(1):5–68, 10 2015. ISSN 0893-9454. doi: 10.1093/rfs/hhv059. URL https://doi.org/10.1093/rfs/hhv059.
Hou, Kewei; Xue, Chen, and Zhang, Lu. Replicating Anomalies. The Review of Financial Studies, 33(5):2019–2133, 12 2018. ISSN
0893-9454. doi: 10.1093/rfs/hhy131. URL https://doi.org/10.1093/rfs/hhy131.
Huij, Joop; Lansdorp, Simon; Blitz, David, and van Vliet, Pim. Factor Investing: Long-Only versus Long-Short. SSRN Electronic
Journal, page 19, Mar 2014. doi: 10.2139/ssrn.2417221. URL https://papers.ssrn.com/abstract=2417221.
Hunstad, Michael and Dekhayser, Jordan. Evaluating the Efficiency of “Smart Beta” Indexes. The Journal of Index Investing, 6
(1):111–121, 2015. ISSN 2154-7238. doi: 10.3905/jii.2015.6.1.111. URL https://jii.pm-research.com/content/6/1/111.
Israel, Ronen; Jiang, Sarah, and Ross, Adrienne. Craftsmanship Alpha: An Application to Style Investing. The Journal of Portfolio
Management, 44(2):23–39, Dec 2017. doi: 10.3905/jpm.2017.2017.1.075.
Jegadeesh, Narasimhan and Titman, Sheridan. Returns to Buying Winners and Selling Losers: Implications for Stock Market
Efficiency. The Journal of Finance, 48(1):65, 1993. doi: 10.2307/2328882.
Joshipura, Mayank and Joshipura, Nehal. The Volatility Effect: Evidence from India. Applied Finance Letters, 5, 06 2016. doi:
10.24135/afl.v5i1.32.
Lakonishok, Josef; Shleifer, Andrei, and Vishny, Robert W. Contrarian Investment, Extrapolation, and Risk. The Journal of
Finance, 49(5):1541–1578, 1994. doi: https://doi.org/10.1111/j.1540-6261.1994.tb04772.x. URL https://onlinelibrary.wiley.
com/doi/abs/10.1111/j.1540-6261.1994.tb04772.x.
Mclean, R. David and Pontiff, Jeffrey. Does Academic Research Destroy Stock Return Predictability? The Journal of Finance, 71(1):
5–32, 2016. doi: https://doi.org/10.1111/jofi.12365. URL https://onlinelibrary.wiley.com/doi/abs/10.1111/jofi.12365.
Plyakha, Yuliya; Uppal, Raman, and Vilkov, Grigory. Equal or Value Weighting? Implications for Asset-Pricing Tests. SSRN
eLibrary, 2014. doi: 10.2139/ssrn.1787045.
Raju, Rajan. Implementing a Systematic Long-only Quality Strategy in the Indian Market. 2019. URL http://ssrn.com/abstract=
3490999.
Raju, Rajan and Chandrasekaran, Abhijit. Implementing a Systematic Long-only Momentum Strategy: Evidence From India.
SSRN eJournal, October 2019. URL https://ssrn.com/abstract=3510433o.
Jan 2022 34
Electronic copy available at: https://ssrn.com/abstract=4000418
Ross, Adrienne; Moskowitz, Tobias; Israel, Ronen, and Serban, Laura. Implementing Momentum: What Have We Learned? SSRN
Electronic Journal, 01 2017. doi: 10.2139/ssrn.3081165.
Sharpe, William F. Mutual Fund Performance. The Journal of Business, 39(1):119–138, 1966. ISSN 00219398, 15375374. URL
http://www.jstor.org/stable/2351741.
Jan 2022 35
Electronic copy available at: https://ssrn.com/abstract=4000418