Professional Documents
Culture Documents
com
Decision Support
a
Russell Investments Japan Co., Ltd., Product & Research Division,Place Canada 7-3-37 Akasaka, Minato-ku, Tokyo 107-0052, Japan
b
Nikko Asset Management Co., Ltd., 1-1-3 Yurakucho, Chiyoda-ku, Tokyo 100-0006, Japan
c
School of Commerce, Senshu University, 2-1-1 Higashimita, Tama-ku, Kawasaki, Kanagawa 214-8580, Japan
Abstract
One of the typical issues in financial literature is that the market tends to be overly pessimistic about value stocks, many
of which are past losers. Therefore, over-reactions might capture by measuring earnings surprise vary with past return lev-
els. In this paper, we propose a new index for an effective investment strategy to capture the return-reversal effect using
both Data Envelopment Analysis (DEA) and Inverted DEA in order to consider the above characteristics of the market.
Our investment strategy using the new index exhibits better performance than the naive return-reversal strategy that only
uses past returns or earnings surprise. In addition, the correlations between our new index and commonly used value indi-
ces are insignificant, and the value indices cannot represent the over-valued (under-valued) situations perfectly. Hence,
considering both proposed and value indices like book-to-price one, we could select value stocks more effectively than
by using only one of these indices.
Ó 2007 Elsevier B.V. All rights reserved.
0377-2217/$ - see front matter Ó 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.ejor.2007.05.033
S. Kadoya et al. / European Journal of Operational Research 189 (2008) 120–131 121
stems from the above stock price pattern observed ian investment strategy is an efficient investment
in the Japanese market. method. Matsumura (1998) reports similar results
In this paper, we propose a new mispricing index for the Japanese market. Furthermore, Levis and
for stock investment using the concept of Data Liodakis (2001) also present similar results for the
Envelopment Analysis (DEA) and Inverted DEA UK market. One possible reason for the effectiveness
(IDEA) that has been proposed by Yamada et al. of the value strategy is some noise traders’1 cognitive
(1994), and attempt to select value stocks whose errors. Some noise traders believe that good stocks
future performances are considered to be high. are stocks of excellent companies. Such traders are
Then, we show that commonly used value indices too optimistic about growth stocks and too pessimis-
cannot represent over-valued or under-valued situa- tic about value stocks. On the other hand, informa-
tions of stocks perfectly, and our new index can pro- tion traders who can evaluate stock accurately, such
vide value-added information in this regard. as managers of pension funds, do not play the role of
DEA is a multi-criteria evaluating method that adjusting the cognitive errors of noise traders
can select the most favorable alternatives from through arbitrage because their clients are more for-
among large sets when there is no parametric giving of losses on stocks of good companies than
assumption among variables using mathematical losses on stocks of companies with poor fundamen-
programming technique. Therefore, we can deal tals. This phenomenon can be explained by the moti-
with multi-horizon past returns and several surprise vation of managers of pension funds in order to
indices that are harmonically defined in various show their prudent investments. The details of pru-
ways. DEA presents a composite score—referred dent investment effects on stock returns are dis-
to as efficiency—for each subject. This helps to sim- cussed by Shefrin and Statman (1995).
plify the complexity of analysis by evaluating the Fama and French (1992) show that individual
multi-criteria. In the field of financial investment, stock size and BP differences can explain return
Murthy et al. (1997) applies DEA to empirical study spreads among US stocks. They also propose a
of performance evaluation. However, to our best three-factor pricing model and insist on the exis-
knowledge, apart from performance evaluation tence of three common factors in the financial mar-
study, few proceedings have been undertaken on ket. One of the three factors is related to BP.
DEA application study in the field. Therefore, if their hypothesis is true, the value effect
This paper is organized as follows: In Section 2, will be consistent with the risk-return framework in
we review preceding studies on investment strategies modern finance study. However, Lakonishok et al.
and their applications in the field of operational (1994) also argue that the BP ratio may capture var-
research. Section 3 presents the idea of our mispric- ious undesirable factors. For example, a low BP
ing index and investment strategy. In Section 4, we may describe a company’s many intangible assets
outline the formulation of the DEA model and pro- that are not reflected in book values. Therefore,
pose a new mispricing index. Section 5 deals with we should consider various indices other than BP.
data description. Empirical experiments design is The central issue of our study is not to discuss the
presented in Section 6, and the results are shown relationship between the value factor and return-
in Section 7. Finally, Section 8 concludes this study. reversal phenomenon,2 but to select over-valued
stocks effectively. Therefore, our argument focuses
2. Preceding studies on portfolio selection techniques.
In this study, we propose a new method to
Recently, academic attention has been focused on capture the return–reversal effect using resulting
a particular anomaly in finance: the high perfor- evidence by Matsumura (1998). Our proposed
mance of value stocks. Lakonishok et al. (1994) method provides an effective contrarian investment
point out that many past losers (low performance
stocks) are future winners (high performance
1
stocks). Most of the past losers and winners are high A noise trader is defined as an investor who does not possess
BP stocks and low BP stocks, respectively. There- sufficient information to evaluate a stock accurately or trade a
fore, Lakonishok et al. (1994) insist that low BP is stock in accordance with his liquidity demand.
2
Watabe and Kobayashi (2001) discuss this problem. They
the result of the trade carried out by optimistic trad- conclude that value effect cannot be explained by the return-
ers. In other words, the return-reversal phenomenon reversal phenomenon and that the return-reversal effect cannot be
is observed in the financial market, and the contrar- negated by subtracting the value effect.
122 S. Kadoya et al. / European Journal of Operational Research 189 (2008) 120–131
consideration the co-linearity between indices. Meric uated as large as possible in the mathematical prob-
and Meric (2001) also indicates this feature. lem solving. Therefore, human judgment is excluded.
In order to expand Eq. (1), we introduce variable If a stock has a relative low efficiency score, that
weights u1, u2, . . . , ul for the historical returns. There- stock is considered as under-valued stock. Our pro-
fore, there are l kinds of return horizons in this case. posed score is conservative because the values of
We also introduce k kinds of valuation weights weights u1, u2, . . . , ul and v1, v2, . . . , vk are set to max-
v1, v2, . . . , vk for multiple surprise indices. Therefore, imize (3). When there are multi-criteria for stock
Eq. (1) can be expanded as follows: value and if even one of these criteria indicates that
, the corresponding stock is relatively over-valued, we
Xl X
k
DEA Idx ¼ uj ð1 þ rmj Þ vi Surmi ; ð3Þ would like to consider the stock as not relatively
j¼1 i¼1 under-valued. In this respect, DEA is a favorable
where m is an individual stock, i represents the system that provides a conservative estimation from
financial data (e.g., sales), and j is the term horizon. this perspective. However, there are two ways in
In addition, rmj represents stock m’s average which we value stocks: whether the stock is over-
monthly returns for the past j terms. Surmi is the sur- valued or whether it is under-valued. Problem (5)
prise index for stock m’s financial data i and is de- is only one aspect and the method is not adequate
fined as follows: to select over-valued stocks.
Therefore, we introduce the inverted version of
Actualmi Forecastmi
Surmi ¼ 1 þ ; ð4Þ Eq. (3), as follows:
Sizem ,
X k X l
Actualmi: actual financial data i (e.g., earnings, IDEA Idx ¼ vi Surmi uj ð1 þ rmj Þ: ð6Þ
sales, etc.) for stock m; i¼1 j¼1
Forecastmi: consensus forecasts of Actualmi; and This problem is referred to as Inverted DEA
Sizem: stock m’s market value. (IDEA).3 The securities having low IDEA scores
represent over-valued stocks in view of the conser-
DEA determines the optimal weights vj, ui to vative estimation because the output variables and
maximize Eq. (3) for each DMU (in this case, input variables replace each other. If we deal with
DMUs are stocks) under normalizing constraints. only one surprise index and historical return hori-
This is written as the following mathematical pro- zon in Eq. (6), the IDEA score will not be valuable
gramming problem: because it will represent only the inversion formula.
, However, we would like to evaluate stocks by multi-
Xl X
k
max uj ð1 þ rmj Þ vi Surmi ; criteria, which measure the most pessimistic under-
uj ;vi valued stocks as the highest value 1. Therefore,
j¼1 i¼1
, IDEA_Idx provides an evaluation from the view-
X
l X
k
s:t: uj ð1 þ rpj Þ vi Surpi 6 1; ð5Þ point of the inverse of DEA_Idx.
j¼1 i¼1 After all, these DEA and IDEA approaches to
ðp ¼ 1; . . . ; nÞ capture mispricing stocks are adequate for under-
u1 ; . . . ; ul ; v1 ; . . . ; vk P 0: valued stock selection and over-valued stock search,
respectively. Therefore, for the contrarian invest-
Here, n represents the number of stocks. Problem ment strategy, we employ a new mispricing index
(5) is intended to obtain the maximum efficiency that is defined as the difference between the DEA
value for stock m. The constraints in (5) work to and IDEA efficiency scores.
ensure that the efficiency score of any stock never We suggest the new mispricing index4 as follows:
exceeds 1. Therefore, even the most efficient stocks
Miss-pricing Index ¼ DEA Idx IDEA Idx: ð7Þ
can only have the optimal value of 1. The maxi-
mized object represents the possible high efficiency
3
value of each stock, while simultaneously adhering The concept of IDEA is explained in Appendix 1.
4
to the rules that restrict a stock from having an We can define new index as ratio (DEA_Idx/DEA_Idx).
efficiency that exceeds 1. However, both DEA_Idx and IDEA_Idx are standardized
between 0 and 1 and the difference defined in Eq. (7) is logical.
DEA is a data oriented evaluating technique. In this study, we also test the new index, defined as the ratio of
Then, human bias causes over-valuation in the both indices. Then, the tests yield similar results that have been
market and over-reaction in this study’s aspect is val- omitted from the paper for the reason of space.
124 S. Kadoya et al. / European Journal of Operational Research 189 (2008) 120–131
first test aims to check the validity of the contrarian belong to the top portfolio within the top category
investment strategy using our proposed index. The using the DEA model. Next, we evaluate the second
second test is performed to verify that our new index group using DEA. Evaluating the stocks that belong
is not significantly related to the BP index and that to the second group, we calculate the efficiency
our new misprice index provides value-added infor- within the second portfolio together with the top
mation for naı̈ve value stock portfolios that are con- portfolio. Finally, we evaluate the bottom portfolio
structed by using only the BP index. The third test stocks across all three-category stocks using DEA.
shows that we can construct an effective contrarian This method can give low BP stocks a handicap
investment strategy using both the BP index and DEA score for the evaluation. Therefore, low BP
our proposed index. The following is the test stocks are treated as over-valued stocks right from
methodology. the start of the evaluation. When we assign high
First, the stocks are sorted in ascending order by BP stocks a handicap score for the evaluation using
index (7) and are divided into quintile groups. The IDEA, we should sort out all the stocks based on BP
average returns of the quintile groups are computed in descending ranking and trisect all securities
and the spread return between the top and bottom depending on the order. Then, we applied the same
groups or between the second and fourth groups procedure for the evaluation. Hence, using both
are examined to ascertain the feasibility of the con- the DEA and IDEA scores, we can calculate our
trarian investment. Here, contrarian investment proposed score (Eq. (7)) as a handicapped index.
refers to the strategy in which the most over-valued All the stocks are sorted into quintiles based on this
stock portfolio (the bottom group by quintile) is handicapped index. Subsequently, we test the spread
sold and the most under-valued portfolio (the top return between the top and bottom portfolio returns.
group by quintile) is bought. Therefore, if the differ- If the spread return is more significant than both the
ence between the top group and bottom group per- spread returns non-handicapped proposed index
formances is significant, it can be said that the quintiles and the BP quintiles, then the proposed
contrarian investment strategy is valid. method can be said to provide additional value for
Next, we conduct two additional tests to discuss the investment strategy based on the BP index.
the advantages of our method. One of these tests is
an investigation of the two-stage sorted 25 equal- 7. Results
weighted asset portfolios’ performances as well as
the Fama and French (1993) size and book-to-mar- 7.1. Asymmetry of surprise affection
ket sorted portfolios. The first sorting is based on
BP. In this case, all the stocks are sorted into quin- In this subsection, we show the asymmetric
tiles. The second sorting is a conditional one within impacts of earnings and other financial data sur-
each BP quintile (first sorting) and is based on our prises on the future returns of the stocks.
proposed index. Subsequently, all the stocks are At first, we construct 25 equal-weighted asset
sorted into quintiles within the BP quintile groups. portfolios by two-stage sorting, as mentioned in Sec-
Therefore, 25 equal-weighted asset portfolios are tion 6. The first sorting is unconditional and based
formed. These portfolios are used to test whether on the magnitude of the surprise that is measured
the differences between the future returns of the port- on the relevant date. The second sorting is condi-
folios are significant. If our index provides value- tional, within each surprise quintile, and is based
added information for the existing index (BP), the on the past return levels. The future returns of the
future returns of the sorted layers will be signifi- 25 portfolios formed are shown below in Table 2.
cantly different. The second additional test is con- The numbers in the upper part of the row are the
ducted using DEA and IDEA, considering the returns and those in the lower part (in parentheses)
hierarchic category. The details of the methodolo- are t-statistics in the table. The figures represent
gies are shown by Cooper et al. (1999). First, we sort the monthly future return averages during the exper-
out all the stocks based on BP in ascending ranking. iment period. Here, we use a surprise measured by
Then, we trisect all the securities depending on the the current profit (Sur3) and the past return levels
magnitude of the BP. The top portfolio consists of measured by r1. Due to space constraints, we refrain
the lowest BP group stocks and the bottom portfolio from citing other results that were obtained using
consists of the highest BP group stocks. In order to other surprise measures and past return indices.
exclude the BP effect, we evaluate the stocks that Table 2 shows that the average returns decline with
126 S. Kadoya et al. / European Journal of Operational Research 189 (2008) 120–131
Table 2 the upper part of the row are the returns and those in
Future returns of cross sorting portfolios by Sur3 and r1 the lower part (in parentheses) are t-statistics.
r1 In this table, we also show the results using other
Top 2 3 4 Bottom quintile portfolios based on other indices (r1, r2, r3,
Sur3 Top 2.5% 2.2% 1.6% 1.2% 0.9% Sur1, Sur2, and Sur3). The ‘‘1–5’’ column in the table
(1.88) (2.02) (1.80) (1.44) (0.85) shows the return differences between the top and
2 2.2% 1.1% 0.9% 0.5% 0.2% bottom portfolios. The ‘‘2–4’’ column also repre-
(2.23) (1.35) (1.12) (0.66) (0.25) sents the spread return between the second and
3 1.4% 1.1% 0.5% 0.3% 1.0%
(1.61) (1.48) (0.58) (0.38) (1.07)
fourth quintiles. These numbers are monthly aver-
4 1.6% 0.9% 0.4% 0.6% 1.4% ages of the differences. The t-values show that the
(1.68) (1.16) (0.63) (0.70) (1.34) differences are statistically meaningful in most cases.
Bottom 2.2% 1.1% 0.9% 0.5% 0.2% However, the t-values of the spread returns based on
(1.26) (1.08) (1.08) (0.49) (0.55) the mispricing index are larger than the others. These
results show that our proposed index is more infor-
both r1 base quintile numbers and Sur3 base quintile mative and helpful for the contrarian investment
numbers, in increasing order. Hence, in order to strategy than other naive indices like past returns.8
select future high return stocks, we should consider
both past return levels and earnings surprise.
7.3. BP effect and mispricing index
7.2. Comparison of quintile portfolio performances
We show the two-stage sorting portfolio perfor-
Quintile groups are formed by using each effi- mances based on BP and the proposed index. The
ciency score on the relevant date for each year. results are shown in Table 4. Here, the numbers in
Therefore, the portfolio rebalances are performed the upper part of the rows are the returns and those
on a year by year basis. The quintile portfolio perfor- in the lower part (in parentheses) are t-statistics.
mances are also measured. The computation period In this table, ‘‘1–5’’ represents the spread returns
is from the day following the relevant date to the between the top and bottom portfolios. The t-statis-
next year’s reference date. The spread returns are tics show that the spread returns between the top
also calculated using monthly portfolio return aver- and bottom portfolios are significant within each
ages. The same procedure is applied to form quintile BP quintile, except the BP top portfolio. The future
portfolios based on surprise indices and past return return of the portfolio that belongs to both the BP
levels. These spreads are also computed. The spreads top quintile and the proposed index top quintile is
between the top and the bottom portfolios and those 2.1%, and the future return of the portfolio that
between the second and fourth portfolios are showed belongs to both the bottom quintiles is 0.6%. There-
in Fig. 1. Here, the spread returns are cumulative fore, the new index that provides value-added infor-
accounting numbers of monthly difference in mation for portfolio selection and the future return
returns. The lines in this figure show the stability of the contrarian investment strategy, using both
of the spread returns based on the proposed index. BP and the new index, should be higher than that
The statistical significances are also tested and the of the value stock selection strategy using only BP.9
results are presented in Table 3. Here, the numbers in Finally, we show the result of the test using DEA
and IDEA, taking into consideration the hierarchic
0.8
0.7
category. The cumulative spread return between the
0.6 top and bottom quintiles is represented in Fig. 2.
0.5 This figure shows an almost constant trend of
0.4
0.3
0.2 8
We conducted 18 tests using various combinations of past
0.1 returns and surprise indices. Essentially, t-statistics improved
0 with an increase in the number of inputs and outputs (see Table B
20000531 20010531 20020531 20030530 20040531
in Appendix 3).
Spread Return (proposing Index quitile,2—4) 9
Spread Return (proposing Index quitile,Top—Bottom) We also calculate rank correlations between our proposed
index and valuation indices, BP, EP, Cash-to-Price (CP) and
Fig. 1. Spread returns between quintiles based on proposed Sales-to-Price (SP). These numbers are enough low and the
index. averages absolute ones are about 0.20.
S. Kadoya et al. / European Journal of Operational Research 189 (2008) 120–131 127
Table 3
Comparing contrarian investment strategy performances
Top 2 3 4 Bottom 1–5 2–4
Proposed index 1.69% 1.23% 0.81% 0.48% 0.30% 1.39% 0.75%
(1.81) (1.65) (1.14) (0.68) (0.34) (3.90) (4.27)
Sur1 1.46% 0.70% 0.55% 0.64% 1.15% 0.31% 0.05%
(1.65) (0.94) (0.77) (0.87) (1.34) (1.94) (0.44)
Sur2 1.49% 0.88% 0.54% 0.50% 1.10% 0.39% 0.38%
(1.70) (1.17) (0.79) (0.68) (1.20) (2.36) (2.83)
Sur3 1.53% 0.82% 0.62% 0.53% 1.00% 0.53% 0.29%
(1.77) (1.10) (0.87) (0.70) (1.17) (3.31) (1.97)
r1 1.18% 1.15% 1.04% 0.72% 0.42% 0.76% 0.42%
(1.15) (1.52) (1.50) (1.05) (0.49) (1.45) (1.77)
r2 1.67% 1.31% 0.99% 0.52% 0.03% 1.63% 0.79%
(1.67) (1.76) (1.40) (0.70) (0.04) (2.81) (3.76)
r3 1.69% 1.30% 0.98% 0.61% 0.07% 1.76% 0.69%
(1.68) (1.67) (1.35) (0.83) (0.09) (2.49) (2.78)
Table 4
Future returns of two-step sorted portfolios
Proposed index
Top 2 3 4 Bottom 1–5
BP quintile Top 2.1% 1.7% 1.9% 1.4% 2.1% 0.1%
(2.0) (2.0) (2.2) (1.7) (2.0) (0.2)
2 1.8% 1.6% 1.4% 1.0% 0.8% 1.0%
(1.9) (2.0) (1.8) (1.4) (1.1) (2.6)
3 1.6% 1.1% 1.1% 0.7% 0.6% 1.0%
(1.7) (1.6) (1.5) (1.0) (0.7) (2.8)
4 1.0% 0.4% 0.2% 0.3% 0.2% 0.8%
(1.2) (0.7) (0.3) (0.4) (0.2) (2.1)
Bottom 1.0% 0.1% 0.3% 0.4% 0.6% 1.6%
(1.0) (0.1) (0.4) (0.4) (0.6) (2.2)
1
0.9
8. Concluding remarks
0.8
0.7
0.6 In this study, we drew a comparison between past
0.5 returns and the surprise indices of business results
0.4 and proposed a new mispricing index. The results
0.3
0.2
revealed that our new index is more helpful in for-
0.1 mulating a contrarian investment strategy than
0 naive indices such as past returns and surprise indi-
20000531 20010531 20020531 20030530 20040531
ces. However, the new index is uncorrelated with
Fig. 2. Spread returns between quintiles based on the handi- other commonly used value indices like BP. There-
capped proposed index. fore, we can select value stocks using both the new
128 S. Kadoya et al. / European Journal of Operational Research 189 (2008) 120–131
129
1146 Min 0.023 0.883 0.773 0.926 0.949 0.980 0.000 0.000 0.000 0.000 0.000 0.000 0.853
130
r1 r2 r3 Sur1 Sur2 Sur3 v1 v2 v3 u1 u2 u3 IDEA_Idx
IDEA_Idx
2000 942 Mean 1.003 0.998 1.003 1.000 1.006 1.004 0.123 0.336 0.546 0.000 0.051 0.726 0.785
942 Std 0.038 0.019 0.014 0.219 0.038 0.033 0.335 0.477 0.494 0.007 0.191 0.195 0.031
942 Max 1.227 1.136 1.091 4.627 1.301 1.268 1.094 1.049 1.025 0.216 0.803 0.829 1.000
942 Min 0.914 0.953 0.974 0.013 0.675 0.696 0.000 0.000 0.000 0.000 0.000 0.000 0.558
2001 1067 Mean 1.010 1.010 0.998 1.013 1.000 1.004 0.529 0.256 0.220 0.077 0.378 0.438 0.898
1067 Std 0.026 0.016 0.012 0.130 0.022 0.016 0.491 0.430 0.407 0.030 0.030 0.039 0.022
33, 3–56.
Lakonishok, J., Shleifer, A., Vinshny, R.W., 1994. Contrarian
1–5 (%)
Meric, G., Meric, I., 2001. Risk and return in the world’s major
stock markets. Journal of Investing 10 (1), 63–66.
Quintile portfolio performance
Shefrin, H., Statman, M., 1995. Making sense of beta, size and
1.47
1.51
1.55
1.34
1.64
1.66
1.38
1.75
1.70
1.54
1.53
1.62
1.66
1.68
1.82
1.60
1.71
1.77
1.69
r3
r3
r3
r3
r3
r3
r3
r3
r3
r2
r2
r2
r2
r2
r2
r2
r2
r2
Japanese).
r1
r1
r1
r1
r1
r1
r1
r1
r1
r1
Sur3
Sur3
Sur3
Sur3
Sur3
Sur3
Sur3
Sur3
Sur3
Sur3
DEA input (IDEA output)
Sur2
Sur2
Sur2
Sur2
Sur2
Sur2
Sur2
Sur2
Sur2
Sur2
Sur1
Sur1
Sur1
Sur1
Sur1
Sur1
Sur1
Sur1
Sur1
Sur1
Table B
Model
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19