Il Liquidity and Stock Return 7

Illiquidity, Illiquidity Risk and Stock Returns: Evidence from Japan
By

Jing Fang
Department of Financial Engineering
Haitong Securities Co., Ltd.
Shanghai, 200021, China
Email: fangjing@htsec.com

Qian Sun*
Institute for Financial & Accounting Studies
Xiamen University
Xiamen, 361005, China
Email: qsun@xmu.edu.cn

Changyun Wang
School of Finance
Renmin University of China
Beijing 100872
Email: wangcy@sfruc.edu.cn

First Draft: January 2005
This Draft: January 2006

Keywords: Illiquidity; Amihud Ratio; Liquidity-adjusted CAPM; Japan

JEL classification: G12

* Corresponding author: Institute for Finance & Accounting Studies, Xiamen University,
Xiamen, 361005, China; Tel: (86-592) 2185995, Fax: (86-592) 2181787; Email:
qsun@xmu.edu.cn
1

Abstract
This paper extends the illiquidity and stock return studies conducted by Amihud (2002)
and Acharya and Pedersen (2005) to the Japanese stock market. Our comprehensive study across
firms and over time indicates that illiquidity, as measured by the Amihud ratio, is not as
consistently priced in Japan as it is in the US, especially for the period between 1990 and 1999.
The liquidity-adjusted CAPM proposed by Acharya and Pedersen does not explain Japanese
stock returns better than the standard CAPM. Our results also show that the liquidity effect
captured in the Amihud ratio may be sensitive to market states.

Keywords: Illiquidity; Amihud Ratio; Liquidity-adjusted CAPM; Up market; Down market

JEL classification: G12
2

I. Introduction
Liquidity or illiquidity is of concern because it has important practical as well as
academic implications. Using the new illiquidity measure proposed by Amihud (2002), we
examine the relationship between illiquidity and stock returns in the Japanese market.
According to Amihud and Mendelson (1980) and Amihud (2002), illiquidity reflects the
impact of order flow on price. Since illiquidity is not observed directly but rather has a number
of aspects that cannot be captured in a single measure, various proxies for illiquidity have been
used in previous studies. Some easily obtained proxies are turnover, trading volume or value,
firm size, etc; however, as pointed out by Lesmond (2002), these proxies may capture the effect
of variables not related to liquidity. On the other hand, some finer and more accurate measures
based on market microstructure data, such as bid-ask spread, amortized effective bid-ask spread,
price response to signed order flow and probability of information-based trading (PIN), are not
generally available, especially over a long period of time.
Although there is no perfect measure of liquidity, a simple and intuitive measure aiming
to balance the limits of data availability and accuracy has been developed in the recent work of
Amihud (2002). This measure only requires the input of daily data to construct and is applicable
to all securities and time periods.
1

The measure proposed by Amihud (2002) is the daily ratio of absolute stock return to its
dollar trading volume averaged over a given period (Amihud ratio hereafter). Intuitively, this can
be interpreted as the daily stock price response associated with one dollar of trading volume.
This is consistent with Kyles (1985) concept of illiquidity, i.e. the response of price to order
flow, and Silbers (1975) thinness measure, i.e. the ratio of absolute price change to absolute
excess demand for trading. After comparing a few alternative liquidity measures, Hasbrouck
(2002) concludes that the Amihud ratio appears the best.
With this new measure, Amihud (2002) examines the relationship between illiquidity and
stock returns and finds that illiquidity not only affects stock returns cross-sectionally but also
over time. Many studies have documented that illiquidity can explain differences in the expected

1
Pastor and Stambaugh (2003) use daily data to construct a liquidity measure based on signed order flow. Lesmond
(2002) developed a liquidity estimate based on the percentage of zero return trading days over a certain period, such
as a year.
3
returns across stocksfor example, Amihud and Mendelson (1986), Glosten and Harris (1988),
Hasbrouck (1991), Brennan and Subrahmanyam (1996), Eleswarapu (1997),Chalmers and
Kadlec (1998), Easley et al. (2002), Pastor and Stambaugh (2003), among others. Using mainly
market microstructure data from the US and various estimation techniques, these authors report a
positive relationship between illiquidity and stock returns across companies.
Few studies, however, have examined the illiquidity and stock return relationship over
time.
2
As pointed out by Amihud (2002), this is probably due to the fact that illiquidity measures
based on microstructure data for long time periods are not available in most markets around the
world. In contrast, the Amihud ratio only uses daily data, which is available for most markets
over long time periods. With this new measure, Amihud (2002) postulates and tests the
hypothesis that over time, the ex ante stock excess return increases in expected illiquidity while
unexpected illiquidity lowers the contemporary stock return. His empirical results are consistent
with this hypothesis.
The Amihud ratio is also employed by Acharya and Pedersen (2005) in testing their
liquidity-adjusted CAPM. They propose that not only the level of illiquidity but also the
uncertainty of illiquidity (illiquidity risk) should be priced. Their results indicate that illiquidity
betas are priced and liquidity-adjusted CAPM is better than the standard CAPM in terms of
goodness-of-fit.
However, the Amihud ratio has not been employed to test the relationship between
illiquidity and stock returns outside the U.S. Yet, if the Amihud ratio is an effective measure of
illiquidity, if illiquidity does have a general impact on stock returns across firms and over time,
and if the liquidity-adjusted CAPM is generally applicable, then the results obtained in Amihud
(2002) and Acharya and Pedersen (2005) should be replicable using data outside the U.S.
We extend the study of illiquidity and stock returns using the Amihud ratio to the
Japanese market, the second largest stock market in the world and next only to the U.S. market in
terms of both capitalization and number of securities. As pointed out in Chan, Hamao and
Lakonishok (1991), the study of the Japanese stock market alongside that of the U.S. is of
importance to the evaluation and comparison of empirical models of the cross-sectional stock
returns. The confirmation of the same determinants in these two countries would strengthen

2
Bekaert, Harvey, and Lundblad (2003) use pooled data to study the relationship between liquidity and expected
returns across firms over time in emerging markets.
4
confidence in the evidence found in the U.S. market, while the distinctiveness of the
determinants would induce further exploration of asset pricing theories. In addition, evidence
from the Japanese market may shed further light on the subsumption of explanatory variables
and robustness with regard to time period and sample selection.
Chan, Hamao and Lakonishok (1991) relate cross-sectional differences in returns on
Japanese stocks to the underlying behavior of earnings yield, size, book-to-market ratio, and cash
flow yield. They uncover a significant relationship between these variables and expected returns
in the Japanese market, which is largely consistent with findings in the U.S. Using market
microstructure data from Japan, Lehmann and Modest (1994) offer a birds eye view into trading
and liquidity on the Tokyo Stock Exchange (TSE) and compared it with that on the NYSE. Hu
(1997) finds a negative relationship between turnover and expected returns of the TSE stocks.
Bremer and Hiraki (1999) find evidence linking short-term returns of individual TSE stocks and
lagged trading volume, which is consistent with the results found in the U.S. stock market.
Hodoshima et al. (2000) examine cross-sectional return and beta in Japan. Hamori (2001) studies
seasonality and stock returns in Japan. Using market microstructure data, Ahn, Cai, Hamao and
Ho (2002) find that stock returns are positively related to illiquidity measures. However, they
also find a number of results that are different from those found on the NYSE.
We examine (1) if the Amihud ratio is correlated with other readily available, traditional
liquidity or illiquidity proxies in Japan; (2) if the Amihud ratio is positively related to stock
returns across companies listed on the TSE; (3) if expected (or unexpected) illiquidity is
positively (or negatively) related to expected (contemporaneous) stock returns over time; and (4)
if liquidity-adjusted CAPM performs better than the standard CAPM in Japan.
Our findings are mixed. While the cross-sectional relationship between illiquidity and
stock returns in the Japanese stock market is consistent with that generally found by Amihud in
the U.S., this is not consistent across all sub-sample periods, especially the second sub-sample
period between 1990 and 1999. While we do find that unexpected illiquidity has a negative
impact on contemporaneous stock returns, we do not find positive impact of expected illiquidity
on expected stock returns. Moreover, we fail to find evidence that the liquidity-adjusted CAPM
performs better than the standard CAPM.
Our study also takes into consideration factors unique to the Japanese market in
constructing our test design and examining alternative model specifications. In addition, we
5
examine if the return-illiquidity relationship is sensitive to market states or the up and down
markets.
The rest of the paper is organized as follows. The next section constructs our illiquidity
measure and relates it to some other traditional ones. Section III examines the cross-sectional
relationship between the Amihud ratio and stock returns. Section IV looks at the time series
effect of illiquidity on stock returns. Section V studies if liquidity-adjusted CAPM holds better
than the standard CAPM and Section VI concludes.

II. Amihud Ratio
All data used in this study has been obtained or computed from the PACAP Japan
Database. The daily stock illiquidity
i
d
ILL (the Amihud ratio based on Amihud (2002)) is
computed as the ratio of absolute daily return to daily trading value.

i
d
i
d
i
d
VAL R ILL / | | = , (1)

where
i
d
R is the return for stock i on day d.
i
d
VAL is the trading value for stock i on day d in
millions of yen, and
i
d
ILL represents the absolute percentage price change per million yen of
trading value.
Our sample period extends from 1975 to 2000. Following Lehmann and Modest (1994)
and Bremer and Hiraki (1999), only the first section stocks in the TSE are included in our study
because the stocks in different sections satisfy different listing criteria and are likely to have very
different trading and liquidity characteristics. For example, the first section stocks are much
larger and much more actively traded than those in the second section. Since most Japanese firms
use March as their fiscal year-end and financial reports may not be available until June, we use
daily stock returns and trading values from July 1
st
in the current year till June 30
th
in the
following year to compute the annual illiquidity. For example, the annual stock illiquidity in
1975 is averaged from July 1
st
, 1975 to June 30
th
, 1976, so on and so forth. The annual
illiquidity is

i
yd
i
yd
D
d
iy
i
y
VAL R D ILL
iy
/ | | / 1
1
=
= , (2)
6
where D
iy
is the number of trading days in year y. Similarly, the monthly illiquidity is

i
md
i
md
D
d
im
i
m
VAL R D ILL
im
/ | | / 1
1
=
= . (3)

Following Amihud, a stock admitted to our sample must meet the following criteria: 1) it
must have valid observations of daily return and trading value for more than 200 days in year y
so that the illiquidity estimate is more reliable;
3
2) the year-end stock price must be greater than
100 so that stock returns are not affected too much by the minimum tick size of 1
4
. Amihud
(2002) confines his sample to stocks with a year-end price greater than $5 to reduce the possible
estimation noise caused by minimum tick size. Further eliminating outliers with annual
illiquidity at the highest and lowest 1 percent of the distribution results in our final sample,
ranging from 565 firms in 1975 to 1099 firms in 1999, as presented in Table 1. Since
independent variables are lagged one-year behind the dependent variable in our model
specification, the sample period employed in our regression analysis is from 1976 to 1999.
[Insert Table 1 here]
We further relate the Amihud ratio to three traditional liquidity proxies via cross-section
regressions employed year by year from 1976 to 1999. The three proxies are market
capitalization, trading value, and turnover. The results are evident that the annual Amihud ratio is
strongly and negatively related to all three traditional liquidity measures
5
. This is consistent with
common sense: the larger the firm size, the larger the trading value or the higher the trade
turnover, and the less illiquid the stock is.
The annual market illiquidity is the average illiquidity across stocks in market portfolio
M in year y
=
M
i
y
M
y
ILL ILL . (4)

3
We have tried to include all firms with valid observations for more than 150 days in year y, the results are
qualitatively the same.
4
In addition to tick size, the TSE also has price limit rules. There are both the limit for maximum daily price change
and the limit for maximum price change between trades. However, the limit between trades is not relevant because
our illiquidity measure uses daily data. The daily price limit is quite large, ranging from 10% to 30%. They are
rarely hit in reality. Therefore, we do not particularly consider them in our sample selection process. The
information of tick size schedule and price limit rules for TSE stocks as of 2004 is available from the authors.
5
The results are not reported to save space but are available upon request from the authors.
7
[Insert Figure 1 here]
Panel A of Figure 1 presents the annual market illiquidity over the period 1975-1999. It
appears that market illiquidity is declining from 1975 to 1990 and inclining after that.
Correspondingly, as shown in Panel B of Figure 1, the Nikkei 225 Index has an upward trend
from 1975 to 1989, followed by a downward trend from 1990 to 1992 and then oscillates
thereafter. According to Securities Market in Japan (2001), a publication of the Japan Securities
Research Institute, the development of the Japanese securities market from 1975 to 1999 can be
divided into several stages: (1) 1975-1984 is the period of coping with the oil crisis; (2) 1985-
1989 is the period of the economic bubble; (3) 1990-1999 is the period of financial reform
involving debate on, and enforcement of, the Financial System Reform Law. Roughly, the first
two periods coincide with the rapid development of the stock market and a declining trend in
illiquidity, while the last one is associated with market slowdown and an increasing trend
towards illiquidity. Therefore, when we divide the whole sample period into two subsample
periods in the subsequent tests, the first one is from 1976 to 1989 and the second one is from
1990 to 1999.

III. Cross-Sectional Relationship between Illiquidity and Stock Returns
For comparison purposes, we first follow Amihud (2002) to estimate the following Fama-
MacBeth type cross-sectional regression model for each month during our sample period, where
monthly stock return, R
i
m
is a function of illiquidity and a set of control variables,

i
y j
X
1 ,

i
n
j
i
y j jy
i
y y y
i
m
X k ILLM k k R + + + =

=

2
1 , 1 1 0
. (5)
With one year lag of all the right-hand variables, our monthly return sample runs from
July 1976 to June 2000, a total of 288 months, while the yearly independent variables run from
1975 to 1998 (a year begins every July and ends in June). Since the annual stock illiquidity
varies dramatically over time, following Amihud (2002), the illiquidity variable is further scaled
by market illiquidity for stock i in year y to obtain the mean-adjusted illiquidity
6

/
i i M
y y y
ILLM ILL ILL = . (6)

6
We have tried illiquidity measures without such adjustment, the results are largely the same. For the sake of
consistency with Amihud (2002), we report the results with the adjustment.
8
Other stock characteristics or control variables included in the regression are: (1) firm size,
Ln
i
y
CAP
1
, which is the logarithm market capitalization for stock i at the end of year y-1; (2) beta,
1i
y-1
, which is the beta estimated in year y-1; (3) total risk,
i
y
STD
1
, which is the standard
deviation of the daily return on stock i in year y-1 (multiplied by 10
2
); (4) dividend yield,
i
y
DP
1
,
which is the sum of the dividends during year y-1 divided by the end of y-1 price; (5) past stock
returns, which include
i
y
PR
1
1
, the return on stock i during the last 100 days before the year end
of y-1 (June 30 every year) and
i
y
PR
2
1
, which is the return on stock i over the rest of the period,
between the beginning of the year y-1 (July 1) and 100 days before its end.
Ln
i
y
CAP
1
is used as a control for the well-known size effect. However, as mentioned
earlier, size may also be a proxy for liquidity. In Amihud (2002), this correlation is -0.614. In our
case, it is -0.581, as shown in Table 2.
i
y
STD
1
is included since investors portfolios may not be
well diversified.
i
y
DP
1
has been documented as an important determinant of stock returns in the
U.S. Previous stock returns are included to control for possible momentum effects (see Brennan
et al., 1998) and
1i
y-1
is used as a control for the market or systematic risk.
The beta is estimated using the Fama-French (1992) methodology. In June of each year,
stocks are ranked by their market capitalization and sorted into 25 portfolios. The market model
is estimated using daily data for the year of each portfolio with the Scholes and Williams (1977)
adjustment used to obtain the portfolio beta. This portfolio beta is then assigned to each
individual stock in the portfolio as its beta risk for that year.
We further put forward two additional control variables: cash flow yield (
i
y
CP
1
), which is
the ratio of earnings plus depreciation per share in year y-1 versus the year end share price for
stock i, and book-to-market ratio (
i
y
BM
1
), which is the ratio of the book value to market value of
equity for stock i at the end of year y-1. Amihud (2002) does not include
i
y
BM
1
in his study
because Easley et al. (2002) and Loughran (1997) find it has no effect on NYSE stocks. However,
Chan et al. (1991) report that cash flow yield and book to market ratio are the two variables with
most significant (positive) impact on expected returns in Japan. Also, the cash flow yield may be
9
a better alternative than dividend yields in Japan because the latter are minuscule for Japanese
firms.
7

Table 2 presents the summary statistics for all the variables put forward above and the
correlation matrix between these variables. In each year, the annual mean, standard deviation
across stocks, skewness, median, minimum and maximum are calculated for sample stocks and
then these annual statistics are averaged over 24 years. Similarly, the correlations between the
variables are calculated each year across stocks and then the yearly correlation coefficients are
averaged over the years. Notice that the correlations are generally low except for the one
between illiquidity and size, which is -0.581.
To facilitate the comparison, cross-sectional regression results presented in Table 3 are
formulated after Table 2 in Amihud (2002). The first four columns in the table report the results
for the four-variable model:

i
y y
i
y y
i
y y
i
y y y
i
m
PR k PR k k ILLM k k R
2
1 4
1
1 3
1
1 2 1 1 0
+ + + + = (7)

while the last four columns report for the seven-variable model:

i
y y
i
y y
i
y y
i
y y
i
y y
i
y y
i
y y y
i
m
DP k STD k CAP k PR k PR k k ILLM k k R
1 7 1 6 1 5
2
1 4
1
1 3
1
1 2 1 1 0
ln

+ + + + + + + = (8)

The mean of the 288 estimated coefficients is calculated for each independent variable, followed
by a t-test conducted on the hypothesis that states that the mean should be equal to zero (columns
1 and 5). To control for the famous January effect
8
and to verify whether the cross-sectional
relationship is stable over time, tests are also performed for the sample, excluding January (264
months; columns 2 and 6), and for the sub-samples 1976-1989 (168 months, columns 3 and 7)
and 1990-1999 (120 months, columns 4 and 8).

7
Earnings yield may be distorted because only accelerated depreciation are allowed in financial reporting for tax
purposes. The cash flow yield can avoid the earnings distortion from firms with large capital investments.
Correspondingly, Chan et al. (1991) find that the earnings yield is not a significant determinant of stock returns in
Japan.
8
Hamori (2001) documents the January effect in Japan for the entire period between 1971 and 1997 although it
tends to disappear in the later part of the sample period.
10
For equation (7) regressions, the mean estimated coefficient of illiquidity for all months
is 0.0014 and significant at the 5 percent level. When January is excluded, the coefficient is
0.0016 and significant at the 1 percent level. These are consistent with Amihuds results and
suggest the existence of a positive cross-sectional relationship between illiquidity and stock
returns in general. However, the relationship appears unstable over the two sub-periods. The
mean illiquidity coefficient in the first sub-sample period is 0.0024, much larger compared to the
all-month sample and highly significant at the 1 percent level, while it is only 0.0003 for the
second sub-period and not significant at all. In contrast, the results reported in Amihud (2002)
show that the illiquidity is significant in both sub-sample periods.
9
Our estimated mean beta
coefficient is negative, though not statistically significant, in all the four-variable regressions. In
contrast, Amihud shows that the similar beta estimates for NYSE stocks are all positive and
significant. However, the insignificant beta estimates are consistent with previous studies for the
Japanese market. Hodoshima, GarzaGomez and Kunimura (2000) find that a regression of
return on beta without differentiating positive and negative market excess returns produces a flat
relationship between return and beta in Japan for the period of 1956-1995. While the mean
estimated coefficients for past returns,
i
y
PR
1
1
and
i
y
PR
2
1
, are all positive and mostly significant
for NYSE stocks, as reported in Amihud (2002), our estimates for
i
y
PR
1
1
are all negative but
only marginally significant for the first sub-sample period and the sample period excluding
January. Our estimates for
i
y
PR
2
1
are all insignificant.
For equation (8) regressions, the results are consistent with equation (7) regressions in the
sense that illiquidity is positive and significantly priced in all but the second sub-sample period.
However, for all equation (8) regressions, the mean coefficient estimates for various control
variables are insignificant except for
i
y
STD
1
with the sample excluding January, which is
negative and significant at the 5 percent level. This is a bit surprising given the findings of
Amihud (2002) that all estimates but beta are significant in his seven-variable regressions. This
suggests that the determinants of stock returns in the U.S. may not be the same as those in Japan.
It is conceivable that
i
y
PR
1
1
and
i
y
PR
2
1
may have prediction power for stock returns in the
U.S. but not necessarily in Japan. On the other hand, Bremer and Hiraki (1999) document that

9
We have also tried sub-sample periods 1976-1990 vs. 1991-1999 and 1976-1988 vs. 1989-1999. The results are
similar.
11
TSE stocks with short-term price reversals (stocks with losses in week t-1 experience price
reversals in week t). As mentioned earlier,
i
y
DP may not be a good cross-sectional determinant
for stock returns because many Japanese firms simply do not pay or pay very little dividends. In
addition,
i
y
CP and
i
y
BM are documented by Chan et al. (1991) as major determinants for
Japanese stock returns. Therefore, we replace
i
y
PR
1
1
and
i
y
PR
2
1
with
i
m
PR
1
, the one-month
lagged return for stock i,
i
y
DP
1
with
i
y
CP
1
and further add in
i
y
BM
1
in our regressions. In
addition, since we have monthly data for illiquidity
i
m
ILL
1
, we use it to replace the annual
illiquidity so that the right-hand side variables are not just annual variables.
Table 4 presents the regression results for the following specifications:

i
m y
i
y y
i
m y y
i
m
PR k k ILLM k k R
1 3
1
1 2 1 1 0
+ + + = (9)

i
y y
i
y y
i
y y
i
y y
i
m y
i
y y
i
m y y
i
m
CP k STD k CAP k BM k R k k ILLM k k R
1 7 1 6 1 5 1 4 1 3
1
1 2 1 1 0
ln

+ + + + + + + = (10)

For equation (9) regressions, the results are similar to the equation (7) regression results reported
in Table 3.
i
m
ILLM
1
shows a positive and significant impact on expected stock returns for the
all-month sample, the sample excluding January, and the first sub-sample period, but nothing
significant for the second sub-sample period. While
1i
y-1
, is still not significant,
i
m
PR
1
is
negative and highly significant in all four samples, suggesting monthly price reversals for
Japanese stocks.
For equation (10) regressions, we find that
i
y
BM
1
is indeed positive and significant in all
but the second sub-sample period regressions. This seems consistent with Chan et al. (1991).
However,
i
y
CP
1
is only significant in the second sub-sample period.
i
m
ILLM
1
is still positive but
its significance level is reduced. For the all-month sample, the t-value becomes marginally
insignificant. For the sample excluding January, the t-values are only significant at the 10 percent
level. For the first sub-sample, it is significant at the 5 percent level, while the illiquidity
coefficient is mostly significant at the 1 percent level for those samples in equation (7), (8) and
12
(9) regressions (as shown in Tables 3 and 4).

For the second sub-sample period,
i
m
ILLM
1
is still
insignificant but
1i
y-1
and Ln
i
y
CAP
1
become significant.
On the whole, our results suggest that illiquidity is priced in the Japanese market but not
so in the second sub-sample period. This pattern is robust across different specifications. From
Figure 1, we see that the first sub-sample period corresponds to a booming market trend with
declining illiquidity, while the second sub-sample period coincides with a down and oscillating
market and a trend of increasing illiquidity. A further examination of the monthly market excess
return shows that, for 168 months in the first sub-sample period, 111 months are associated with
the positive market excess return and 57 with negative ones. For 120 months in the second sub-
sample period, only 54 months are associated with the positive market excess return and 66 with
negative ones. Therefore, the ratio of negative excess market return months over positive ones is
much higher in the second sub-sample period (66/54) than in the first (57/111).
Using stock return data in Japan from 1956 to 1995, Hodoshima, et al. (2000) find that
regression of return on beta without differentiating positive and negative market excess returns
produces a flat relationship between beta and return. However, significant conditional positive or
negative relationships between beta and return are found once the observations are separated into
up and down market groups, where the up (or down) market refers to observations associated
with positive (or negative) market premium, R
m
R
f
>0 (R
m
R
f
<0).
10
They explain that the
expected market excess return should never be negative, but actual observations used in the
regression are often negative. Similarly, one may argue that the expected illiquidity premium
should never be negative but the realized premium may well be so. If the realized illiquidity
premium is positively correlated with excess market return, then the estimated relationship
between the Amihud ratio and stock returns may be distorted. Therefore, we further separate our
sample into up and down markets and repeat the cross-sectional regressions to see if illiquidity is
priced differently in those market states. To save space, we only repeat the regressions for
equations (9) and (10) and present the results in Table 5.
11


10
Chan and Lakonishok (1993), Grundy and Malkiel (1996), and Pettengill et al. (1995) investigate the relationship
between return and beta by taking into account whether the market excess return is positive or negative in the US
market and find that the beta and stock returns are significantly related.
11
The up and down market results for equations (7) and (8) are qualitatively the same.
13
Panels A and B of Table 5 present the results for equations (9) and (10) in up and down
markets, respectively. Here, the beta is mostly significant. Specifically, the beta is positive for
the up market but negative for the down market. This is very much consistent with the findings
of Hodoshima et al. (2000). In addition,
i
m
ILLM
1
is insignificant for all samples in the down
market. This suggests that the Amihud ratio cannot be properly priced in the down market.
Moreover, even during the up market, illiquidity is still not priced in the second sub-sample
period. Therefore, the insignificant mean estimate for the illiquidity coefficient in the second
sub-sample period is not totally or even mainly due to the concentration of more negative excess
market returns in the period. The second sub-sample period may be a special period because it
coincides with a lot of changes in the Japanese financial system. Hamao, Mei and Xu (2003)
document that idiosyncratic volatility in Japanese stocks has fallen, coinciding with a slowdown
in the capital allocation process within the Japanese economy. They opined that Japanese
corporate managers may have chosen to bail out large companies rather than allocate capital to
young companies and that this caused the stock prices of Japanese stocks in the 1990s to be more
correlated compared to those of the 1980s. It is possible that corporate behavior as well as
investor risk tolerance may have experienced some big changes in the 1990s, rendering
illiquidity risk not being captured by the Japanese data during this period.
However, in contrast to equation (9) results, the estimated mean coefficient of
i
m
ILLM
1

for equation (10) is mostly insignificant in the up and down market. On the other hand, beta,
i
y
STD
1
, Ln
i
y
CAP
1
,
i
y
CP
1
and
i
y
BM
1
are largely significant. Recall that in table 4, we find that
once additional variables such as
i
y
BM
1
, Ln
i
y
CAP
1
, etc. are controlled, the significance of the
illiquidity coefficient is reduced. Now the illiquidity effect is almost all subsumed once the up-
market and down-market are further controlled. It is possible that the correlation between the
Amihud ratio and other control variables, especially size, may become stronger once the sample
is divided into up and down markets.

IV. Time Series Effect of Illiquidity
Amihud (2002) argues that stocks are not only riskier but also less liquid than short-term
treasury securities. Hence, stock return in excess of the T-bill rate (risk premium) includes a
premium for illiquidity. It follows that if investors anticipate higher market illiquidity, they will
14
expect higher returns. More specifically, expected stock returns should be positively related to
expected illiquidity while unexpected illiquidity should be negatively related to
contemporaneous unexpected stock return.
Following Amihud (2002), the market illiquidity used in the time-series test is the
logarithmic form of the average illiquidity across stocks. Since the yearly time series is short,
with only 24 data points from 1976-1999, we focus on the monthly data (288 observations). The
expected and unexpected market illiquidity are estimated through the AR(1) model
m
M
m
M
m
v ILL c c ILL + + =
1 1 0
ln ln , (11)
where
M
m
ILL ln is the monthly market illiquidity as defined in Section II and v
m
is the residual
representing the unexpected market illiquidity
M
m
ILLU ln . Investors determine the expected
illiquidity
M
m
ILLE ln at the beginning of a month based on information from the previous month,
M
m
M
m
ILL c c ILLE
1 1 0
ln ln

+ = , (12)
The market price is then set at the beginning of the month through the following model to
generate the expected return,
m
M
m m
M
m
f
m
M
m
u ILL g g u ILLE f f R R + + = + + =
1 1 0 1 0
ln ln . (13)
In this equation,
0 1 0 0
c f f g + = and
1 1 1
c f g = ,
M
m
R is the return of market portfolio M (all stocks
in our sample) in month m
12
and R
f
m
is the monthly call money rate or one-month Gensaki rate
for month m, as in Chan et al. (1991). This is retrieved from the monthly key economic file and
u
m
can be decomposed into the unexpected illiquidity
M
m
ILLU ln and an error term w
m
. After
controlling for the January effect, the time-series regression of the excess market return on the
market illiquidity is as follows
13

0 1 1 2 3
ln ln
M f M M
m m m m m m
R R g g ILL g ILLU g JAN w
= + + + + . (14)
The two testable hypotheses are:

H1: expected market illiquidity is positively related to expected market excess return (g
1
>0).
H2: unexpected market illiquidity is negatively related to contemporaneous market excess
return (g
2
<0).

12
It is calculated as the log difference of two consecutive month-end daily closing prices.
13
A similar specification has been used by French, Schwert and Stambaugh (1987) in testing the effect of risk on
excess stock return.
15
Amihud (2002) further puts forward and tests the flight to liquidity hypothesis. Amihud,
Mendelson and Wood (1990)
14
point out that there are two effects on expected stock returns
when expected market illiquidity rises. On the one hand, the stock price declines and expected
returns rise for all stocks, while on the other, capital flies from less liquid to more liquid stocks.
These two effects reinforce each other for illiquid stocks but offset each other for liquid ones.
Increasing market illiquidity not only negatively affects prices for illiquid stocks but also induces
investors to switch to more liquid stocks, which further depresses the price for illiquid stocks.
However, increasing market illiquidity leads to an increase in demand for liquid stocks, which
mitigates their price decline. Therefore, the illiquidity effect should be stronger for small stocks
and weaker for large stocks because firm size is negatively correlated with illiquidity. Replacing
market portfolio return series with size portfolio return series
15
, we can rewrite the equation (14)
as follows
p
m m
p M
m
p M
m
p p f
m
p
m
w JAN g ILLU g ILL g g R R + + + + =
3 2 1 1 0
ln ln , (15)

where p denotes one of the 25 size-based portfolios (portfolio 25 contains the largest stocks) and
p
m
R is the average return across stocks in portfolio p for month m. The testable hypotheses are:
H3: The expected illiquidity effect
p
g
1
should be positive and decrease as firm size does
( 0
25
1
20
1
15
1
10
1
5
1
> > > > > g g g g g ).
H4: The unexpected illiquidity effect
p
g
2
should be negative and increase as firm size does
( 0
25
2
20
2
15
2
10
2
5
2
< < < < < g g g g g ).

Amihud (2002) includes two additional variables, default yield premium and term yield
premium, in his time series test. Since we do not have default yield data for Japan, only term
yield premium TM
m
is used in the expanded specifications

0 1 1 2 3 1
ln ln
M f M M
m m m m m m m
R R g g ILL g ILLU g JAN aTM u

= + + + + + (16)

14
Amihud et al. (1990) study the relationship between market liquidity and market return for the 1987 crash by
estimating the effects of changes in the bid-ask spread, the initial spread, and the change in quote size or the change
in stock prices for three periods around the crash. They reason that the price decline reflects, in part, a reassessment
of market liquidity, while price recovery results from liquidity improvement to some degree.
15
25 size portfolios are the same as those formed in Section III,
16
and
0 1 1 2 3 1
ln ln
p f p p M p M p p p
m m m m m m m
R R g g ILL g ILLU g JAN a TM u

= + + + + + . (17)

The term yield premium
3 G
m m m
TM YL R = is computed as the difference between the yield to
maturity of 10-year government bonds (
m
YL ) and the three-month Gensaki rate in month m
(
3 G
m
R ). The additional hypothesis about the premium is that 0 > a .
Kendall (1954) points out that the estimated coefficient
1
c from the finite samples is
biased downward in AR(I) models such as equation (11). He proposes a simple but accurate bias
correction approximation procedure: the estimated coefficient
1
c is augmented by the term
T c / ) 3 1 (
1
+ , where T is the sample size.
Our estimation of equation (11) provides the following results for the time-series test
with monthly illiquidity,
m
M
m
M
m
v ILL ILL + + =
1
ln 918 . 0 213 . 0 ln . Applying Kendalls bias
correction method, the adjusted slope coefficient is 0.931 (and the intercept is adjusted
accordingly)
16
. The monthly unexpected illiquidity is calculated as a residual from the above
autoregressive model using the adjusted coefficients.
Table 6 presents the estimation results for the whole sample period from July 1976 to
June 2000 (288 months). Several observations are evident. First, including the term yield
premium TM
m
in the estimation has no impact on the results. The coefficient estimate is not
significantly different from zero and the inclusion of TM
m
has almost no effect on the estimated
coefficients of other variables. Second, expected illiquidity
p
m
ILL
1
ln

has no impact on expected
return at all and there is no monotonic pattern for the estimated coefficients across size portfolios
either. No matter for the market or individual portfolios, the estimated coefficient is insignificant.
This is inconsistent with our H1 that expected illiquidity should be positively and significantly
related to the expected stock return and H3 that expected illiquidity effect is decreasing along
with firm size. Third, unexpected illiquidity
p
m
ILLU ln is negatively associated with expected
stock return premiums and the estimated coefficient is significant and monotonically increasing

16
The corresponding estimate in Amihud (2002) is
m
M
m
M
m
v ILL ILL + + =
1
ln 945 . 0 313 . 0 ln
and the adjusted slope
coefficient is 0.954.
17
along with firm size. This finding is consistent with H2 that contemporaneous stock return is
negatively associated with unexpected illiquidity and H4 that the unexpected illiquidity effect is
negative and increases as firm size does. Finally, we find that the January effect is more
significant for smaller stock portfolios but insignificant for the portfolio with the largest firms.
While the last two observations are consistent with the findings in Amihud (2002), the first two
are not.
We repeat the regressions for equations (16) and (17) over the two sub-sample periods.
Since equation (17) results are similar to that of equation (16) and the estimated coefficient for
TM
m
is never significant, table 7 only presents the regression results for equation (16).
Unlike the cross-sectional regression results presented in tables 3 to 5, our time series
regression results are consistent across the two sub-sample periods. They are largely the same as
those presented in table 6. We also performed the same tests using portfolios formed on book-to-
market value and using yearly data. The results are again very much alike.
17
So the expected
illiquidity proxied by the lagged Amihud ratio cannot help to predict future stock returns in
Japan but the unexpected illiquidity derived from first order autoregression of the Amihud ratio
does have significant impact on contemporaneous returns. Why expected illiquidity does not
help predict stock returns in Japan is puzzling and deserves further studies.

V. Liquidity-Adjusted CAPM
Acharya and Pedersen (2005) first set up a one-beta CAPM for net returns

1 1 1 1
1 1
1 1
cov ( , )
( )
var ( )
i i M M
i i f t t t t t
t t t t M M
t t t
R C R C
E R C R
R C
+ + + +
+ +
+ +

= +
, (18)
where
i
t
i
t
i
t i
t
p
p d
R
1
+
= is the expected gross return of an asset,
i
t
i
t i
t
p
c
C
1
= is the relative illiquidity
cost,
i
t
i
i
i
t
i
t
i
i
M
t
p s
p d s
R
1
) (

+
= is the market return, and
i
t
i
i
i
t
i
i
M
t
p s
c s
C
1
= is the relative market

illiquidity.
i
t
d and
i
t
c refer to dividend paid and illiquidity cost (the cost of selling the asset)

17
We do not report the results to save space.
18
incurred for security i during each time t, respectively.
i
t
p denotes the price for security i at the
end of time t and
i
s denotes the number of shares for security i. ) (
f M
t
M
t
R C R E = is the
risk premium. Further expanding and rewriting equation (18), Acharya and Pedersen derive the
following liquidity-adjusted CAPM
18

) ( var
) , ( cov
) ( var
) , ( cov
) ( ) (
1 1
1 1
1 1
1 1
1 1
M
t
M
t t
M
t
i
t t
t
M
t
M
t t
M
t
i
t t
t
i
t t
f i
t t
C R
C C
C R
R R
C E R R E
+ +
+ +
+ +
+ +
+ +
+ + =

) ( var
) , ( cov
) ( var
) , ( cov
1 1
1 1
1 1
1 1
M
t
M
t t
M
t
i
t t
t
M
t
M
t t
M
t
i
t t
t
C R
R C
C R
C R
+ +
+ +
+ +
+ +
. (19)

Assuming constant conditional covariances of innovations in illiquidity and returns, equation (19)
can be expressed as

i i i i i
t
f
t
i
t
C E R R E
4 3 2 1
) ( ) ( + + = (20)

This simple theoretical framework suggests that (1) the liquidity risk affect asset price through
the last three terms in the equation, and (2) the effects of liquidity level ) (
1
i
t t
C E
+
and liquidity risk
are separate. The market risk is captured by
p 1
. The three liquidity risk factors are the
commonality-in-liquidity
p 2
, individual stock return sensitivity to market illiquidity
p 3
, and
individual stock illiquidity sensitivity to market return
p 4
.
The commonality-in-liquidity, ) , ( cov
1
M
t
i
t t
C C
, should be positively associated with the

expected return because investors need to be compensated for the risk that their assets may
become illiquid when the market as a whole is illiquid. The return sensitivity to market illiquidity,
) , ( cov
1
M
t
i
t t
C R
, should have a negative impact on the expected return because investors are
willing to accept a lower return on an asset normally associated with a high return in times of
market illiquidity. The illiquidity sensitivity to market return, ) , ( cov
1
M
t
i
t t
R C
, should also have a

negative impact on the expected return because investors are willing to accept a lower return on
an asset with high liquidity when the market return is low.

18
For model assumptions and derivation details see Acharya and Pedersen (2005).
19
Using CRSP data from 1962 to 1999, Acharya and Pedersen show that the required return
of a security i does increase in the covariance between its illiquidity and market
illiquidity, ) , ( cov
1
M
t
i
t t
C C
and decreases in the covariance between the securitys return and

market illiquidity, ) , ( cov
1
M
t
i
t t
C R
as well as its illiquidity and market return, ) , ( cov

1
M
t
i
t t
R C
. In
addition, they show that the liquidity-adjusted CAPM explains the data better than the standard
CAPM. Furthermore, they present some evidence that liquidity risk is important over and above
the effects of market risk and the level of liquidity.
In order to test the liquidity-adjusted CAPM, we estimate illiquidity cost and various
betas. First, we compute return and normalized illiquidity for each firm each month. Second, we
form a market portfolio and 25 test portfolios sorted by variations in illiquidity, size or BM ratio
and compute monthly returns and normalized illiquidity for these portfolios. Thirdly, we estimate
illiquidity innovations and liquidity betas for each portfolio. We then test the liquidity-adjusted
CAPM by applying GMM to estimate equation (20) and its variations.
19

Since the Amihud ratio is measured in percent per yen, while the relative liquidity cost
in the liquidity-adjusted CAPM is the ratio of the cost of selling over ex-dividend price (
i
t
i
t
p c
1
/

)
specified in yen cost per yen invested, we employ a normalized monthly stock illiquidity
computed as follows
1
0.25 0.30 *
i i M
m m m
C ILL CAPR

= + , (21)
where
i
m
ILL , for stock i in month m is multiplied by the market capitalization index of market
portfolio M in month m-1,
M
m
CAPR
1
is the ratio of the capitalization of the market portfolio at the
end of month m-1 over that at the end of June 1975. According to Acharya and Pedersen (2005),
M
m
CAPR
1
can adjust for inflation and thus make the normalized illiquidity relatively stationary
over time. The coefficients 0.25 and 0.30 are chosen such that the cross-sectional distribution of
normalized illiquidity (
i
m
c ) has approximately the same level and variance of the effective spread.
Since the normalized illiquidity is still a noisy measure, we use portfolios instead of individual
stocks in our analyses to alleviate this problem.

19
The GMM point estimates are the same as those derived using OLS (either in a pooled regression or using the
Fama-MacBeth (1973) method). However, GMM is more efficient because it takes into consideration possible
heteroscedesticity and serial correlation.
20
The market portfolio is formed every month during the sample period from 1975 to 1999
using stocks with beginning-of-month prices higher than 100 yen and at least 15 days of return
and volume data in a month. Meanwhile, 25 illiquidity (LQ) and 25 illiquidity volatility (LQV)
portfolios are formed for each year between 1976 and 1999, drawing on the sample presented in
the preceding section, sorting by their annual illiquidity and annual standard deviation of
illiquidity, respectively, for the previous year.
The equally-weighted portfolio return
p
m
R and portfolio-adjusted illiquidity
p
m
C for each
portfolio p in each month m, are defined as
=
=
p
I
i
i
m
p
p
m
R
I
R
1
1
, p=1,2,, 25, (22)
and
=
=
p
I
i
i
m
p
p
m
C
I
C
1
1
, p=1,2,, 25, (23)
where
i
m
R and
i
m
C are the monthly return and monthly normalized illiquidity for stock i in each
month m, p is the rank of the portfolio, and
p
I is the number of stocks in portfolio p. The
equally weighted return and illiquidity are used in line with the work of Amihud (2002) and
Chordia, Roll and Subrahmanyam (2000). Furthermore, the equally weighted market return and
market illiquidity compensate for the over-representation of large liquid stocks.
Our normalized illiquidity for LQ-sorted portfolios ranges from 0.25% to 0.92%, with a
mean of 0.41% and standard deviation of 0.18%. We do not have bid-ask spread data for TSE to
compare with our normalized illiquidity. However, Ahn, Cai, Hamao and Ho (2002) provide
some statistics on the three groups of stocks formed out of their sample of 204 Nikkei 225
component stocks from Jan. 5 to Mar. 31, 2000. The average spread is 0.56% for Group 1 with
an average price less than or equal to 2000, 0.31% for Group 2 with an average price between
2001 and 3000, 0.34% for Group 3 with an average price between 3001 and 30000. The
average spread for the whole sample is 0.40%, almost identical to our estimation of an average
normalized illiquidity of 0.41%.
Since illiquidity is persistent, innovations in illiquidity ) (
1
p
m m
p
m
C E C

are used to
compute liquidity betas, which are obtained from the following regression for each portfolio p as
well as the market portfolio M
21
) * 30 . 0 25 . 0 ( * 30 . 0 25 . 0
1 1 1 0 1
M
m
p
m
M
m
p
m
CAPR ILL a a CAPR ILL

+ + = +
p
m
M
m
p
m
u CAPR ILL a + + +

) * 30 . 0 25 . 0 (
1 2 2
, (24)
where
p
m
u is used as the standardized illiquidity innovation ) (
1
p
m m
p
m
C E C

. The three unadjusted
portfolio illiquidities,
p
m
ILL ,
p
m
ILL
1
and
p
m
ILL
2
, are multiplied by the same capitalization ratio
for the market portfolio
M
m
CAPR
1
. This is to ensure that we measure innovations in illiquidity
only, not the changes in
M
m
CAPR .
Meanwhile, innovations in market portfolio return are also estimated in an AR(2)
specification that sets a control for market characteristics available at the beginning of the month,
in spirit to Sadka (2003); in particular, market returns and volatility, the average market
illiquidity, the log of trading value and turnover for the market are all measured over the past six
months, while the log of market capitalization is measured with a one-month lag.
With the above estimates, we compute,
p 1
,
p 2
,
p 3
and
p 4
, for each portfolio and
using all monthly return and illiquidity observations, for the whole sample period 1976-1999
20
as
well as the sub-sample periods 1976-1989 and 1990-1999.
[Insert Table 8]
Panel A of Table 8 presents descriptive statistics for odd-numbered LQ-sorted portfolios.
Consistent with Acharya and Pedersen (2005), we see that the sort on the past illiquidity does
produce portfolios with monotonically increasing average illiquidity ( ( )
p
E C ) from portfolio 1
through portfolio 25. Also, stocks with high average illiquidity tend to have high market beta
(
p 1
), high volatility of illiquidity innovation ( ) (
p
c ), high book-to-market ratio (BM), high
volatility of portfolio returns ( ) (
p
R ), and small market capitalizations (CAP). However, for
the three illiquidity betas-- the main variables of our interest-- the results are largely inconsistent
with those found by Acharya and Pedersen. First, while Acharya and Pedersen find that all betas
have predicted signs and monotonic trends with significant t-values for most portfolios, we only
find
p 1
and
p 4
with significant t-values. In addition, we find
p 2
has a negative sign for some
of the portfolios, which is inconsistent with the expectation. Furthermore, we find that neither
p 2
nor
p 3
has an obvious trend in illiquidity level. Only
p 4
has the expected negative sign

20
We use the entire time-series to compute the four betas as Black, Jensen and Scholes (1990) and Pastor and
Stambaugh (2003) have done. Therefore, it is not directly comparable to the yearly betas computed in Section III.
22
and a monotonically decreasing trend with illiquidity. Second, we do not see turnover (TRN) and
portfolio excess returns ) (
f p
R R E show any obvious trend with the LQ-sorted portfolios as
those found in Acharya and Pedersen. Also, the portfolio excess return is not monotonically
increasing at the expected illiquidity level, ( )
p
E C , indicating that the high illiquidity level may
not be associated with high portfolio excess returns. Panel B of Table 8 presents the same
statistics for the sub-sample periods 1976-1989 and 1990-1999. They are similar to those
reported in Panels A.
Following Acharya and Pedersen, we also compute the similar statistics for odd-
numbered portfolios sorted by a standard deviation of the Amihud ratio (LQV-sorted portfolios).
In general, the results are qualitatively the same as those presented in Panel A and B. We further
computed the betas for portfolios sorted by SZ and BM, respectively. Again
p 2
and
p 3
are
generally insignificant and do not show the trends documented by Acharya and Pedersen. For
BM-based portfolios, even
p 4
does not show any trend. All these are not reported to save space.
Overall, the above results suggest that commonality-in-liquidity (
p 2
) and individual
return sensitivity to market illiquidity (
p 3
) may not have the predicted effect in Japan as they
have in the U.S. However, the illiquidity sensitivity to market return (
p 4
) seems to have the
same predicted effect in Japan as it does in the U.S. These results are not sensitive to the sub-
sample periods and the various portfolio-sorting criteria.
We further compute the correlation matrices for the betas of LQ-sorted portfolios in the
whole sample period as well as sub-sample periods. Surprisingly, we find that
p 1
and
p 2
are
negatively correlated. Although the absolute values of the correlation coefficients are generally
lower than those reported in Acharya and Pedersen (2005), they are still high enough to cause
multicollinearity in cross-sectional regressions, especially the correlation coefficients between
p 1
,
p 3
and
p 4
.
21

Table 9 presents the GMM estimates of equation (20) and its variations. Panel A reports
the regressions for LQ-sorted portfolios. Assuming ) (
f M
t
M
t
R C R E = is the same for all
betas in equation (20), we define the net beta as

21
The detailed results are not reported but available upon request.
23

1 2 3 4 np p p p p
= + , (25)
and the liquidity-adjusted CAPM reduces to
( ) ( )
p f p np
m m m
E R R E C = + + . (26)
As a comparison, we also estimate the standard CAPM to be

p f
m
p
m
R R E
1 1
) ( + = . (27)
The estimation results for equations (26) and (27) are shown in line 1 and line 2, respectively.
22

For equation (26), while the estimated coefficient of the illiquidity level is highly positive and
significant, the net beta coefficient is 0.0432 with a t-value of only 0.06, which is totally
insignificant. For equation (27), the standard market beta coefficient is 0.9285 and not
statistically significant. This suggests that it is the illiquidity level rather than illiquidity risk that
has some impact on expected returns. In contrast, both net beta,
np
, and standard market beta,
p 1
, are positive and highly significant in the results of Acharya and Pedersen (2005). The
adjusted R
2
for (26) and (27) are 0.425 and 0.331, while the corresponding ones reported in
Acharya and Pedersen are 0.809 and 0.638.
Next, we isolate the effect of liquidity betas from the level of liquidity ) (
p
m
C E and the
market beta
p 1
by testing the following:
np p p
m
f
m
p
m
C E R R E + + + =
1 1
) ( ) ( , (28)
np p f
m
p
m
R R E + + =
1 1
) ( . (29)
The results are presented in lines 3 and 4, respectively. The net beta,
np
, is significantly
positive in both equations (28) and (29) but the estimated coefficient of the liquidity level, , is
negative in equation (28), a result which is counter-intuitive. While Acharya and Pedersen also
report a negative , theirs is statistically insignificant. In our case, however, is significant at the
1 percent level. When Acharya and Pedersen say that liquidity risk matters over and above
market risk and liquidity level, they admit that multicollinearity problems make their evidence
weak. Our results are weaker when compared to theirs. The derived market beta coefficient,

22
Acharya and Pedersen (2005) report results both with estimated at its calibrated value and with as a free
parameter to be estimated. We omit the results with a calibrated to save space. In all regressions with calibrated ,
the results are always worse than those where is allowed to be freely estimated in terms of adjusted R
2
and the
significance level of estimated betas.
24
based on equation (25), is positive in their case but negative in ours for both equations (28) and
(29),
) ( 49 . 155 40 . 0 49 . 155 59 . 155
4 3 2 1 1 p p p p np p
+ = +
and
09 . 38 21 . 0 09 . 38 30 . 38
1 1
+ = +
p np p
) (
4 3 2 p p p

This is counter-intuitive. Moreover, we find a negative and significant .
Finally, an unconstrained version that allows different risk premia for different betas by
using the following form is estimated.
1 1 2 2 3 3 4 4
( ) ( )
p f p p p p p
m m m
E R R E C = + + + + + . (30)
The estimated is significantly negative. The estimated risk premia for
p 1
,
p 2
, and
p 3
are
not statistically significant while the risk premia for
p 4
is negative and very significant.
However, no strong inference can be made due to severe multicollinearity among the four betas
and the results are not reported.
As found in previous sections, the liquidity effect on stock returns is different across sub-
sample periods. Therefore, we repeat the GMM regressions using the betas presented in Panels B
of Table 8 for the two sample periods, (1976-1989) and (1990-1999), and report the results in
Panels B and C of Table 9, respectively.
The results are not much better for the first sub-sample period shown in Panel B. The net
beta is not significant in line 1 while the market beta in line 2 is highly significant. The higher
adjusted R
2
(0.349) for line 1 regression than that for line 2 regression (0.257) is obviously due
to the level of liquidity included in equation (26) but not due to the net beta. Results in lines 3
and 4 are a little better than those reported in Panel A in the sense that the market beta
coefficients after adjustment are positive (222.70 - 222.32 > 0 and 108.31 107.68 > 0).
However, the estimated coefficient for the liquidity level is still significantly negative. The
results for the second sub-sample period shown in Panel C are no better than those shown in
Panel B. Therefore, separating the data into two sub-sample periods does not produce more
supportive evidence for the liquidity-adjusted CAPM.
We repeat similar regressions for LQV-sorted portfolios and size- and BM-sorted
portfolios using factors such as value-weighted instead of equal-weighted portfolios, etc. The
results are no better than those reported in Panel A of Table 9 either.
25
However, when we separate the data into positive and negative market premium groups
and run pooled GLS regressions for the up and down market respectively, there is some mild
evidence that liquidity-adjusted CAPM has a better goodness-of-fit than the standard CAPM in
Japan. This again suggests that CAPM-type of model may be sensitive to up and down markets
in Japan, as documented by Hodoshima, et al. (2000).
23

VI. Summary and Conclusion
Using a new illiquidity measure proposed by Amihud (2002), we conduct a
comprehensive study on the relationship between illiquidity, illiquidity risk and stock returns
among stocks listed on the Tokyo Stock Exchange. We first examine the cross-sectional
relationship between illiquidity and stock returns and find that illiquidity has a positive impact on
stock returns in Japan in general but not in the second sub-sample period of 1990-1999. Even
after using a control to account for up and down markets, we still fail to find a significant
relationship between illiquidity and stock returns in the second sub-sample period. Therefore, the
results are not totally consistent with those of Amihud (2002).
Next, we look at the time series relationship between illiquidity and stock returns. Again,
the results are not totally consistent with those found by Amihud. While unexpected illiquidity
does have a negative impact on contemporaneous stock returns, the expected illiquidity does not
have any impact on expected stock returns. In addition, evidence for the fly to liquidity
hypothesis associated with expected illiquidity is not supportive. Separating the whole sample
into two sub-periods produces similar results.
Finally, we test to see if the liquidity-adjusted CAPM proposed by Acharya and Pedersen
(2005) can explain data better than the standard CAPM. Acharya and Pedersen find some
supportive evidence that their liquidity-adjusted CAPM is superior to the standard CAPM using
data from the NYSE. However, our results are not supportive of this claim. Even when we
separate the sample into two sub-periods, it is still not supportive. Only when we divide the
sample into up and down markets do the results show some mild support for the liquidity-
adjusted CAPM.
Overall, our mixed results indicate that the liquidity-stock return relationship found in the
U.S. cannot be consistently replicated in Japan. Specifically, the period 1990-1999 in Japan may

23
All results not reported here are available upon request.
26
be a unique period, with a lot of transitions rendering the cross-sectional relationship between
illiquidity and stock returns undetectable. Also, the expected illiquidity measured by a lagged
Amihud ratio does not have any impact on the expected stock premium. Our findings also
indicate that the liquidity-adjusted CAPM may be too simple to be universally applicable.
Another possibility is that the Amihud ratio does not measure illiquidity well in Japan.
Our study contributes to the extant literature in two directions. First, we provide a
comprehensive study on the liquidity -stock return relationship over a long period of time for the
second largest stock market in the world. Second, we show that the Amihud ratio as a measure to
capture the illiquidity effect may be sensitive to time periods and the presence of up and down
markets, which deserves further explorations.

27
References

Acharya, V. & Pederson, L. (2005). Asset Pricing with Liquidity Risk. Journal of Financial
Economics 77, 375-410.

Ahn, J., Cai, J., Hamao, Y. & Ho, R(2002). The components of the bid-ask spread in a limit-
order market: evidence from the Tokyo Stock Exchange. Journal of Empirical Finance 9, 399-
430.

Amihud, Y. (2002). Illiquidity and Stock Returns: Cross-Section and Time-Series Effects.
Journal of Financial Markets 5, 31-56.

Amihud, Y. & Mendelson, H. (1980). Dealership market: market making with inventory. Journal
of Financial Economics 8, 311353.

Amihud, Y. & Mendelson, H. (1986). Asset pricing and the bidask spread. Journal of Financial
Economics 17, 223249.

Amihud, Y., Mendelson, H. & Wood, R. (1990). Liquidity and the 1987 stock market crash.
Journal of Portfolio Management 16, 6569.

Bekaert, G., Harvey, C. & Lundblad, C. (2003). Liquidity and Expected Returns: Lessons from
Emerging Markets. New York: Columbia University.

Black, F., Jensen, M. & Scholes, M. (1990). The capital asset pricing model: some empirical
tests, in Michael Jensen (ed.), Studies in the Theory of Capital Markets. Praeger, New York.

Bremer, M. & Hiraki, T. (1999). Volume and individual security returns on the Tokyo Stock
Exchange. Pacific-Basin Finance Journal 7, 351370.

Brennan, M., Chordia, T. & Subrahmanyam, A. (1998). Alternative factor specifications,
security characteristics, and the cross-section of expected stock returns. Journal of Financial
Economics 49, 345373.

Brennan, M. & Subrahmanyam, A. (1996). Market microstructure and asset pricing: on the
compensation for illiquidity in stock returns. Journal of Financial Economics 41, 441464.

Chalmers, J. & Kadlec, G. (1998). An empirical examination of the amortized spread. Journal of
Financial Economics 48, 159188.

Chan, L., Hamao, Y. & Lakonishok, J. (1991). Fundamental and Stock Returns in Japan. Journal
of Finance 46, 1739-1764.

Chan, L. & Lakonishok, J. (1993). Are the Reports of Betas Death Premature? Journal of
Portfolio Management 19(4), 51-62.

28
Chordia, T., Roll, R. & Subrahmanyam, A. (2000). Commonality in Liquidity. Journal of
Financial Economics 56, 3-28.

Easley, D., Hvidkjaer, S. & OHara, M. (2002). Is information risk a determinant of asset returns?
Journal of Finance 57, 2185-2221.

Eleswarapu, V. (1997). Cost of transacting and expected returns in the NASDAQ market.
Journal of Finance 52, 21132127.

Fama, E. & French, K. (1992). The cross section of expected stock returns. Journal of Finance
47, 427465.

Fama, E. & MacBeth, J. (1973). Risk, return and equilibrium: empirical tests. Journal of
Political Economy 81, 607636.

French, K., Schwert, G. & Stambaugh, R. (1987). Expected stock returns and volatility. Journal
of Financial Economics 19, 329.

Glosten, L. & Harris, L. (1988). Estimating the components of the bidask spread. Journal of

Grundy, K. and Malkiel, B. (1996). Reports of betas death have been greatly exaggerated.
Journal of Portfolio Management 22(3), 36-44.

Hamao, Yasushi, Jianping Mei & Yexiao Xu (2003). Idiosyncratic risk and the creative
destruction in Japan, NBER working paper.

Hamori, S. (2001). Seasonality and stock returns: some evidence from Japan. Japan and the
World Economy 13, 463-481.

Hasbrouck, J. (1991). Measuring the information content of stock trades. Journal of Finance 46,
179207.

Hasbrouck, J. (2002). Inferring Trading Costs from Daily Data: US Equities from 1962 to 2001.
New York City: New York University.

Hodoshima, J., Garza-Gomez, X. & Kunimura, M. (2000). Cross-sectional regression analysis of
return and beta in Japan. Journal of Economics and Business 52, 515-533.

Hu, S. (1997). Trading turnover and expected stock returns: the trading frequency hypothesis and
evidence from the Tokyo Stock Exchange. Working Paper, National Taiwan University.

Japan Securities Research Institute( 2001). Securities Market in Japan 2001

Kendall, M. (1954). Note on bias in the estimation of autocorrelation. Biometrica 41, 403404.

29
Kyle, A. (1985). Continuous auctions and insider trading. Econometrica 53, 13151335.

Lehmann, B. & Modest, D. (1994). Trading and liquidity on the Tokyo Stock Exchange: a birds
eye view. Journal of Finance 49, 951-984.

Lesmond, D. (2002). Liquidity of emerging markets. Working paper, A.B. Freeman School of
Business. New Orleans: Tulane University.

Pastor, L. & Stambaugh, R. (2003). Liquidity risk and expected stock returns. Journal of
Political Economy 111, 642-685.

Pettengill, G., Sundaram, S. & Mathur, I. (1995). The conditional relation between beta and
returns. Journal of Financial and Quantitative Analysis 30(1), 101-116.

Sadka, R. (2003). Momentum, Liquidity Risk, and Limits to Arbitrage. Evanston, IL:
Northwestern University.

Scholes, M. & Williams, J. (1977). Estimating betas from non-synchronous data. Journal of

Silber, W.L. (1975). Thinness in capital markets: the case of the Tel Aviv Stock Exchange.
Journal of Financial and Quantitative Analysis 10, 129142.

30
Table 1 Sample Selection Process

This table reports the sample selection process. The sample period covers 1975-1999. The stocks
included in the sample must have valid observations of return and trading value data for more
than 200 days and have year-end prices greater than 100 yen, outliers with annual illiquidity at
the highest or lowest 1% tails of the distribution are eliminated.

Year Trading days Original stocks Stocks with
price>100
Stocks with trading
days>200
Final
sample
1975 286 985 915 577 565
1976 286 994 932 656 643
1977 285 1002 939 729 714
1978 286 1011 1009 757 742
1979 285 1022 1021 731 716
1980 285 1030 1027 684 670
1981 285 1041 1037 671 658
1982 286 1058 1047 759 744
1983 287 1077 1074 826 809
1984 285 1086 1084 878 860
1985 279 1109 1109 953 934
1986 274 1129 1129 963 944
1987 273 1152 1152 1027 1006
1988 249 1170 1170 1060 1039
1989 246 1184 1184 1057 1036
1990 246 1192 1192 1020 1000
1991 247 1223 1223 1031 1010
1992 246 1231 1230 1055 1034
1993 247 1236 1236 1115 1093
1994 249 1239 1239 1119 1097
1995 247 1252 1250 1170 1147
1996 245 1296 1293 1135 1112
1997 247 1332 1248 1150 1127
1998 245 1350 1278 1147 1124
1999 245 1350 1278 1122 1099

31
Table 2 Summary Statistics and Correlation of Stock Characteristics
This table presents the summary statistics and correlation matrix for the stock characteristics
used in the cross-sectional regression.
i
y
ILLM is the mean-adjusted illiquidity for stock i across
the days in year y.
i
y
CAP ln
is the logarithm for the market capitalization of stock i at the end of
year y.
1i
y
is the market beta for stock i in year y as estimated using the Fama-French (1992)
method with the Scholes and Williams (1977) adjustment.
i
y
STD
is the standard deviation of
return for stock i across days in year y and multiplied by 10
2
.
i
y
DP is the ratio of dividend per
share to share price for stock i in year y.
i
y
CP
is the ratio of earnings per share plus depreciation
to share price for stock i in year y.
i
y
BM is the ratio of book value to market value of equity for
stock i at the end of year y.
1i
y
PR is the past returns for the last 100 days for stock i in year y
calculated as the log ratio of its daily closing price.
2i
y
PR is the past returns for the rest of the
days for stock i in year y calculated as the log ratio of its daily closing price. The period covers
1975-1998. The stocks included in the sample must have valid observations of return, trading
value data for more than 200 days and have year-end prices greater than 100 yen; outliers with
annual illiquidity at the highest or lowest 1% tails of the distribution are eliminated.
Panel A: Summary Statistics
Variable Mean of
annual
means
Mean of annual
standard
deviation
Mean of
annual
skewness
Median of
annual
means
Minimum
of annual
means
Maximum
of annual
means
i
y
ILLM
0.093 0.110 1.853 0.081 0.018 0.195
i
y
CAP
206.947 428.704 5.562 213.487 58.419 417.027
1i
y

1.042 0.185 -0.008 1.048 0.959 1.143
i
y
STD
2.309 0.641 0.642 2.168 1.871 3.354
i
y
DP (%)
1.189 0.706 0.457 1.108 0.548 2.198
i
y
CP (%)
7.381 11.439 4.979 6.973 3.978 11.474
i
y
BM
0.514 0.254 -0.689 0.506 0.265 1.051
1i
y
PR (%)
-1.235 20.664 0.160 1.802 -46.750 36.658
2i
y
PR (%)
3.612 17.076 0.639 5.198 -27.634 24.335
Panel B: Correlation
Variable
i
y
CAP ln
1i
y

i
y
STD
i
y
DP
i
y
CP
i
y
BM
1i
y
PR
2i
y
PR
i
y
ILLM -0.581 0.358 0.217 0.007 -0.006 0.135 -0.093 0.073
i
y
CAP ln -0.453 -0.258 -0.174 -0.014 -0.226 0.054 -0.067
1i
y
0.231 -0.008 -0.027 0.058 -0.073 -0.058
i
y
STD -0.291 -0.154 0.078 -0.127 0.153
i
y
DP
0.269 0.374 -0.063 -0.016
i
y
CP 0.079 0.040 0.001
i
y
BM -0.330 0.037
1i
y
PR -0.020
32
Table 3 Cross-Sectional Illiquidity Effects on Stock Return
This table presents the means of the estimated coefficients from the monthly cross-sectional regression of stock return on the respective variables:
i
y y
i
y y
i
y y
i
y y y
i
m
PR k PR k k ILLM k k R
2
1 4
1
1 3
1
1 2 1 1 0
+ + + + = , and
i
y y
i
y y
i
y y
i
y y
i
y y
i
y y
i
y y y
i
m
DP k STD k CAP k PR k PR k k ILLM k k R
1 7 1 6 1 5
2
1 4
1
1 3
1
1 2 1 1 0
ln

+ + + + + + + = .
The coefficients are averaged for the whole sample period, one including and one excluding January, and for the two sub-periods, from 1976 to 1989 and from
1990 to 1999.
i
y
ILLM
1
is the ratio of stock illiquidity of stock i to market illiquidity in year y-1.
i
y 1
is the market beta for stock i in year y-1 as estimated using

the Fama-French (1992) method with the Scholes and Williams (1977) adjustment.
i
y
PR
1
1
is the past returns for the last 100 days for stock i in year y-1
calculated as the log ratio of its daily closing price.
i
y
PR
2
1
is the past return for the rest of the days for stock i in year y-1 calculated as the log ratio of its daily
closing price.
i
y
CAP
1
ln

is the logarithm of the market capitalization for stock i at the end of year y-1.
i
y
STD
1
is the standard deviation of return for stock i
across the days in year y-1.
i
y
DP
1
is the ratio of dividend per share to share price for stock i in year y-1. The monthly returns are from 1976 to 1999 and the
stock characteristics are from 1975 to 1998. The t-statistics are reported in parentheses.
Variable All months Excl. Jan 1976-1989 1990-1999 All months Excl. Jan 1976-1989 1990-1999
Constant 0.0082 0.0105 0.0120 0.0028 0.0060 -0.0017 0.0401 -0.0418
(1.552) (1.973)** (1.902)* (0.307) (0.254) (-0.068) (1.329) (-1.133)
i
y
ILLM
1

0.0014 0.0016 0.0022 0.0003 0.0014 0.0017 0.0019
0.0005
(2.493)** (2.758)*** (3.061)*** (0.314) (2.526)** (3.118)*** (2.923)*** (0.607)
i
y 1

-0.0015 -0.0064 0.0029 -0.0077 -0.0003 -0.0019 -0.0028
0.0031
(-0.311) (-1.331) (0.520) (-0.877) (-0.110) (-0.610) (-0.838) (0.543)
i
y
PR
1
1

-0.0060 -0.0053 -0.0076 -0.0037 -0.0030 -0.0018 -0.0037
-0.0022
(-1.676)* (-1.447) (-1.858)* (-0.578) (-0.937) (-0.540) (-0.939) (-0.388)
i
y
PR
2
1

0.0004 0.0003 0.0074 -0.0094 0.0017 0.0030 0.0073
-0.0060
(0.085) (0.060) (1.251) (-1.333) (0.391) (0.669) (0.226) (-0.901)
i
y
CAP
1
ln

0.0001 0.0005 -0.0008
0.0015
(0.166) (0.558) (-0.701) (1.089)
i
y
STD
1

-0.0014 -0.0029 -0.0011
-0.0017
(-0.053) (-2.287)** (-0.824) (-0.696)
i
y
DP
1

0.0718 0.0940 0.0904
0.0458
(1.051) (1.340) (0.985) (0.448)
Note: ***, ** and * denote significance at the 1, 5, and 10 percent levels, respectively.
33
Table 4 Cross-Sectional Illiquidity Effects on Stock Return

This table presents the means of the estimated coefficients from the monthly cross-sectional regression of stock return on the respective variables.
i
m y
i
y y
i
m y y
i
m
PR k k ILLM k k R
1 3
1
1 2 1 1 0
+ + + = .
i
y y
i
y y
i
y y
i
y y
i
m y
i
y y
i
m y y
i
m
CP k STD k CAP k BM k PR k k ILLM k k R
1 7 1 6 1 5 1 4 1 3
1
1 2 1 1 0
ln

+ + + + + + + = .
The coefficients are averaged for the whole sample period, one including and one excluding January, and for the two sub-periods, from 1976 to
1989 and from 1990 to 1999.
i
m
ILLM
1
is the mean adjusted illiquidity for stock i in month m-1.
i
y 1
is the market beta for stock i in year y-

1estimated using the Fama-French (1992) method with the Scholes and Williams (1977) adjustment.
i
m
PR
1
is the monthly return for stock i in
month m-1.
i
y
BM
1
is the book-to-market ratio for stock i at the end of year y-1.
i
y
CAP
1
ln

is the logarithm for the market capitalization of stock i at
the end of year y-1.
i
y
STD
1
is the standard deviation of return for stock i across the days in year y-1.
i
y
CP
1
is the ratio of earnings per share plus
depreciation to share price for stock i in year y-1. The monthly returns are from 1976 to 1999, and the stock characteristics are from 1975 to 1998.
The t-statistics are reported in parentheses.
Constant 0.0058 0.0067 0.0116 -0.0024 -0.0027 -0.0111 0.0630 -0.0940
(1.123) (1.293) (1.618) (-0.342) (-0.111) (-0.439) (1.911)* (-2.839)***
i
m
ILLM
1

0.0008 0.0008 0.0017 -0.0004 0.0005 0.0006 0.0012 -0.0004
(2.289)** (2.278)** (3.524)*** (-0.911) (1.594) (1.730)* (2.451)** (-0.840)
i
y 1

0.0014 -0.0022 0.0045 -0.0029 0.0034 0.0021 -0.0016 0.0102
(0.273) (-0.419) (0.719) (-0.336) (0.999) (0.634) (-0.388) (1.772*)
i
m
PR
1

-0.0595 -0.0533 -0.0335 -0.0956 -0.0659 -0.0600 -0.0406 -0.1010
(-6.557)*** (-5.871)*** (-3.056)*** (-6.439)*** (-7.675)*** (-6.989)*** (-3.904)*** (-7.227)***
i
y
BM
1

0.0067 0.0069 0.0090 0.0036
(2.852)*** (2.761)*** (2.558)** (1.280)
i
y
CAP
1
ln

0.0002 0.0006 -0.0019 0.0031
(0.221) (0.622) (-1.541) (2.645)***
i
y
STD
1

-0.0004 -0.0017 -0.0003 -0.0006
(-0.331) (-1.301) (-0.251) (-0.223)
i
y
CP
1

0.0041 0.0042 0.0025 0.0065
(1.210) (1.133) (0.497) (1.474)
34
Table 5 Cross-Sectional Illiquidity Effects on Stock Return in Up/Down Markets

The table presents the means of the estimated coefficients from the monthly cross-sectional regression of stock return on the respective variables.
i
m y
i
y y
i
m y y
i
m
PR k k ILLM k k R
1 3
1
1 2 1 1 0
+ + + = .
i
y y
i
y y
i
y y
i
y y
i
m y
i
y y
i
m y y
i
m
CP k STD k CAP k BM k PR k k ILLM k k R
1 7 1 6 1 5 1 4 1 3
1
1 2 1 1 0
ln

+ + + + + + + = .
The coefficients are averaged for the whole sample period, one including and one excluding January, and for the two sub-periods, from 1976 to
1989 and from 1990 to 1999, in up and down markets, respectively.
i
m
ILLM
1
is the mean adjusted illiquidity for stock i in month m-1.
i
y 1
is the
market beta for stock i in year y-1 estimated using the Fama-French (1992) method with the Scholes and Williams (1977) adjustment.
i
m
PR
1
is the
monthly return for stock i in month m-1.
i
y
BM
1
is the book-to-market ratio for stock i at the end of year y-1.
i
y
CAP
1
ln

is the logarithm for the
market capitalization of stock i at the end of year y-1.
i
y
STD
1
is the standard deviation of return for stock i across the days in year y-1.
i
y
CP
1
is the
ratio of earnings per share plus depreciation to share price for stock i in year y-1. The monthly returns are from 1976 to 199, and the stock
characteristics are from 1975 to 1998. The t-statistics are reported in parentheses.

Panel A: Up market
Constant 0.0174 0.0162 0.0251 0.0018 0.0838 0.0774 0.1157 0.0186
(2.648)*** (2.445)** (3.052)*** (0.169) (2.804)*** (2.429)** (3.171)*** (0.363)
i
m
ILLM
1

0.0010 0.0010 0.0022 -0.0012 0.0005 0.0006 0.0012 -0.0009
(2.315)** (1.955)* (3.688)*** (-1.489) (1.098) (1.217) (2.029)* (-1.115)
i
y 1

0.0236 0.0224 0.0096 0.0522
0.0117 0.0121 0.0014 0.0327
(3.693)*** (3.555)*** (1.330) (4.391)*** (2.687)*** (2.825)*** (0.320) (3.621)***
i
m
PR
1

-0.0791 -0.0712 -0.0489 -0.1406
-0.0858 -0.0788 -0.0550 -0.1484
(-5.965)*** (-5.323)*** (-3.502)*** (-5.243)*** (-6.803)*** (-6.219)*** (-4.220)*** (-5.768)***
i
y
BM
1

0.0050 0.0052 0.0049 0.0050
(1.580) (1.550) (1.184) (1.144)
i
y
CAP
1
ln

-0.0028 -0.0025 -0.0035 -0.0015
(-2.593)** (-2.117)** (-2.555)** (-0.827)
i
y
STD
1

0.0068 0.0048 0.0022 0.0162
(4.026)*** (2.894)*** (1.227) (4.861)***
i
y
CP
1

0.0000 -0.0005 0.0021 -0.0042
(0.002) (-0.104) (0.347) (-0.640)

35
Panel B: Down market

Variable All months Excl. Jan 1976-1989 1990-1999 All months Excl. Jan All months Excl. Jan
Constant -0.0098 -0.0053 -0.0144 -0.0059 -0.1179 -0.1216 -0.0388 -0.1862
(-1.237) (-0.655) (-1.085) (-0.622) (-3.166)*** (-3.153)*** (-0.602) (-4.645)***
i
m
ILLM
1

0.0005 0.0006 0.0008 0.0002 0.0005 0.0006 0.0010 0.0000
(0.960) (1.190) (0.939) (0.354) (1.037) (1.272) (1.368) (0.017)
i
y 1

-0.0282 -0.0329 -0.0053 -0.0480
-0.0078 -0.0103 -0.0073 -0.0082
(-3.668)*** (-4.237)*** (-0.441) (-5.156)*** (-1.520) (-2.001)** (-0.917) (-1.229)
i
m
PR
1

-0.0333 -0.0310 -0.0039 -0.0587
-0.0393 -0.0365 -0.0129 -0.0622
(-2.957)*** (-2.699)*** (-0.230) (-4.081)*** (-3.757)*** (-3.403)*** (-0.767) (-4.967)***
i
y
BM
1

0.0091 0.0090 0.0167 0.0025
(2.535)** (2.413)** (2.653)** (0.673)
i
y
CAP
1
ln

0.0042 0.0044 0.0012 0.0068
(3.124)*** (3.169)*** (0.494) (5.027)***
i
y
STD
1

-0.0101 -0.0098 -0.0052 -0.0143
(-5.795)*** (-5.393)*** (-2.427)** (-5.571)***
i
y
CP
1

0.0097 0.0100 0.0033 0.0152
(1.841)* (1.829)* (0.359) (2.631)**


36
Table 6 Time-Series Illiquidity Effects on the SZ-Portfolio Return
The excess monthly market return is regressed on monthly market illiquidity
0 1 1 2 3 1
ln ln
M f M M
m m m m m m m
R R g g ILL g ILLU g JAN aTM u

= + + + + +
,
where
M
m
R
is the monthly equally-weighted market return,
f
m
R is the one-month Gensaki monthly rate,
M
m
ILL ln
is the
expected monthly market illiquidity,
M
m
ILLU ln
is the unexpected monthly market illiquidity,
3 G
m m m
TM YL R =
is the term
yield premium, and
m
JAN
is a January dummy that equals 1 in January and zero otherwise. The test on 25 SZ portfolios is:
0 1 1 2 3 1
ln ln
p f p p M p M p p p
m m m m m m m
R R g g ILL g ILLU g JAN a TM u

= + + + + +
,
where
p
m
R
, p = 5, 10, 15, 20, and 25, are the equally weighted monthly returns on the SZ portfolio p. The period of estimation is
from 1976 to 1999.
Portfolio Constant
1
ln
M
m
ILL

ln
M
m
ILLU

m
JAN

1 m
TM

R
2

Market 0.016 0.005 -0.127 0.034 0.354
(1.43) (1.14) (-11.81)*** (3.38)*** (0.347)
Portfolio 5 0.012 0.003 -0.138 0.045 0.303
(0.82) (0.61) (-10.21)*** (3.53)*** (0.295)
Portfolio 10 0.021 0.008 -0.127 0.039 0.280
(1.46) (1.42) (-9.60)*** (3.11)*** (0.272)
Portfolio 15 0.017 0.006 -0.113 0.026 0.253
(1.32) (1.28) (-9.18)*** (2.22)** (0.245)
Portfolio 20 0.019 0.007 -0.101 0.019 0.221
(1.50) (1.41) (-8.44)*** (1.71)* (0.213)
Portfolio 25 0.025 0.008 -0.080 0.009 0.140
(1.85)* (1.57) (-6.36)*** (0.74) (0.131)
Market 0.015 0.005 -0.127 0.034 0.004 0.354
(1.33) (1.12) (-11.75)*** (3.38) *** (0.13) (0.345)
Portfolio 5 0.012 0.003 -0.138 0.045 -0.000 0.303
(0.77) (0.61) (-10.13)*** (3.52)*** (-0.01) (0.293)
Portfolio 10 0.022 0.008 -0.127 0.039 -0.010 0.280
(1.46) (1.44) (-9.50)*** (3.09)*** (-0.27) (0.269)
Portfolio 15 0.018 0.006 -0.113 0.026 -0.006 0.253
(1.30) (1.29) (-9.09)*** (2.21)** (-0.19) (0.242)
Portfolio 20 0.019 0.007 -0.101 0.019 -0.001 0.221
(1.39) (1.39) (-8.38)*** (1.71)* (-0.04) (0.210)
Portfolio 25 0.027 0.008 -0.079 0.009 -0.014 0.141
(1.88)* (1.61) (-6.27)*** (0.73) (-0.41) (0.128)

37
Table 7 Time-Series Illiquidity Effects on the SZ-Portfolio Return in Sub-periods
The excess monthly market return is regressed on monthly market illiquidity
0 1 1 2 3
ln ln
M f M M
m m m m m m
R R g g ILL g ILLU g JAN w
= + + + + ,
where
M
m
R
is the monthly equally-weighted market return,
f
m
R is the one-month Gensaki monthly rate,
M
m
ILL ln
is
the expected monthly market illiquidity,
M
m
ILLU ln
is the unexpected monthly market illiquidity, and
m
JAN
is a
January dummy that equals 1 in January and zero otherwise.
The test on 25 SZ portfolios is:
p
m m
p M
m
p M
m
p p f
m
p
m
w JAN g ILLU g ILL g g R R + + + + =
3 2 1 1 0
ln ln
,
where
p
m
R
, p = 5, 10, 15, 20, and 25, are the equally weighted monthly returns on the SZ portfolio p.
Panel A: 1976~1989
Portfolio Constant
1
ln
M
m
ILL

ln
M
m
ILLU

m
JAN

R
2

Market 0.009 -0.000 -0.092 0.032 0.371
(1.03) (-0.13) (-9.84)*** (4.11)*** (0.353)
Portfolio 5 -0.003 -0.005 -0.099 0.040 0.361
(-0.28) (-1.20) (-8.38)*** (3.94)*** (0.349)
Portfolio 10 0.012 0.002 -0.085 0.026 0.271
(1.09) (0.39) (-7.20)*** (2.56)** (0.257)
Portfolio 15 0.010 0.001 -0.075 0.022 0.209
(0.83) (0.26) (-6.12)*** (2.08)** (0.195)
Portfolio 20 0.017 0.003 -0.065 0.017 0.153
(1.35) (0.75) (-5.06)*** (1.58) (0.137)
Portfolio 25 0.021 0.005 -0.050 0.015 0.067
(1.35) (0.87) (-3.11)*** (1.08) (0.050)
Panel B: 1990~1999
Portfolio Constant
1
ln
M
m
ILL

ln
M
m
ILLU

m
JAN

R
2

Market 0.021 0.010 -0.159 0.042 0.371
(0.79) (1.04) (-7.39)*** (1.85)* (0.353)
Portfolio 5 0.029 0.014 -0.172 0.057 0.321
(0.91) (1.13) (-6.46)*** (2.04)** (0.302)
Portfolio 10 0.025 0.013 -0.167 0.064 0.327
(0.80) (1.10) (-6.48)*** (2.37)** (0.308)
Portfolio 15 0.023 0.012 -0.148 0.037 0.309
(0.81) (1.09) (-6.39)*** (1.50) (0.289)
Portfolio 20 0.017 0.009 -0.134 0.027 0.290
(0.66) (0.94) (-6.18)*** (1.18) (0.270)
Portfolio 25 0.027 0.011 -0.107 0.003 0.231
(1.10)* (1.14) (-5.24)*** (0.15) (0.209)
38
Table 8 Portfolio Properties
This table reports the properties of the oddnumbered portfolios of the equally weighted LQ portfolios formed each year during 19761999. The
market beta (
p 1
) and the liquidity betas (
p 2
,
p 3
and
p 4
) are computed using monthly return and illiquidity observations of each test portfolio
and the equally weighted market portfolio for the whole period as well as for the two sub-periods. The average illiquidity
( )
p
E C
, the variation of
illiquidity innovations
) (
p
c
, the average excess return ) (
f p
R R E , the variation of excess return ) (
p
R , the turnover TRN, the market
capitalization CAP and the BM ratio are computed for each portfolio as time-series averages of the respective monthly characteristics. The t-
statistics are reported in parenthesis.
Panel A: Whole Sample period
Portfolio
p 1

(100)
p 2

(100)
p 3

(100)
p 4

(100)
( )
p
E C

(%)
) (
p
c

(%)
) (
f p
R R E
(%)
) (
p
R
(%)
TRN
(%)
CAP
(B )
BM
1
70.68
(16.80)
0.0000
(0.59)
0.01
(0.15)
-0.01
(-7.89)
0.25 0.00 0.50 5.69 4.14 52830.59 0.38
3
86.30
(28.87)
0.0000
(1.30)
-0.01
(-0.13)
-0.02
(-7.93)
0.26 0.00 0.38 5.69 3.85 20804.57 0.41
5
90.25
(34.28)
0.0000
(0.52)
-0.00
(-0.04)
-0.05
(-9.85)
0.27 0.01 0.45 5.72 3.83 12642.74 0.45
7
96.19
(41.43)
0.0001
(0.60)
-0.03
(-0.35)
-0.08
(-9.58)
0.28 0.01 0.39 5.91 3.76 8007.29 0.50
9
99.92
(46.42)
0.0000
(0.20)
-0.02
(-0.25)
-0.11
(-9.51)
0.29 0.01 0.38 6.05 3.99 6121.48 0.50
11
105.08
(45.22)
0.0002
(0.85)
-0.05
(-0.59)
-0.15
(-7.99)
0.31 0.02 0.53 6.38 3.82 4876.13 0.50
13
108.52
(49.40)
0.0000
(0.10)
-0.02
(-0.28)
-0.23
(-8.44)
0.33 0.03 0.52 6.52 3.95 4121.10 0.53
15
108.95
(47.26)
0.0003
(0.45)
-0.06
(-0.76)
-0.35
(-8.22)
0.37 0.05 0.38 6.58 3.29 3575.85 0.56
17
110.63
(46.74)
0.0001
(0.17)
-0.07
(-0.87)
-0.53
(-8.76)
0.42 0.07 0.60 6.69 3.91 2707.37 0.56
19
107.72
(46.40)
0.0008
(0.68)
-0.10
(-1.23)
-0.59
(-7.00)
0.48 0.09 0.49 6.52 3.25 2281.27 0.55
21
111.29
(41.73)
0.0000
(0.01)
-0.08
(-0.88)
-0.89
(-7.44)
0.56 0.13 0.61 6.84 3.04 1848.15 0.59
23
112.96
(40.99)
-0.0012
(-0.64)
-0.07
(-0.79)
-1.02
(-7.15)
0.66 0.15 0.92 6.94 3.45 1426.22 0.57
25
113.51
(32.35)
-0.0034
(-1.00)
-0.12
(-1.30)
-1.78
(-7.05)
0.92 0.26 1.14 7.28 4.22 1521.53 0.56
39

Panel B: Sub-sample periods
Sub-period 1976~1989 1990~1999
Portfolio
p 1

(100)
p 2

(100)
p 3

(100)
p 4

(100)
) (
p
c E

(%)
p 1

(100)
p 2

(100)
p 3

(100)
p 4

(100)
) (
p
c E

(%)
1 82.50 0.0000 -0.03 -0.01 0.25 66.23 0.0000 0.05 -0.01 0.25
(8.82) (0.25) (-0.31) (-4.94) (15.72) (0.33) (0.48) (-6.16)
3 90.82 0.0000 -0.03 -0.02 0.26 83.89 0.0000 0.03 -0.02 0.26
(14.28) (0.00) (-0.41) (-5.07) (26.16) (0.82) (0.24) (-6.19)
5 93.69 -0.0001 0.00 -0.06 0.26 88.16 0.0000 0.03 -0.04 0.27
(17.77) (-1.09) (0.01) (-6.93) (28.79) (0.28) (0.22) (-6.91)
7 89.88 -0.0001 -0.04 -0.09 0.27 96.84 0.0001 0.02 -0.07 0.29
(19.15) (-0.92) (-0.66) (-6.07) (38.11) (0.33) (0.12) (-6.83)
9 96.95 -0.0003 -0.01 -0.12 0.28 99.62 0.0000 0.01 -0.11 0.31
(22.61) (-1.47) (-0.19) (-5.95) (40.83) (0.09) (0.10) (-6.74)
11 100.82 -0.0001 0.01 -0.14 0.29 104.96 0.0002 -0.02 -0.15 0.34
(21.67) (-0.55) (0.09) (-4.98) (41.45) (0.48) (-0.18) (-5.45)
13 107.16 0.0001 -0.09 -0.20 0.30 107.58 -0.0002 0.04 -0.23 0.37
(26.98) (0.32) (-1.22) (-5.11) (39.67) (-0.30) (0.27) (-5.81)
15 102.91 0.0003 -0.04 -0.35 0.32 109.40 0.0000 -0.03 -0.34 0.44
(22.77) (0.62) (-0.57) (-6.95) (42.06) (-0.01) (-0.22) (-5.10)
17 99.73 -0.0004 -0.08 -0.37 0.34 112.44 -0.0002 -0.03 -0.57 0.53
(22.23) (-0.70) (-1.09) (-6.40) (41.98) (-0.10) (-0.21) (-5.75)
19 101.16 0.0001 -0.07 -0.40 0.37 108.23 0.0004 -0.07 -0.63 0.63
(22.22) (0.06) (-0.98) (-4.35) (42.55) (0.22) (-0.50) (-4.70)
21 105.87 -0.0011 -0.06 -0.55 0.40 111.24 -0.0005 -0.03 -0.97 0.78
(22.34) (-1.21) (-0.78) (-5.33) (34.40) (-0.17) (-0.22) (-4.91)
23 110.84 -0.0006 -0.07 -0.62 0.44 111.99 -0.0023 -0.02 -1.11 0.97
(20.20) (-0.59) (-0.82) (-5.22) (36.97) (-0.67) (-0.16) (-4.71)
25 106.00 0.0000 -0.07 -0.84 0.52 113.94 -0.0059 -0.08 -2.01 1.48
(16.32) (0.02) (-0.79) (-4.90) (27.14) (-0.96) (-0.54) (-4.75)

40
Table 9 Test of Liquidity-adjusted CAPM on Portfolio Return
This table reports the estimated coefficients from cross-sectional regressions of the liquidity-adjusted CAPM for 25 equally weighted LQ portfolios using monthly data during
19761999 with an equally weighted market portfolio. We test the model by running excess monthly portfolio returns on the monthly normalized portfolio illiquidity, the market
beta, the liquidity betas and the net beta in the following special cases:
np p
m
f
m
p
m
c E R R E + + = ) ( ) (
,
p f
m
p
m
R R E
1 1
) ( + =
,
np p p
m
f
m
p
m
c E R R E + + + =
1 1
) ( ) (
,
and
np p f
m
p
m
R R E + + =
1 1
) (
, where
p p p p np 4 3 2 1
+ =
. The t-statistics are reported in the parentheses.
Panel A: 1976-1999
Line Constant ( )
p
E C

p 1

p 2

p 3

p 4

np
R
2
1 -0.4598 3.2035 0.0432 0.425
(-0.640) (3.882)*** (0.060)
2 -0.3955 0.9285 0.331
(-0.522) (1.147)
3 2.4967 -7.8724 -155.5927 155.4944 0.471
(3.126)*** (-5.193)*** (-6.228)*** (6.257)***
4 0.5853 -38.3027 38.0930 0.435
(0.876) (-3.443)*** (3.426)***
Panel B: 1976~1989
Line Constant ( )
p
E C

p 1

p 2

p 3

p 4

np
R
2

1 -1.713 4.998 1.398 0.349
(-1.392) (3.699)*** (1.306)
2 -2.100 3.281 0.257
(-1.521) (2.364)**
3 1.702 -5.081 -222.321 222.701 0.397
(1.434) (-2.810)*** (-4.710)*** (4.712)***
4 0.214 -107.677 108.313 0.361
(0.207) (-3.737)*** (3.729)***
Panel C: 1990~1999
Line Constant ( )
p
E C

p 1

p 2

p 3

p 4

np
R
2

1 0.690 1.148 -1.401 0.487
(0.784) (1.570) (-1.315)
2 0.640 -0.987 0.366
(0.659) (-0.829)
3 1.659 -1.799 -89.926 88.293 0.522
(1.853)* (-1.990)** (-2.426)** (2.403)**
4 1.025 -17.203 15.762 0.489
(1.226) (-1.052) (0.965)
41
Figure 1 Time-Series Pattern of Annual Market Illiquidity and the Nikkei 225

This figure shows the time-series pattern of the annual market illiquidity
M
y
ILL in Panel A and Nikkei
225 in Panel B during the sample period.
M
y
ILL is calculated as the cross-sectional average of annual
stock illiquidity for all the sample stocks during 1975~1999. The stocks included in the sample must have
valid observations of return and trading value data for more than 200 days in a year and have year-end
prices greater than 100 yen; outliers with annual illiquidity at the highest or lowest 1% tails of the
distribution are eliminated. Nikkei 225 is the index value at the end of each fiscal year.

Panel A: Annual Market Illiquidity
0
0.05
0.1
0.15
0.2
0.25
1
9
7
5
1
9
7
6
1
9
7
7
1
9
7
8
1
9
7
9
1
9
8
0
1
9
8
1
1
9
8
2
1
9
8
3
1
9
8
4
1
9
8
5
1
9
8
6
1
9
8
7
1
9
8
8
1
9
8
9
1
9
9
0
1
9
9
1
1
9
9
2
1
9
9
3
1
9
9
4
1
9
9
5
1
9
9
6
1
9
9
7
1
9
9
8
1
9
9
9
Year
I
L
L
m
y

Panel B: Nikkei 225

0
5000
10000
15000
20000
25000
30000
35000
19
75
19
76
19
77
19
78
19
79
19
80
19
81
19
82
19
83
19
84
19
85
19
86
19
87
19
88
19
89
19
90
19
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
Year
N
i
k
k
e
i

2
2
5

Il Liquidity and Stock Return 7

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Il Liquidity and Stock Return 7

Uploaded by

Copyright:

Available Formats

Illiquidity, Illiquidity Risk and Stock Returns: Evidence from Japan

= is the relative market

, should be positively associated with the

, should also have a

and decreases in the covariance between the securitys return and

as well as its illiquidity and market return, ) , ( cov

is the market beta for stock i in year y-1 as estimated using

is the market beta for stock i in year y-

You might also like