You are on page 1of 19

DOI: 10.1111/fima.

12272

ORIGINAL ARTICLE

Can mispricing explain the value premium?


Jeffrey F. Jaffe1 Jan Jindra2 David J. Pedersen3
Torben Voetmann4,5

1 Finance Department, The Wharton School,

University of Pennsylvania, Philadelphia, Abstract


Pennsylvania Empirical research finds that stocks with low market-to-book (MTB)
2 The Securities and Exchange Commission, San
ratios outperform stocks with high MTB ratios. Rhodes-Kropf, Robin-
Francisco, California son, and Viswanathan separate the MTB ratio into mispricing and
3 Rutgers School of Business—Camden, Camden,
growth options components. We report that the mispricing com-
New Jersey
ponent, but not the growth options component, predicts abnormal
4 University of San Francisco, San Francisco,

California
returns for up to 5 years. We also find that the mispricing compo-
5 The Brattle Group, San Francisco, California nent, but not the growth options component, provides incremental
information relative to existing asset pricing models. Moreover, after
Correspondence controlling for mispricing, value no longer beats growth. Overall, our
David J. Pedersen, Rutgers School of Business—
evidence is consistent with a behavioral explanation of the value
Camden, 227 Penn Street, Camden, New Jersey
08102. premium.
Email: David.Pedersen@Rutgers.edu
KEYWORDS
Funding information value premium, mispricing, growth options, return predictability
David Whitcomb Center for Research in Financial
Services of Rutgers University J E L C L A S S I F I C AT I O N S
G14

1 INTRODUCTION

For over 80 years, investors have noted the profitability of value strategies, that is, the tendency of stocks with high
value-to-price ratios to outperform stocks with low value-to-price ratios (Graham & Dodd, 1934). The past 25 years has
seen an ongoing debate within the academic community over the source of this value premium. One side argues that
differences in the book equity-to-market equity (BTM) ratios reflect differences in the systematic risks (Fama & French,
1992). According to this risk-based explanation, high BTM stocks are riskier than low BTM stocks and, consequently,
earn a return premium. The other side offers a behavioral explanation. High BTM stocks are undervalued, whereas low
BTM stocks are overvalued (Lakonishok, Shleifer, & Vishny, 1994). Thus, the value premium is a manifestation of the
correction of these mispricings.
A challenge in distinguishing between these two alternatives is the identification of mispriced stocks. In this paper,
we rely on the work of Rhodes-Kropf, Robinson, and Viswanathan (RKRV) (2005) to decompose a firm’s market equity-
to-book equity (MTB) ratio into a mispricing component and a growth options component. RKRV’s (2005) model yields


c 2019 Financial Management Association International

Financial Management. 2019;1–19. wileyonlinelibrary.com/journal/fima 1


2 JAFFE ET AL .

one of the most common misvaluation measures in the literature.1 However, one aspect of the literature using the
RKRV-based (2005) measure is that the studies assume the measure reflects misvaluation. To our knowledge, no one
has systematically investigated the asset pricing implications of RKRV’s (2005) decomposition.
The goal of this paper is to use RKRV’s (2005) identification of the mispricing and growth options components of
MTB to examine the value premium. First, we test whether the mispricing and growth options components of MTB pre-
dict future returns. Using a large sample of U.S. stocks, we find that underpriced stocks, according to the RKRV (2005)
model, subsequently outperform overpriced stocks consistent with the hypothesis that the RKRV (2005) model iden-
tifies mispricing. This conclusion holds for excess returns and for abnormal returns based on the capital asset pricing
model (CAPM), the five-factor model of Fama and French (2015) (FF-5), and the M-4 factor model of Stambaugh and
Yuan (2017) (M-4). The mispricing alphas from the factor models range from 24 to 47 basis points per month.
The ability of the growth options component to predict future returns is far less robust. A portfolio composed of long
positions on firms with low-growth options and short positions on firms with high-growth options generates an insignif-
icant excess return. The CAPM alpha on this portfolio is also insignificant. However, the alpha is negative and significant
once we control for the additional risk factors in the FF-5 model. Although the M-4 alpha of the long–short-growth
options portfolio is also negative, the effect is only marginally significant. Thus, even when significant, the alphas from
growth options are negative suggesting a contrarian value strategy.
Next, we determine whether each MTB component adds incremental information to the existing asset pricing mod-
els. To do so, we follow standard steps in asset pricing and construct factors from the mispricing and growth options
components of MTB. The time series premium on the mispricing factor is roughly equivalent to the premium on the
standard value factor (HML), while the growth options factor does not earn a significant premium. Following Barillas
and Shanken (2017), we then regress each of our two MTB factors on the factors in the existing asset pricing models.
Barillas and Shanken (2017) argue that a significant alpha indicates that the new factor contains information useful to
investors implying that one should augment the existing model with the new factor. For the mispricing factor, we find
a positive and significant alpha, ranging between 17 and 43 basis points per month, relative to the CAPM, FF-5, and
M-4 models. In particular, the FF-5 alpha indicates that an investor who already trades the value factor, along with the
other FF-5 factors, can benefit from the information contained in the mispricing factor. Thus, the value factor does not
subsume the mispricing factor.
Again, the growth options results are less consistent. The CAPM alpha of the growth options factor is positive and
marginally significant. Its FF-5 alpha is negative and significant, and its M-4 alpha is insignificant. Overall, it is unclear
whether the growth options factor adds anything to the existing asset pricing models.
We then consider whether the mispricing factor can explain the well-known value premium. Consistent with the
literature, we find that the Fama-French (2015) book-to-market Value-minus-Growth (Value – Growth) portfolio earns
a significant premium of 43 basis points per month over our sample period. The standard market and size factors are
unable to explain this premium. However, when we also control for the mispricing factor, the alpha on the “Value –
Growth” portfolio is statistically insignificant and falls to –2 basis points per month. In other words, the value premium
disappears. Using the Barillas-Shanken (2017) framework, we next regress the value factor (i.e., HML) on the market
factor, the size factor (i.e., SMB), and the mispricing factor. The alpha from this three-factor model is a statistically
insignificant 6 basis points per month. Because the alpha on HML is a statistically significant 48 basis points per month
in a model that only includes the market and size factors, we conclude that the mispricing factor subsumes the value
factor. Overall, the evidence strongly suggests that mispricing explains the value premium.
There is little evidence that growth options explain the value premium. The alpha on the “Value – Growth” portfolio
is positive and significant based on a model that includes the market factor, the size factor, and the growth options

1 As of December 2017, there are 101 citations of RKRV (2005) in the following journals: Journal of Finance, Review of Financial Studies, Journal of Financial

Economics, Journal of Financial and Quantitative Analysis, Management Science, Journal of Corporate Finance, and Financial Management. By our count, 31 of these
articles use the RKRV-based (2005) measure as an independent variable, and five articles use it as a dependent variable.
JAFFE ET AL . 3

factor. Likewise, the alpha from a regression of HML on the market, size, and growth option factors is positive and
significant.
We also examine the long-run predictive power of RKRV’s (2005) measure of mispricing. The Undervalued-minus-
Overvalued (Undervalued – Overvalued) portfolio generates statistically significant alphas that average 0.235%
per month over the 5-year period following the misvaluation measurement implying a cumulative alpha of 14.10%
(=0.235% × 60). However, the results indicate that the RKRV (2005) model has little predictive power after the first
5 years.
Our findings contribute to the literature on behavioral explanations of the value premium. Lakonishok et al. (1994)
suggest that investors following a naïve strategy of projecting future growth rates based on past performance con-
tribute to the value premium. Lakonishok et al. (1994) conclude that because the actual future growth rates of growth
stocks relative to those of value stocks generally turn out to be lower than what they were in the past or what the mar-
ket expected them to be, growth stocks underperform value stocks. Moreover, the authors find little support for the
notion that value strategies are fundamentally riskier. Using a proxy for the likelihood of financial distress, Griffin and
Lemmon (2002) report that the standard distress risk explanation for the value premium does not hold for the subset of
firms for which one would expect it to be the strongest. Consistent with a behavioral explanation, the authors find that
the value premium is largest in the set of firms for which rational arbitrage is less likely to be effective (e.g., small firms
with low analyst coverage). We incorporate the RKRV (2005) measure of mispricing into the asset pricing literature
and conclude that the mispricing premium explains the value premium.
We also contribute to the literature that investigates the value premium by decomposing MTB (or BTM). Daniel
and Titman (2006) decompose BTM into two components: one associated with past accounting performance and one
orthogonal to past performance. They find future returns are only related to the orthogonal component. Fama and
French (2008) decompose BTM into three components: (a) a historical BTM, (b) changes in book equity, and (c) changes
in price. They find all three components are related to future returns. Gerakos and Linnainmaa (2017) decompose BTM
into two components: one related to changes in firm size and the other orthogonal to changes in firm size. They find that
future returns are only related to the component derived from changes in firm size. Although Gerakos and Linnainmaa
(2017) do not enter the debate between behavioral and risk-based explanations, Daniel and Titman (2006) and Fama
and French (2008) conclude that their results could be consistent with either risk or mispricing. According to Fama and
French (2008, p. 2,994), “whether breaking BTM into its components enhances estimates of expected returns sheds no
light on the time worn conundrum of whether expected returns are the result of rational or irrational pricing.” We shed
light on this issue by decomposing MTB into a mispricing component and a growth options component.
We are aware of two studies that analyze the predictive power of RKRV’s (2005) model using smaller samples of
stocks following corporate events. Hertzel and Li (2010) find that future abnormal returns are negatively related to
the degree of overvaluation for a sample of firms with seasoned equity offerings. Fu, Lin, and Officer (2013) determine
that future abnormal returns for overvalued acquirers are negative, but that abnormal returns for nonmerging over-
valued control firms in the same industry are not significantly negative. We believe our paper is the first to show that
the misvaluation identified by RKRV’s (2005) model predicts future returns across a large, diverse sample of stocks.
Moreover, we find that RKRV’s (2005) measure of mispricing adds information to the existing asset pricing models and
can explain the value premium.
The rest of the paper is organized as follows. The data are described in Section 2. Section 3 reports our empirical
results, whereas Section 4 provides our conclusions.

2 DATA

We start with all stocks for which Compustat and the Center for Research in Security Prices (CRSP) provide sufficient
data for our analysis. Following RKRV (2005), we remove stock years where either the ratio of MTB is above 100 or
the market value of equity (ME) is below $10 million. As in RKRV (2005), a firm’s market value of equity is measured as
4 JAFFE ET AL .

of the end of June of each year and all accounting measures, including book equity, are measured as of the firm’s fiscal
year-end in the preceding calendar year.
Table 1 presents the summary statistics for our sample. The second column of the table reports the average market
value of equity across our firms for each year of the sample. The average market value of equity generally increases
over the sample period with some periods of slight declines, consistent with the returns that U.S. stocks earned over
this period. The next column provides the average market value of the equity-to-book value of equity (MTB) ratio. MTB
generally increases early in the sample period, reaching a high point in 2000. The rise in the market-to-book over the
sample period is far less than the rise in the market value of equity.
RKRV (2005, eq. (3)) decompose the natural logarithm of MTB into three components:

( ) ( ) ( ) ( )
mit − bit = mit − v 𝜃it ; 𝛼jt + v 𝜃it ; 𝛼jt − v 𝜃it ; 𝛼j + v 𝜃it ; 𝛼j − bit , (1)
⏟⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏟ ⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ ⏟⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏟
FSE TSSE LRVTB

where mit is the natural logarithm of the market value of equity of firm i in year t, bit is the natural logarithm of the
firm’s book value of equity in year t, v(𝜃 it ; 𝛼 jt ) is the firm’s fundamental value of equity in year t based on a set of firm
characteristics (𝜃 it ) and their contemporaneous multiples (𝛼 jt ), and v(𝜃 it ; 𝛼 j ) is the firm’s fundamental value of equity in
year t based on firm characteristics and their long run multiples (𝛼 j ).2 RKRV (2005) label the difference between the
first two terms on the right-hand side of the equation as firm-specific error (FSE), the difference between the next two
terms as time series sector error (TSSE), and the difference between the last two terms as the long-run value-to-book
ratio (LRVTB), expressed in logarithms. FSE is an estimate of misvaluation using only contemporaneous data on a firm’s
sector. TSSE estimates the difference in firm equity value based on deviations between the contemporaneous multiples
of a firm’s industry and the long-run multiples of that industry. LRVTB is the difference between a firm’s fundamental
value of equity, as estimated from long-run multiples in the firm’s industry, and the firm’s book value of equity. Our
paper measures a firm’s mispricing (or misvaluation) by the sum of FSE and TSSE (FSE + TSSE) and growth options by
LRVTB.
We estimate the components of Equation (1) above. We begin by using the following model from RKRV (2005, eq.
15) to estimate the contemporaneous multiples (𝛼 ijt ):

mit = 𝛼0jt + 𝛼1jt bit + 𝛼2jt ln(NI)+ it + 𝛼3jt I(<0) ln(NI)+ it + 𝛼4jt LEVit +𝜀it , (2)

where mit is the natural logarithm of the market value of equity of firm i in year t, bit is the natural logarithm of the
firm’s book value of equity in year t, NI+ it is the absolute value of its net income in year t, I(<0) is an indicator function for
negative net income observations, and LEVit is the leverage ratio defined as one minus the ratio of firm i’s market value
of equity to its market value of assets in year t. Following RKRV (2005), we define the market value of a firm’s assets
as the market value of equity plus the book value of assets minus the sum of the book value of equity and deferred
tax assets. The model is estimated cross-sectionally each year for each of the 12 Fama and French industries. Long-
run multiples (𝛼 j ) are calculated each year by industry as the average annual multiple over the period from 1968 to the
current year. We start our sample period in 1973 to allow for a minimum of 6 years in each long-run multiple calculation.
Because we have asset pricing model data through June 2016, the final RKRV (2005) regression is for the year 2015.
Our approach differs from RKRV (2005) in one respect. To estimate mispricing in year t, RKRV (2005) employ the
full time series of accounting information including data both prior to and subsequent to year t. Hertzel and Li (2010,

2 RKRV’s (2005) eq. 3 (our Equation (1)) refers to all accounting measures, including the book value of equity, as pertaining to year t even though they are
based on fiscal year-end data from year t – 1. In this equation, mit , the natural logarithm of the market value of firm i’s equity, is measured at the end of June of
year t. A similar labeling convention is used in RKRV’s (2005) eq. 15 (our Equation (2)).
JAFFE ET AL . 5

TA B L E 1 Sample descriptive statistics

Mispricing Growth options


Year ME MTB (FSE+TSSE) (LRVTB) N
1973 $353.0 1.74 −0.210 0.401 2,132
1974 $272.9 1.37 −0.343 0.314 2,306
1975 $298.2 1.58 −0.194 0.302 2,486
1976 $319.9 1.71 −0.152 0.292 2,564
1977 $314.2 1.67 −0.167 0.335 2,649
1978 $297.8 1.83 −0.114 0.355 2,797
1979 $325.9 1.97 −0.097 0.351 2,803
1980 $366.4 2.97 −0.054 0.368 2,825
1981 $405.6 3.84 0.108 0.462 3,102
1982 $350.7 1.96 −0.126 0.319 2,962
1983 $484.9 5.09 0.312 0.531 3,481
1984 $423.5 2.90 0.058 0.459 3,454
1985 $526.5 3.57 0.101 0.534 3,516
1986 $658.5 5.37 0.258 0.605 3,612
1987 $729.4 5.73 0.274 0.616 3,769
1988 $671.5 3.93 0.133 0.541 3,653
1989 $744.9 3.97 0.136 0.563 3,628
1990 $843.4 4.10 0.140 0.543 3,496
1991 $894.7 5.36 0.140 0.503 3,443
1992 $962.5 4.33 0.199 0.551 3,699
1993 $1,002.2 5.06 0.223 0.636 4,155
1994 $842.6 4.20 0.145 0.605 5,071
1995 $986.9 4.24 0.193 0.636 5,402
1996 $1,167.7 6.21 0.296 0.676 5,787
1997 $1,434.0 4.88 0.246 0.686 6,067
1998 $1,880.7 5.37 0.284 0.712 5,958
1999 $2,366.1 5.22 0.226 0.614 5,574
2000 $2,785.5 9.22 0.238 0.600 5,490
2001 $2,564.9 3.54 0.083 0.554 5,095
2002 $2,346.0 3.43 0.081 0.499 4,599
2003 $2,419.1 6.58 0.133 0.540 4,424
2004 $2,958.2 5.80 0.314 0.629 4,363
2005 $3,139.2 4.17 0.250 0.654 4,353
2006 $3,381.0 4.10 0.267 0.681 4,277
2007 $3,952.2 4.34 0.276 0.692 4,202
2008 $3,370.9 3.98 0.021 0.550 4,082
2009 $2,652.3 2.95 −0.098 0.429 3,715
2010 $3,127.0 3.57 0.086 0.461 3,606
2011 $4,139.2 3.82 0.164 0.540 3,507

(Continues)
6 JAFFE ET AL .

TA B L E 1 (Continued)

Mispricing Growth options


Year ME MTB (FSE+TSSE) (LRVTB) N
2012 $4,247.2 4.42 0.073 0.513 3,394
2013 $5,105.0 4.75 0.221 0.564 3,300
2014 $6,141.2 5.31 0.326 0.609 3,337
2015 $6,183.2 6.41 0.307 0.604 3,427

Note. This table presents annual averages of the sample characteristics. ME is market value of equity (in millions) measured
at the end of June of each year. Market-to-book (MTB) ratio is the ratio of the market value of equity to the book value of
equity. Book value of equity is measured at the fiscal year-end of year t – 1. Firm specific error (FSE), time series sector error
(TSSE), and long-run value to book (LRVTB) are based on the RKRV (2005) methodology. FSE, TSSE, and LRVTB for year t are
calculated using fiscal year-end accounting variables reported for year t – 1 and ME is calculated at the end of June of year t.

p. 942) caution that the original RKRV (2005) approach employs “forward-looking information that is not available to
investors at time t.”3 Consistent with Hertzel and Li (2010), we ignore future data when estimating mispricing.
The fourth column of Table 1 reports the average value of FSE + TSSE or mispricing for the different years of our
sample. As seen in Equation (1) above, RKRV (2005) express FSE + TSSE as a logarithm. Because the average values of
FSE + TSSE are negative in the early years, the average stock is undervalued, according to the RKRV (2005) model, in
these years. The average value of FSE + TSSE rises over time becoming positive in every year except one after 1982.
Thus, the average stock is overvalued according to RKRV (2005) in the later years. Since only contemporaneous data
are used to estimate value in the FSE calculation, the average value of FSE across the whole sample is zero in any year.
Thus, the increase in FSE + TSSE over time is due to an increase in TSSE, which likely reflects the rise of the various
valuation multiples in the US over time.
The fifth column of Table 1 reports the average value of LRVTB or growth options for each of the different years
in our sample. As with FSE + TSSE, RKRV (2005) express LRVTB as a logarithm. The average value for a given year
is always between zero and one implying that the fundamental value is, on average, above the book value. Variation
across the years appears small.
Table 2 examines the time series properties of the components of RKRV’s (2005) decomposition. In Panel A, all
stocks are ranked on MTB at the end of each June and placed into five quintile portfolios based on NYSE breakpoints.
The quintile with the lowest (highest) values for MTB is called Low (High). The panel reports the percentage of firms
remaining in a given quintile in future years. For example, the first number in the first line of the panel indicates that, on
average, 48.56% of the firms in the entire sample remain in the same quintile in any given 2-year period. If firms were
assigned randomly to the mispricing quintiles each year, we would expect that only 20% of the firms would remain in
the same quintile from one year to the next. The first column also indicates that the High quintile has the highest per-
centage of firms remaining in that quintile, 62.49%, from one year to the next. The Low quintile has the second highest
percentage at 60.75%. The remaining columns in Panel A provide the percentage of firms that remain in the same quin-
tile for a number of consecutive years. For example, the last column reports that 18.87% (22.32%) of firms in the Low
(High) quintile in a given year remain in the Low (High) quintile for each of the next four consecutive years. We view the
results of this panel as suggesting persistence in the MTB ratio.
The numbers are highest in the Low and High quintiles, although this may simply occur because these are the two
extreme quintiles. That is, for firms in Quintiles 2, 3, and 4, a large change in either direction of that firm’s MTB ratio
would likely move that firm to a different quintile. However, a large fall (rise) in MTB for a firm in the Low (High) quintile
should keep that firm in the same quintile, all else being equal.

3 RKRV (2005, p. 578) argue, “Thus, accepting the interpretation that this measure proxies for misvaluation does not require the reader to believe that assets

were mispriced in an asset pricing sense. It does not rest on the inability of market participants to make full use of available information. This measure could
proxy for knowledge held by the management that was unknown to the market at that time.”
JAFFE ET AL . 7

TA B L E 2 Analysis of MTB, mispricing, and growth options

Panel A: Quintile assignment based on MTB over time


Percent of sample in same quintile at the beginning of years:
0&1 0, 1, & 2 0, 1, 2, & 3 0, 1, 2, 3, & 4
Entire sample 48.56% 27.92% 17.64% 11.83%
1—low 60.75% 39.66% 27.04% 18.87%
2 38.53% 17.07% 8.06% 4.04%
3 33.90% 13.63% 6.01% 2.78%
4 37.58% 16.93% 8.48% 4.54%
5—high 62.49% 42.54% 30.37% 22.32%
Panel B: Quintile assignment based on mispricing over time
Percent of sample in same quintile at the beginning of years:
0&1 0, 1, & 2 0, 1, 2, & 3 0, 1, 2, 3, & 4
Entire sample 41.89% 22.24% 13.45% 8.78%
1—undervalued 58.15% 37.11% 25.02% 17.60%
2 31.96% 12.82% 5.97% 3.04%
3 29.12% 10.60% 4.54% 2.06%
4 30.78% 12.25% 5.72% 2.94%
5—overvalued 46.46% 26.61% 16.83% 11.29%
Panel C: Quintile assignment based on growth options over time
Percent of sample in same quintile at the beginning of years:
0&1 0, 1, & 2 0, 1, 2, & 3 0, 1, 2, 3, & 4
Entire sample 52.69% 32.92% 22.25% 15.79%
1—low 52.12% 31.78% 20.23% 13.41%
2 44.36% 24.16% 14.45% 9.04%
3 37.48% 17.59% 9.30% 5.25%
4 41.38% 20.35% 11.22% 6.73%
5—high 69.38% 50.72% 38.14% 29.26%

Note. This table reports the descriptive statistics for portfolios sorted on MTB, the mispricing component of MTB (FSE + TSSE),
and the growth options component of MTB (LRVTB) employing NYSE breakpoints. M/B, FSE, TSSE, and LRVTB are calculated
for year t using fiscal year-end accounting variables reported for year t – 1 and ME is calculated as of the end of June of year t.
The panels report the frequency of same quintile assignment. The portfolio sorting begins in June 1973 and ends in June 2015.

Panel B and C works with FSE + TSSE and LRVTB, respectively, with quintiles formed analogously to the quintile
formation in Panel A. In Panel B, the quintile with the largest negative (positive) values for FSE + TSSE is called Under-
valued (Overvalued) and, in Panel C, the quintile with the lowest (highest) values for LRVTB is called Low (High). The
results in Panels B and C are quite similar to those in Panel A for MTB. For example, the numbers in the first row in
Panel B (labeled the “Entire sample”) are slightly below those in the first row of Panel A, whereas the numbers in the
first row of Panel C are slightly above those in the first row of Panel A. Thus, we find similar persistence in all three
panels. And, as in Panel A, the numbers in Quintiles 1 and 5 for either Panel B or Panel C are always higher than those
in Quintiles 2, 3, and 4 for the same panel.
Table 3 reports the correlation coefficients between the terms in the RKRV (2005) model for our sample. Since
FSE + TSSE and LRVTB are expressed in natural logs, we report correlations based on the natural log of all of the
variables in Table 3. Ln(MTB) is positively correlated with both mispricing (FSE + TSSE) and growth options (LRVTB), an
unsurprising result as both mispricing and growth options are components of ln(MTB). Ln(ME) is positively correlated
8 JAFFE ET AL .

TA B L E 3 Correlations

(N = 165,562) ln(MTB) Mispricing Growth Options ln(ME)


ln(MTB) 1.000 0.659 0.696 0.259
(.000) (.000) (.000)
Mispricing 1.000 −0.082 0.408
(.000) (.000)
Growth options 1.000 −0.045
(.000)
ln(ME) 1.000

Note. This table presents the correlations between firm characteristics. Market-to-book (MTB), mispricing (FSE + TSSE), and
growth options (LRVTB) are calculated for year t using fiscal year-end accounting variables reported for year t – 1 and ME is
calculated as of the end of June of year t. ME is the market value of equity (in millions) measured at the end of June of each year.
The sample covers June 1973 to June 2015. p-Values are in parentheses.

with both ln(MTB) and mispricing. Again, these results are unsurprising as ME is the numerator of the MTB ratio
and mispricing is a component of ln(MTB). Finally, both mispricing and ln(ME) are negatively correlated with growth
options.

3 MISPRICING, GROWTH OPTIONS, AND THE CROSS-SECTION OF


STOCK RETURNS

Over the years, a great deal of research has concluded that stocks with low MTB ratios outperform stocks with high
MTB ratios. As previously mentioned, RKRV (2005) decompose the MTB ratio into FSE + TSSE and LRVTB. RKRV
(2005) assert that the first term is a measure of mispricing and the second term is a measure of a firm’s growth options.
However, their paper does not test whether either component is related to the cross-section of stock returns. In this
section, we determine whether the mispricing or the growth options component of market-to-book predicts future
returns, whether factors constructed from these MTB components contain incremental information relative to existing
asset pricing models, and whether these components help to explain the value premium.

3.1 Do the two market-to-book components predict future returns?


Table 4 sorts stocks into quintiles based on BTM (Panel A), the mispricing component of the market-to-book (Panel B),
and the growth options component of the market-to-book (Panel C).4 Value-weighted portfolios are formed for each
quintile at the end of each June using NYSE breakpoints and are held from July of the current year through June of the
following year. We work with all stocks on the CRSP with the necessary data to estimate BTM and the market-to-book
components. Our sample period is July 1973–June 2016, which corresponds to the start of our RKRV (2005) estimates
and the end of the available asset pricing model data. The first column of each panel reports the average value of the
sorting variable. The second column indicates the average value-weighted return on each quintile portfolio in excess of
the 1-month Treasury bill rate. The final three columns present the alphas of the quintile portfolios from three factor
models: (a) CAPM, (b) FF-5, and (c) M-4.5,6

4 As is standard in the asset pricing literature when analyzing the value premium, we report our quintile sorts by BTM, rather than by MTB, in Table 4, Panel A.

5 We report the results for FF-5 and M-4 as FF-5 includes a value premium factor (HML) and M-4 includes mispricing factors. As discussed later, the main
results of the analysis are robust to several other factor models.
6 The data for the M-4 model were obtained from Professor Stambaugh’s website: http://finance.wharton.upenn.edu/~stambaug/. FF-5 model data were
obtained from Professor French’s website: http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.
JAFFE ET AL . 9

TA B L E 4 Returns of portfolios sorted on BTM, Mispricing, and growth options

Panel A: Portfolios sorted on BTM


Factor model 𝜶’s:
Ave BTM (N) Excess return CAPM FF-5 M-4
1—value 1.70 0.75% 0.15% −0.15% −0.08%
(783) [1.12] [−1.81] [−0.82]
2 0.89 0.74% 0.24% −0.06% −0.01%
(670) [2.65] [−1.10] [−0.17]
3 0.66 0.71% 0.21% 0.00% 0.04%
(616) [3.11] [0.02] [0.58]
4 0.46 0.61% 0.08% −0.09% −0.04%
(642) [1.29] [−1.57] [−0.70]
5—growth 0.21 0.46% −0.13% −0.01% −0.03%
(950) [−2.19] [−0.35] [−0.59]
Value-growth 0.29% 0.28% −0.14% −0.06%
[1.64] [1.57] [−1.49] [−0.46]
Panel B: Portfolios sorted on mispricing
Factor model 𝜶’s:
FSE+TSSE (N) Excess return CAPM FF-5 M-4
1—undervalued −0.55 0.89% 0.33% 0.18% 0.21%
(1,132) [3.38] [2.84] [2.74]
2 −0.03 0.73% 0.23% −0.01% 0.00%
(687) [2.94] [−0.18] [−0.01]
3 0.20 0.70% 0.18% −0.03% −0.03%
(633) [2.85] [−0.58] [−0.57]
4 0.45 0.55% 0.03% −0.11% −0.09%
(586) [0.55] [−2.37] [−1.78]
5—overvalued 1.05 0.46% −0.15% −0.06% −0.03%
(622) [−3.11] [−1.53] [−0.69]
U–O 0.43% 0.47% 0.25% 0.24%
[3.28] [3.63] [2.78] [2.32]
Panel C: Portfolios sorted on growth options
Factor model 𝜶’s:
LRVTB (N) Excess return CAPM FF-5 M-4
1—low −0.49 0.61% 0.01% −0.18% −0.16%
(539) [0.07] [−2.71] [−2.08]
2 0.03 0.65% 0.13% −0.06% 0.04%
(628) [1.68] [−0.99] [0.62]
3 0.31 0.62% 0.11% 0.01% 0.04%
(567) [1.70] [0.12] [0.55]
4 0.61 0.60% 0.06% −0.06% −0.04%
(643) [1.06] [−1.09] [−0.61]

(Continues)
10 JAFFE ET AL .

TA B L E 4 (Continued)

Panel C: Portfolios sorted on growth options


Factor model 𝜶’s:
LRVTB (N) Excess return CAPM FF-5 M-4
5—high 1.26 0.51% −0.08% 0.08% 0.04%
(1,284) [−1.14] [1.70] [0.65]
Low–high 0.10% 0.08% −0.26% −0.20%
[0.67] [0.57] [−3.04] [−1.82]

Note. This table reports the value-weighted excess returns and alphas of portfolios sorted on BTM, the mispricing component
of MTB (FSE+TSSE), and the growth options component of MTB (LRVTB) employing NYSE breakpoints. BTM, FSE, TSSE, and
LRVTB are calculated for year t using fiscal year-end accounting variables reported for year t – 1 and ME calculated as of
the end of June of year t. Portfolios are formed at the end of June of each year from 1973 to 2015, and the returns for the
subsequent 12 months are used in the analysis. The following models are used to estimate alphas: CAPM, the FF-5 model, and
the M-4 model. The sample period is July 1973–June 2016. t-Statistics are in brackets.

The portfolios that sort by the mispricing component, FSE + TSSE, yield future excess returns similar to those
on portfolios sorted by BTM. The return difference between the top and bottom quintiles is 29 basis points
(t-value = 1.64) per month for BTM and 43 basis points (t-value = 3.28) per month for the mispricing component. In
contrast, the average excess return on the long–short portfolio for the growth options component, LRVTB, is only 10
basis points (t-value = 0.67) per month. Thus, based on the spread of excess returns, the mispricing component is a more
important determinant of the book-to-market’s ability to predict future returns than is the growth options component.
The alpha results yield further insights. The alpha of the “Value – Growth” portfolio for BTM is insignificant in all
three asset pricing models. In contrast, the alpha based on the FSE + TSSE sorts of the “Undervalued – Overvalued”
portfolio is significant in all three models.7 In the FF-5 model, which includes the value premium factor (HML), the
mispricing component generates an alpha of 25 basis points (t-value = 2.78) per month. The mispricing component
generates an alpha of 24 basis points (t-value = 2.32) per month in the M-4 model and covers 11 well-documented
anomalies. Thus, the mispricing component predicts future returns even after controlling for both the value premium
and a number of well-known anomalies.8
However, the mispricing in Panel B of Table 4 appears to be stronger for undervalued stocks than for overvalued
ones. The alphas for the Undervalued quintile are significantly positive for all three factor models. Conversely, although
the CAPM alpha is significantly negative for the Overvalued quintile, the alphas for this quintile are insignificantly dif-
ferent from zero for either the FF-5 model or the M-4 model.
Although the growth options component, LRVTB, does not generate a significant excess return spread, the alpha
results suggest it may be valuable to investors. Consistent with the excess return results, the CAPM alpha of the Low-
minus-High portfolio (Low – High) based on LRVTB sorts is indistinguishable from zero (8 basis points [t-value = 0.57]
per month). Nevertheless, the “Low–High” portfolio generates a statistically significant alpha of –26 basis points
(t-value = −3.04) per month in the FF-5 model and a marginally significant alpha of –20 basis points (t-value = −1.82)
per month in the M-4 model.
Taken together, the results on LRVTB from the FF-5 and M-4 models may have implications for the basic value versus
growth strategy. Prior research has shown that value stocks have beaten growth stocks historically. Our results on
RKRV’s (2005) measure of mispricing, FSE + TSSE, suggest that this mispricing measure is a strong contributor to the

7 In an untabulated analysis, we test the ability of the individual MTB components (FSE and TSSE) to predict future returns. Neither FSE alone nor TSSE alone

show much evidence of predictability. Rather, it is the combination of FSEs and TSSEs that allows investors to identify mispriced stocks.
8 As reported in the appendix, the alpha of the “Undervalued – Overvalued” portfolio is positive and significant, ranging between 23 basis points and 43 basis

points per month, in four alternative asset pricing models: the Q model of Hou, Xue, and Zhang (2017), the Fama-French four factor model, the conditional
market model of Petkova and Zhang (2005), and the six factor model of Barillas and Shanken (2018). The alpha is 27 basis points (t-value = 1.94) per month in
the leverage factor model of Adrian, Etula, and Muir (2014).
JAFFE ET AL . 11

performance differential of value over growth. However, our results on RKRV’s (2005) measure of growth options,
LRVTB, move in the opposite direction as stocks with high LRVTB may actually outperform stocks with low LRVTB. As
seen in Equation (1), high LRVTB contributes to high market-to-book. This result is borne out empirically as well as we
find that ln(MTB) is positively correlated with LRVTB. Thus, a trader buying value stocks and selling growth stocks may
be on the wrong side of the differential in alphas between high and low LRVTB stocks. In other words, investors may be
able to increase the performance of their portfolios by strategically tilting from stocks with low LRVTB to stocks with
high LRVTB.

3.2 Do the two market-to-book components contain incremental information?


Table 4 suggests that both the mispricing component and the growth options component may be valuable to an
investor. To measure their importance from an investor’s viewpoint, we create two factors, UMO and LRVB, from the
two components. We construct these factors following the methodology that Fama and French (1993) use to create
the standard HML factor. First, at the end of each June, we sort stocks into size categories (large and small) based
on the median NYSE size breakpoint. In addition, at the same time, we independently sort stocks by either mispric-
ing (FSE+TSSE) or growth options (LRVTB) into low (i.e., below the 30th NYSE percentile breakpoint), median (i.e.,
between the 30th and 70th NYSE percentile breakpoint), or high (i.e., above the 70th NYSE percentile breakpoint)
portfolios generating six value-weighted size-mispricing portfolios and six value-weighted size-growth options port-
folios. Moreover, we hold these value-weighted portfolios from July to the subsequent June. Then, we define each
factor as the difference between the average return on the two low portfolios and the average return on the two high
portfolios:

( )
UMOt = 0.5 × Small − Low FSE + TSSEt + Large − Low FSE + TSSEt
( )
− 0.5 × Small − High FSE + TSSEt + Large − High FSE + TSSEt , (3)

( )
LRVBt = 0.5 × Small − Low LRVTBt + Large − Low LRVTBt
( )
−0.5 × Small − High LRVTBt + Large − High LRVTBt . (4)

Table 5, Panel A reports the distribution of the UMO and LRVB factors, as well as those of the factors in the FF-5
and M-4 models. Of particular interest is the comparison between the market-to-book component factors and HML.
The average monthly return of UMO (35 basis points) is nearly the same as that of HML (38 basis points). Because the
standard deviation of UMO returns is lower than that of HML returns, the t-value of the UMO average return (3.57)
is slightly higher than that of the average HML return (3.00). In contrast, the LRVB factor does not generate a signifi-
cant monthly return (16 basis points per month, t-value = 1.36). These results are consistent with those based on the
portfolio excess returns in Table 4 because in either table RKRV’s (2005) mispricing component generates significant
positive returns, while its growth options component does not.
Table 5, Panel B reports the correlations between the factors. Both UMO and LRVB are highly correlated with HML.
The correlation between HML and UMO is 0.74, whereas the correlation between HML and LRVB is 0.89. Thus, even
though the LRVB factor does not earn a premium, the correlations suggest it is an important part of the standard value
factor. There is also a high correlation between the UMO and LRVB factors (correlation = 0.61).
To measure the incremental information that UMO and LRVB contain on expected returns relative to HML and
other existing factors, we run spanning regressions in Table 5, Panels C and D, respectively. Barillas and Shanken (2017)
argue that if an asset pricing model cannot price a factor excluded from that model, then a model augmented with that
factor is a better model. Several recent studies use spanning regressions to evaluate factors (Barillas & Shanken, 2018;
Gerakos & Linnainmaa, 2017; Hou, Mo, Xue, & Zhang, 2018). To gauge the marginal value of a new factor, we estimate a
12 JAFFE ET AL .

TA B L E 5 Analysis of mispricing and growth options factors

Panel A: Monthly percentage return


Factor
MKT SMB HML RMW CMA MGMT PERF UMO LRVB
Average monthly return 0.52 0.15 0.38 0.27 0.35 0.65 0.66 0.35 0.16
SD 4.57 3.10 2.92 2.36 1.98 2.83 3.86 2.25 2.73
t-value [2.63] [1.12] [3.00] [2.59] [4.08] [5.28] [3.90] [3.57] [1.36]
Panel B: Correlations between factors
MKT 1.00
SMB 0.26 1.00
HML −0.28 −0.22 1.00
RMW −0.26 −0.44 0.15 1.00
CMA −0.39 −0.12 0.69 0.04 1.00
MGMT −0.52 −0.39 0.72 0.28 0.77 1.00
PERF −0.26 −0.10 −0.30 0.43 −0.05 0.02 1.00
UMO −0.35 −0.05 0.74 0.08 0.55 0.57 −0.28 1.00
LRVB −0.14 −0.23 0.89 0.11 0.63 0.64 −0.47 0.61 1.00
Panel C: Spanning regressions. Dependent variable = UMO
Factor loadings
Asset pricing model dbMKT bSMB bHML bRMW bCMA bMGMT bPERF Adjusted R2
CAPM 0.43 −0.17 11.9%
[4.63] [−8.45]
FF-5 0.17 −0.09 0.11 0.55 −0.02 −0.01 59.6%
[2.57] [−5.98] [4.87] [18.24] [−0.75] [−0.16]
M-4 0.19 −0.10 0.16 0.41 −0.19 46.0%
[2.40] [−5.24] [6.14] [13.18] [−9.69]
Panle D. Spanning regressions. Dependent variable = LRVB
Factor loadings
Asset pricing model dbMKT bSMB bHML bRMW bCMA bMGMT bPERF Adjusted R2
CAPM 0.21 −0.09 1.8%
[1.74] [−3.28]
F-F-5 −0.22 0.09 −0.06 0.80 0.00 0.14 80.5%
[−4.04] [6.99] [−3.29] [31.42] [0.00] [3.62]
M-4 −0.12 0.06 0.03 0.69 −0.32 64.5%
[−1.46] [3.26] [0.99] [22.51] [−16.50]

Note. This table investigates the factors constructed from the mispricing component of MTB (UMO) and the growth options
component of MTB (LRVB). Panel A reports the distributions of UMO and LRVB, along with the distributions of the factors in
the FF-5 and the M-4 models. Panel B provides the correlations among UMO, LRVB, and the factors in the FF-5 and M-4 models.
Panels C and D present the results of spanning regressions in which the dependent variable is UMO and LRVB, respectively.
The independent variables of the spanning regressions are the factors from the CAPM, the FF-5, and the M-4 models. The
sample period is July 1973–June 2016. t-Statistics are in brackets.
JAFFE ET AL . 13

spanning regression by regressing that factor on the existing factors in an asset pricing model and test the significance
of the model’s alpha. A significant alpha indicates the factor includes additional information.
Panel C of Table 5 reports the results of spanning regressions where UMO is the dependent variable. We find that
the UMO factor contains information incremental to the factors in all three asset pricing models. The UMO alpha is
largest in the CAPM model (43 basis points, t = 4.63), an unsurprising result because only the factor loading on the
market is included. However, the UMO alpha in the FF-5 model, which includes the value factor (HML), is still positive
(17 basis points per month) and significant (t-value = 2.57). This means that an investor who already trades the value
factor, along with the market, size, profitability (RMW), and investment (CMA) factors, can benefit from the information
contained in UMO. In addition to UMO’s positive loading on HML, we find that UMO loads negatively on the market
factor and positively on the size factor. UMO appears unrelated to either the profitability or the investment factor.
The mispricing factors in the M-4 model capture 11 identified anomalies. Their MGMT factor is a composite of six
anomalies: (a) net stock issues, (b) composite equity issues, (c) accruals, (d) net operating assets, (e) asset growth, and
(f) investment to assets. Their PERF factor is a composite of five anomalies: (a) distress, (b) O-score, (c) momentum, (d)
gross profitability, and (e) return on assets. The results in Table 5, Panel C indicate that UMO is relevant to an investor
who trades the factors in the M-4 model. UMO loads positively on MGMT and negatively on PERF. The alpha from
this spanning regression is 19 basis points (t-value = 2.40) per month indicating that the M-4 anomaly factors do not
subsume UMO.9 The significant UMO alpha relative to the M-4 model indicates that UMO captures a unique dimension
of mispricing in the market above and beyond the 11 anomalies captured by the M-4 model.
Although UMO reflects a facet of mispricing distinct from those represented by the Stambaugh–Yuan (2017) fac-
tors, UMO is also related to the anomalies underlying their factors (as demonstrated by significant loadings on the
MGMT and PERF factors). We explore this relation in more detail in an untabulated analysis, where we estimate a
series of 11 regressions. Here, UMO is regressed on the market excess return, Stambaugh and Yuan’s (2017) SMB fac-
tor, and the long–short portfolio associated with one of the 11 anomalies included in the Stambaugh and Yuan mispric-
ing factors. For all six anomalies included in the MGMT factor, UMO’s beta on the anomaly spread portfolio is positive
and significant. For all five anomalies included in the PERF factor, UMO’s beta is negative and significant. These results
suggest that UMO is related to all 11 anomalies. Nevertheless, across the 11 regressions, the UMO alpha is positive
and significant in 10 of the models, ranging between 27 basis points and 57 basis points per month. Furthermore, the
UMO alpha remains positive and significant (22 basis points per month) when we include all 11 anomaly portfolios in
the regression. Thus, we conclude that UMO captures multiple forms of mispricing, some of which are correlated with
the Stambaugh–Yuan (2017) mispricing factors and some are not.
Panel D of Table 5 reports the spanning regression results where LRVB is the dependent variable. The LRVB alpha
in the CAPM model is positive (21 basis points) and marginally significant (t = 1.74). However, similar to the portfolio
alpha results in Panel C of Table 4, the LRVB alpha in the FF-5 model is negative (−22 basis points) and significant
(t-value = −4.04). In contrast, the LRVB alpha is insignificant in the M-4 model. Thus, unlike UMO, it does not appear
there is incremental information in LRVB (growth options) relative to the existing asset pricing models.

3.3 Do the two market-to-book components explain the value premium?


Our results thus far indicate that asset pricing models that include the traditional value factor (HML) cannot
fully explain the market-to-book mispricing premium. Alternatively, we now consider whether the mispricing factor
(UMO) explains the well-known value premium. We also test whether the market-to-book growth options factor
(LRVB) explains the value premium.

9As reported in the appendix, the UMO alpha remains positive and significant, ranging between 14 and 38 basis points per month, in five alternative asset
pricing models: the Q model of Hou et al. (2017), the leverage factor model of Adrian et al. (2014), the Fama-French four factor model, the conditional market
model of Petkova and Zhang (2005), and the six factor model of Barillas and Shanken (2018).
14 JAFFE ET AL .

TA B L E 6 The value premium and the components of MTB

Panel A: Fama-French book-to-market quintile portfolios


Factor model 𝜶’s:
Quintile Excess return CAPM MKT and SMB MKT, SMB, and UMO MKT, SMB, and LRVB
1—growth 0.48% −0.11% −0.10% 0.03% −0.02%
[−1.82] [−1.70] [0.66] [−0.50]
2 0.68% 0.13% 0.13% 0.08% 0.10%
[2.20] [2.33] [1.34] [1.77]
3 0.68% 0.18% 0.20% 0.07% 0.11%
[2.39] [2.64] [0.99] [1.93]
4 0.67% 0.16% 0.17% −0.12% 0.04%
[1.78] [1.81] [−1.86] [0.68]
5—value 0.91% 0.35% 0.33% 0.01% 0.17%
[3.00] [2.86] [0.12] [2.30]
Value–growth 0.43% 0.46% 0.43% −0.02% 0.19%
[2.72] [2.84] [2.71] [−0.19] [2.19]
Panel B: HML spanning tests
Factor loadings
Dependent variable bMKT bSMB bUMO bLRVB Adjusted R2
HML 0.48 −0.15 −0.15 9.8%
[3.98] [−5.62] [−3.60]
HML 0.06 0.02 −0.18 0.97 58.4%
[0.73] [0.92] [−6.49] [24.81]
HML 0.28 −0.11 0.03 0.93 81.0%
[5.06] [−8.45] [1.46] [44.34]

Note. This table reports the value-weighted excess returns and alphas of the Fama-French book-to-market quintile portfolios
in Panel A and the HML spanning test results in Panel B. MKT is the market factor, SMB is the size factor, UMO is the factor
constructed from the mispricing component of MTB, and LRVB is the factor constructed from the growth options component
of MTB. The sample period is July 1973–June 2016. t-Statistics are in brackets.

We start in Table 6, Panel A by examining the returns on the Fama-French book-to-market quintile portfolios.10 On
average, the “Value – Growth” portfolio earns a premium of 43 basis points (t-value = 2.72) per month over our sample
period. This premium is not explained by either the market factor or the size factor. The alpha of the “Value – Growth”
portfolio is 46 basis points (t-value = 2.84) per month in the CAPM model and 43 basis points (t-value = 2.71) per month
in a model that includes the market factor and SMB. Once we add the mispricing factor, however, the value premium
disappears. The “Value – Growth” portfolio alpha in a model that includes the market factor, SMB, and UMO is –2 basis
points (t-value = −0.19) per month. We also test whether the growth options factor explains the value premium. It does
not. In a model that includes the market, size, and growth option factors, the alpha of the “Value – Growth” portfolio
is 19 basis points (t-value = 2.19) per month. Taken together, the evidence in Table 6, Panel A indicates that it is the
mispricing component of MTB that explains the premium that value stocks earn over growth stocks.
We also test whether UMO and LRVB subsume the value factor. As previously reported in Table 5, Panel A, HML
has an average return of 38 basis points (t-value = 3.00) per month over our sample period. In Table 6, Panel B, we

10 The book-to-market quintile return data were obtained from Professor French’s website: http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_
library.html. These returns are based on the Fama-French universe of stocks and differ from those in Table 4, Panel A, which are based on the stocks in our
sample.
JAFFE ET AL . 15

report spanning regressions that identify how much of this return is explained by either UMO or LRVB, along with the
market and size factors. In each of the Table 6, Panel B regressions, the dependent variable is HML. If the alpha of a
particular model is significant, it indicates that HML includes additional information not contained in the factors used
on the right-hand side. We start with a simple model that includes the market and size factors. We find that HML loads
negatively on both of these factors. Consequently, HML’s alpha in the two-factor model (48 basis points per month)
is larger than HML’s average excess return. Moreover, the alpha is highly significant (t-value = 3.98). Thus, the market
and size factors do not explain the value factor.
Next, we regress HML on a model that includes UMO, as well as the market and size factors. The alpha in this model
is only six basis points (t-value = 0.73) per month implying that UMO subsumes HML. Given this insignificant alpha,
the highly significant loading of HML on UMO (0.97, t-value = 24.81), and the inability of the market and size factors to
explain HML, we conclude that the market-to-book mispricing factor explains the value factor.11
We also report the results from an HML spanning regression that includes LRVB, MKT, and SMB as the asset pricing
factors. Even though HML has a statistically significant loading on LRVB (0.93, t-value = 44.34), the model’s alpha is
positive (28 basis points per month) and significant (t-value = 5.06). Given HML’s average excess return of 38 basis
points per month, as reported in Panel A of Table 5, this alpha is economically significant. Thus, a three-factor model
that includes the market, size, and the market-to-book growth options component does not explain the HML factor
premium.
Overall, Table 6 provides consistent evidence that the premium that value stocks earn over growth stocks reflects
the correction of mispriced stocks. Once we control for the increase in the value of underpriced stocks and the decrease
in the value of overpriced stocks, there is no difference in returns between the value and the growth stocks. Thus, the
well-documented value premium appears to be a by-product of the mispricing effect.

3.4 How long do mispricing effects persist?


In Section 3.1, we presented evidence that the mispricing component of the RKRV (2005) model forecasts future 1-
year returns. To assess the speed at which the misvaluation is corrected, we examine the model’s long-run predictive
power.12 As before, we rank all of the stocks on FSE + TSSE each June and place them into five quintile portfolios based
on NYSE breakpoints. We measure returns for each of the next 4 years using an approach developed by Jegadeesh and
Titman (1993). Our first sort occurs in June 1973 and we follow the returns on the five quintiles formed from this sort
from July 1973 to June 1978. Our second sort occurs in June 1974 and we follow the returns on the quintiles from this
sort from July 1974 to June 1978. Thus, from July 1974 to June 1975, we have returns on two portfolios for each of
the five quintiles. For each of the five quintiles, the two portfolios are given equal weight in our return calculations over
the July 1974 to June 1975 period for that quintile. Beginning in June 1977, we have five portfolios for each of the five
quintiles. For each of the five quintiles, we combine the securities in the five different portfolios for that quintile at the
end of every June and follow the returns on that combined portfolio for the next 12 months.
To investigate the speed of adjustment, Table 7 examines alpha over various time periods following the ranking date.
For each of the factor models, the first five columns report alphas over each of the first 5 years following the ranking.
To conserve space, each panel only reports alphas for the Overvalued quintile, the Undervalued quintile, and the differ-
ence in alpha between the two extreme quintiles. Panel A presents alphas under the CAPM model. We focus on Years
2–5 because the alphas for the first year were already discussed above. For both the Undervalued quintile and the
“U – O” row, the alphas are significantly positive in each year from Year 2 to Year 5. In addition, the alphas for the Over-
valued quintile are significantly negative in Years 2 and 3. Thus, with the CAPM as the benchmark, the RKRV (2005)
model appears to predict returns for much, if not all, of the 5-year period.

11 In untabulated results, we find that our conclusion remains the same when we estimate the spanning regression with HML as the dependent variable and

UMO and the FF-5 model nonvalue factors (MKT, SMB, RMV, and CMA) as the independent variables.
12 Because the growth options component does not reliably predict future 1-year returns, we do not examine its long-run predictive power.
16 JAFFE ET AL .

TA B L E 7 𝛼’s from time-series regressions for years 1, 2, 3, 4, 5, 1–5, 2–5, and 6–10: Portfolios sorted on mispricing

Panel A: –𝜶’s based on CAPM model during:


Year 1 Year 2 Year 3 Year 4 Year 5 Years 1–5 Years 2–5 Years 6–10
1—undervalued 0.33% 0.36% 0.17% 0.22% 0.32% 0.29% 0.29% 0.09%
[3.38] [4.10] [2.03] [2.75] [4.00] [4.20] [4.08] [1.48]
5—overvalued −0.15% −0.14% −0.11% −0.07% −0.04% −0.10% −0.10% −0.03%
[−3.11] [−2.96] [−2.33] [−1.63] [−0.79] [−2.64] [−2.51] [−0.57]
U–O 0.47% 0.50% 0.27% 0.30% 0.36% 0.39% 0.39% 0.12%
[3.63] [4.09] [2.37] [2.65] [3.17] [3.95] [3.82] [1.20]
Panel B: –𝜶’s based on FF-5 model during:
Year 1 Year 2 Year 3 Year 4 Year 5 Years 1–5 Years 2–5 Years 6–10
1—undervalued 0.18% 0.21% 0.10% 0.16% 0.27% 0.22% 0.21% 0.06%
[2.84] [3.46] [1.56] [2.42] [3.95] [4.92] [4.20] [1.12]
5—overvalued −0.06% −0.06% −0.08% −0.09% −0.01% −0.05% −0.06% −0.05%
[−1.53] [−1.57] [−2.02] [−1.97] [−0.22] [−1.51] [−1.83] [−1.19]
U–O 0.25% 0.27% 0.19% 0.25% 0.28% 0.26% 0.28% 0.11%
[2.78] [3.26] [2.03] [2.66] [2.89] [4.07] [3.76] [1.30]
Panel C. - 𝜶’s based on M-4 model during:
Year 1 Year 2 Year 3 Year 4 Year 5 Years 1–5 Years 2–5 Years 6–10
1—undervalued 0.21% 0.10% 0.08% 0.14% 0.20% 0.18% 0.15% 0.05%
[2.74] [1.30] [1.15] [1.97] [2.72] [3.36] [2.60] [0.82]
5—overvalued −0.03% 0.01% −0.03% −0.05% 0.01% −0.02% −0.02% −0.01%
[−0.69] [0.33] [−0.74] [−1.16] [0.20] [−0.58] [−0.63] [−0.31]
U–O 0.24% 0.08% 0.11% 0.20% 0.19% 0.21% 0.18% 0.06%
[2.32] [0.81] [1.14] [1.94] [1.82] [2.55] [2.06] [0.68]

Note. This table reports the alphas for portfolios sorted on the mispricing component of MTB (FSE+TSSE) employing NYSE
breakpoints. Firms are sorted into portfolios at the end of June each year, and their returns are followed from one to ten years
in the future. The firm-specific error (FSE) and time series sector error (TSSE) are calculated for year t using fiscal year-end
accounting variables reported for year t – 1 and ME is calculated as of the end of June of year t. The following models are
used to estimate alphas: CAPM (Panel A), the FF-5 model (Panel B), and the M-4 model (Panel C). The sample period is July
1973–June 2016. t-Statistics are in brackets.

However, as reported in Panels B and C of Table 7, the results are somewhat weaker for the two multi-factor
models. Combining these two models, the four individual years from Year 2 to Year 5 generate eight alphas for each of
the three rows, yielding a total of 24 alphas overall. Of the 24 alphas reported for Years 2–5, four have t-values above
2 for the Undervalued quintile, one has a t-value below –2 for the Overvalued quintile, and four have t-values above 2
for the “U – O” row.
The next column shows alphas for Years 1–5 taken together. All of the alphas for both the Undervalued quintile and
the “U – O” row are significantly positive. However, the results are weaker for the Overvalued quintile with only the
CAPM alpha being significantly negative. Thus, as with the 1-year results, one can argue that the RKRV (2005) model
is better at identifying Undervalued firms than Overvalued firms.
Perhaps the biggest difference between the 5-year results and the 1-year results is the magnitude of the mispricing
that is revealed. Excluding the CAPM, the average alpha across the other two models for the “Undervalued – Over-
valued” portfolio is 0.235% [(=0.26% + 0.21%)/2] per month over the 5 years implying a cumulative alpha of 14.1%
(=0.235% × 60). Thus, assuming that the market takes 5 years to incorporate the information in RKRV’s (2005) model,
JAFFE ET AL . 17

the long–short strategy generates a substantial return as the mispricing is corrected. The estimate would be greater
still if the CAPM alpha was included.
To see if the RKRV (2005) model has predictive power after the first year, the next column reports alphas for Years
2–5 taken together. All three models have significantly positive alphas for both the Undervalued quintile and the “U
– O” row. However, only the CAPM provides a significantly negative t-value for the Overvalued quintile. We conclude
that the RKRV (2005) model has predictive power for alpha beyond the first year for the Undervalued quintile and for
the difference between alphas of the two extreme quintiles. However, the results from the two multi-factor models do
not suggest that the RKRV (2005) model has predictive power over Years 2–5 for the Overvalued quintile.
The next column examines alphas in Years 6–10 following the ranking. All alphas are insignificantly different from
zero in this column suggesting that the RKRV (2005) model has no predictive power for the entire sample beyond the
first 5 years.
Taken together, Table 7 provides evidence that the RKRV (2005) model forecasts returns over the 5-year period
following sorting. However, the model appears to have no predictive power beyond the first 5 years suggesting that
mispricing based on the RKRV (2005) model is corrected over time.

4 CONCLUSION

It remains an ongoing debate in the academic literature whether the value premium, the tendency for high BTM stocks
to outperform low BTM stocks, is a product of risk or behavioral phenomena. A key challenge in distinguishing between
these explanations is the identification of mispriced stocks. Our paper uses the model in RKRV (2005) to decompose
the MTB ratio into a mispricing component and a growth options component and asks three related questions. First,
do the mispricing and growth options components of MTB predict future returns? In addition, do factors constructed
from the mispricing and growth options components contain information incremental to existing asset pricing models?
Finally, do the mispricing and growth options components explain the value premium?
We examine the predictive power of the RKRV (2005) model by relating both the mispricing component and the
growth options component to future stock price performance employing three different factor models: (a) CAPM, (b)
the FF-5 model, and (c) the M-4 model. Our initial tests provide strong evidence that the mispricing component predicts
future returns. Long–short strategies that buy stocks in the Undervalued quintile and sell stocks in the Overvalued
quintile generate significantly positive alphas over the next year under each of the factor models. However, we do
not find strong evidence for the growth options component because the alpha of a portfolio, that is, long low growth
options and short high growth options, is only significant in one of the models.
Next, we explore whether factors constructed from the components of MTB contain incremental information rela-
tive to the existing asset pricing models. We employ the methodology that Fama and French (2015) use to form their
HML factor to create our mispricing (UMO) and growth options (LRVB) factors. Barillas and Shanken (2017) argue that
if an asset pricing model cannot price a factor excluded from that model, a better model can be created by including the
new factor. We separately regress UMO and LRVB on the factors of CAPM, the FF-5 model, and the M-4 model. The
alpha is significantly positive in each of the three regressions where UMO is the dependent variable implying that UMO
contains information incremental to each of the three asset pricing models. However, the evidence regarding whether
LRVB contains information incremental to the three asset pricing models is weaker. The CAPM alpha is marginally sig-
nificantly positive, the FF-5 alpha is significantly negative, and the M-4 alpha is insignificant.
We then test whether the two MTB components can explain the value factor. Although the standard “Value –
Growth” portfolio earns a significant alpha relative to a model that includes the market and size factors, this value pre-
mium completely disappears when we add the mispricing factor to the model. Similarly, based on the spanning regres-
sion framework, the alpha is significantly positive when we regress the HML factor on the market and size factors, but
becomes insignificant when we regress the HML factor on the market factor, the size factor, and the mispricing factor.
Thus, we conclude that RKRV’s (2005) mispricing component can explain the value premium.
18 JAFFE ET AL .

Conversely, we find no evidence that the growth options component of MTB can explain the value premium. The
alpha of the “Value – Growth” portfolio is positive and significant in a model that includes the market, size, and growth
options factors. The alpha from HML’s spanning regression on the market factor, the size factor, and the growth options
factor is also significantly positive.
We also provide evidence on the relation of RKRV (2005) mispricing to long-run returns. Our results here are similar
to those involving 1-year returns. We find that long–short strategies that buy underpriced stocks and sell overpriced
ones generate significantly positive alphas over the next 5 years under each of the three factor models that we use.
The magnitudes are high here. Excluding the CAPM, the average cumulative 5-year alpha for the “Undervalued – Over-
valued” portfolio is 14.1%. The alphas would be higher still with inclusion of the CAPM. However, although our results
imply predictive power over the first five years, we find no predictive power over Years 6–10. Thus, it appears that any
mispricing that the RKRV’s (2005) model identifies is corrected within 5 years for these samples.
Taken together, we conclude that the mispricing component of the RKRV (2005) model, FSE + TSSE, is generally
able to identify mispricing. In addition, we find that this component contains incremental information relative to exist-
ing asset pricing models. In contrast, we do not find that the BTM ratio contains incremental information relative to
mispricing, FSE + TSSE. In other words, we conclude that mispricing can explain the value premium, consistent with a
behavioral explanation of the premium.

ACKNOWLEDGMENTS

We would like to thank the anonymous referee and Bing Han (Editor) for detailed and insightful suggestions. J. Bao, M.
Cain, S.P. Kothari, and E. Smith provided helpful comments. The Securities and Exchange Commission, as a matter of
policy, disclaims responsibility for any private publication or statement by any of its employees. The views expressed
herein are those of the author and do not necessarily reflect the views of the Commission or of the author’s colleagues
upon the staff of the Commission. D.J. Pedersen acknowledges partial financial support from the David Whitcomb Cen-
ter for Research in Financial Services of Rutgers University. T. Voetmann acknowledges that the views expressed in this
article are solely those of the authors who are responsible for the content, and do not necessarily represent the views
of The Brattle Group.

REFERENCES

Adrian, T., Etula, E., & Muir, T. (2014). Financial intermediaries and the cross-section of asset returns. Journal of Finance, 69,
2557–2596.
Asness, C., & Frazzini, A. (2013). The devil in HML’s details. Journal of Portfolio Management, 39, 49–68.
Barillas, F., & Shanken, J. (2017). Which alpha? Review of Financial Studies, 30, 1316–1338.
Barillas, F., & Shanken, J. (2018). Comparing asset pricing models. Journal of Finance, 73, 715–754.
Daniel, K., & Titman, S. (2006). Market reactions to tangible and intangible information. Journal of Finance, 61, 1605–1643.
Fama, E. F., & French, K. (1992). The cross-section of expected stock returns. Journal of Finance, 47, 427–465.
Fama, E. F., & French, K. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial Economics, 33,
3–56.
Fama, E. F., & French, K. (2008). Dissecting anomalies. Journal of Finance, 63, 1653–1678.
Fama, E. F., & French, K. (2015). A five-factor asset pricing model. Journal of Financial Economics, 16, 1–22.
Fu, F., Lin, L., & Officer, M. (2013). Acquisitions driven by stock overvaluation: Are they good deals? Journal of Financial Eco-
nomics, 109, 24–39.
Gerakos, J., & Linnainmaa, J. T. (2017). Decomposing value. Review of Financial Studies, 31, 1825–1854.
Graham, B., & Dodd, D. (1934). Security analysis. New York, NY: McGraw Hill.
Griffin, J. M., & Lemmon, M. L. (2002). Book-to-market equity, distress risk, and stock returns. Journal of Finance, 57, 2317–
2336.
Hertzel, M., & Li, Z. (2010). Behavioral and rational explanations of stock price performance around SEOs: Evidence from a
decomposition of market-to-book ratios. Journal of Financial and Quantitative Analysis, 45, 935–958.
Hou, K., Mo, H., Xue, C., & Zhang, L. (2018). Which factors? Review of Finance, 23, 1–35.
Hou, K., Xue, C., & Zhang, L. (2017). Replicating anomalies (Working paper). Columbus, OH: The Ohio State University.
Jegadeesh, N., & Titman, S. (1993). Returns to buying winners and selling losers: Implications for market efficiency. Journal of
Finance, 48, 65–91.
JAFFE ET AL . 19

Lakonishok, J., Shleifer, A., & Vishny, R. W. (1994). Contrarian investment, extrapolation, and risk. Journal of Finance, 49, 1541–
1578.
Petkova, R., & Zhang, L. (2005). Is value riskier than growth? Journal of Financial Economics, 78, 187–202.
Rhodes-Kropf, M., Robinson, D., & Viswanathan, S. (2005). Valuation waves and merger activity: The empirical evidence. Journal
of Financial Economics, 77, 561–603.
Stambaugh, R. F., & Yuan, Y. (2017). Mispricing factors. Review of Financial Studies, 30, 1270–1315.

How to cite this article: Jaffe JF, Jindra J, Pedersen DJ, Voetmann T. Can mispricing explain the value premium?
Financial Management. 2019;1–19. https://doi.org/10.1111/fima.12272

APPENDIX
In this appendix, we investigate whether the mispricing effects reported in the portfolio analysis (Table 4) and spanning
tests (Table 5) are robust to the use of five alternative asset pricing models: (a) the Q model of Hou et al. (2017) (Q), (b)
the leverage factor model of Adrian, Etula, and Muir (2014) (AEM), (c) the Fama-French four factor model (FF-4), (d) the
conditional market model of Petkova and Zhang (2005) (PZ), and (e) the six factor model of Barillas and Shanken (2018)
(BS-6). Two of the models (Q and AEM), along with the three models (CAPM, FF-5, and M-4) used in the main analysis,
come from Gerakos and Linnainmaa (2017) who study the ability of existing asset pricing models to explain the value
premium.13 FF-4, which includes the market factor, SMB, HML, and the momentum factor (UMD), is included as an
alternative Fama-French specification. PZ includes two interaction terms: one between the market factor and an indi-
cator of low expected market risk premium and the other between the market factor and an indicator of high expected
market risk premium. We include this model to determine whether time-varying risk explains our results. Finally, we
include BS-6, which includes the market factor, the size factor, Q’s investment and profitability factors, the momentum
factor, and the monthly HML factor of Asness and Frazzini (2013), to test whether the mispricing component of MTB
contains additional information relative to an alternative HML factor.
The Table A1 reports the alphas of the “Undervalued – Overvalued” portfolio and spanning test alphas of the UMO
factor based on these five additional asset pricing models. In four of the models (Q, FF-4, PZ, and BS-6) in the portfolio
analysis and all five models in the spanning regression analysis, the mispricing alpha is positive and significant. The
alpha is marginally significant in the AEM model of the portfolio analysis. Taken together, these results, along with those
in our main analysis, suggest the mispricing effect is robust to controlling for tradable factors in several asset pricing
models.

TA B L E A1 Alternative asset pricing models

Portfolio alphas: “Undervalued – Overvalued” Spanning test alphas: Dependent variable = UMO
Q AEM FF4 PZ BS6 Q AEM FF4 PZ BS6
0.39 0.27 0.28 0.43 0.23 0.31 0.21 0.29 0.38 0.14
[3.93] [1.94] [3.33] [3.37] [2.50] [3.69] [2.00] [4.92] [4.28] [2.13]

Note. This table reports alphas for the “Undervalued – Overvalued” portfolio (see Table 4, Panel B) and the UMO spanning
regression (see Table 5, Panel C) based on five alternative factor models: the Q model of Hou et al. (2017) (Q), the leverage
factor model of Adrian et al. (2014) (AEM), the Fama-French four factor model (FF-4), the conditional market model of Petkova
and Zhang (2005) (PZ), and the six factor model of Barillas and Shanken (2018) (BS-6). The sample period is July 1973–June
2016. t-Statistics are in brackets.

13 Gerakos and Linnainmaa (2017) include in their analysis seven additional models that use nontradable factors. Because Barillas and Shanken (2017) argue
that the spanning test is only valid for asset pricing models with tradable factors, we restrict the robustness tests to those models in Gerakos and Linnainmaa
(2017) with tradable factors. Even though it includes a nontradable factor, the leverage factor model of Adrian et al. (2014) is included in our analysis because
the authors document how to create a factor-mimicking portfolio for their nontradable leverage factor.

You might also like