Professional Documents
Culture Documents
Impact of Liquidity On Market Capitalization Edelwise
Impact of Liquidity On Market Capitalization Edelwise
CHAPTER - 1
Most commonly, quarterly or annual advance tax payments draw liquidity out of the system
as a lot of liquid money gets locked with the government.
On the other hand, any large payouts by the government or higher corporate sector spending
can increase the liquidity in the system.
Funds inflows
A strong economic performance and the relative under-performance in the developed
countries attracted the attentions of many large global investors who were drawn towards
investing here ( FDI as well as portfolio investments).
This resulted in healthy capital inflows in the last few years. These capital inflows put a lot of
pressure on the liquidity management here as uncontrolled capital flows can result in rising
inflation, currency appreciation, loss of competitiveness and reduction in monetary control.
Tools to control liquidity
The RBI monitors the liquidity situation periodically and takes necessary steps to control the
situation from time to time. The RBI uses various direct and indirect policies to control the
shortterm and long-term liquidity position.
These are various instruments used by the RBI to control liquidity:
Cash reserve ratio: The RBI uses the cash reserve ratio (CRR) as a tool to control the medium
to long-term liquidity issues. An increase in the CRR results in an increase in the amount of
money that banks have to maintain with the RBI as a percentage of their deposits.
This reduces the overall liquid funds with the bank and hence reduces the overall liquidity.
Liquidity adjustment factor: The liquidity adjustment factor (LAF) was introduced a decade
ago as a part of financial reforms.
LAF helps in managing a shortterm liquidity situation resulting from the large and volatile
capital flows (inflows as well as outflows). Reverse repo rate: The RBI uses the reverse repo
rate for short-term liquidity management and to smoothen interest rates in the call/money
market.
The repos also help in keeping the interest rates in a predictable range, as provided by the
prevailing repo rate and reverse repo rate.
In times of excess visible liquidity, the call rates hover around the reverse repo rate, whereas
in times of tight liquidity, the call rate will hover around the repo rate.
Liquidity impacts inflation
An uncontrolled and unmanaged liquidity situation can have a severe impact on inflation,
rates of interest, STOCK MARKETS, and foreign exchange rates.
Since the conditions in the global markets and foreign fund flows are quite volatile, the job of
the RBI in controlling the liquidity condition has become more challenging.
The RBI has taken small steps in changing the monetary policy since the beginning of this
year. These steps have shown good results in terms of maintaining interest rates, liquidity and
GDP growth.
The inflation rate is still ruling high due to various factors and analysts believe that further
monetary actions from the RBI along with a good rainfall and base effect will moderate it in
the next couple of quarters.
DEFINITION
Market Capitalization
The total dollar market value of all of a company's outstanding shares. Market capitalization
is calculated by multiplying a company's shares outstanding by the current market price of
one share. The investment community uses this figure to determine a company's size, as
opposed to sales or total asset figures.
EXPLAINS 'Market Capitalization'
If a company has 35 million shares outstanding, each with a market value of $100, the
company's market capitalization is $3.5 billion (35,000,000 x $100 per share).
Company size is a basic determinant of asset allocation and risk-return parameters for stocks
and stock mutual funds. The term should not be confused with a company's "capitalization,"
which is a financial statement term that refers to the sum of a company's shareholders' equity
plus long-term debt.
The stocks of large, medium and small companies are referred to as large-cap, mid-cap, and
small-cap, respectively. Investment professionals differ on their exact definitions, but the
current approximate categories of market capitalization are:
Large Cap: $10 billion plus and include the companies with the largest market capitalization.
Mid Cap: $2 billion to $10 billion
Small Cap: Less than $2 billion
Market capitalization
The New York Stock Exchange onWall Street, the world's largest stock exchange per
total market capitalization of its listed companies.[1]
Market capitalization or market cap is the total dollar market value of the shares outstanding
of a publicly traded company; it is equal to the share price times the number of shares
outstanding.[2][3] As outstanding stock is bought and sold in public markets, capitalization
could be used as a proxy for the public opinion of a company's net worth and is a determining
factor in some forms of stock valuation. The investment community uses this figure to
determine a company's size, as opposed to sales or total asset figures.
The total capitalization of stock markets or economic regions may be compared to
other economic indicators. The total market capitalization of all publiclyTRADED
COMPANIES in the world was US$51.2 trillion in January 2007 [4] and rose as high as
US$57.5 trillion in May 2008[5] before dropping below US$50 trillion in August 2008 and
slightly above US$40 trillion in September 2008.[5]
Since 2009, when Bitcoin became the first decentralized cryptocurrency and numerous
cryptocurrencies (altcoins) have been created, the 'market cap' term has also come into
common use to describe the total dollar market value of the total amount of cryptocurrency in
circulation (available supply).[6][7] The term sometimes can refer to the estimated market value
of the total amount of cryptocurrency that will ever be in circulation (total supply).[8]
Calculation
capitalization, N is the number of shares outstanding, and P is the price per share.
For example, if some company has 4 million shares outstanding and the price per share is
$20, its market cap is then $80 million. If the price per share rises to $21, the market cap
becomes $84 million. If it drops to $19 per share, the market cap falls to $76 million.
Related measures
Market cap reflects only the equity value of a company. It is important to note that a firm's
choice of capital structure has a significant impact on how the total value of a company is
allocated between equity and debt. A more comprehensive measure is enterprise value (EV),
which gives effect to outstanding debt, preferred stock, and other factors. For insurance firms,
a value called the embedded value (EV) has been used.
CHAPTER - 2
Objective of Study
Scope and Rationale of Study
Methodology
Limitation of Study
CHAPTER - 3
10
such a state variable and securities differ in their return covariances with market liquidity,
then liquidity betas should be priced. One natural approach to investigating this question is to
follow the majority of the characteristic stock liquidity literature and estimate measures of
systemic liquidity by aggregating microstructure data, but this approach suffers from at least
two practical problems. First, the large volume of data per unit of time makes it difficult to
11
compute even the most basic aggregate liquidity measure. Second, even the longest time
series of transaction data is short compared to the availability of lower frequency data. Pastor
and Stambaugh (2002) devise a measure of the price reversal (resiliency) dimension of
market-wide liquidity utilizing daily returns over a long time period (1962-1999). Controlling
for the usual risk factors, they find a positive relationship between stock returns and the
covariance of return with their measure of market-wide liquidity. Using other dimensions of
liquidity such as depth and breadth to construct market-wide liquidity measures appears to
remain an unexplored area of research. This study asks three questions. First, are measures of
aggregate liquidity using depth and breadth priced, as Pastor and Stambaugh (2002) find for
their resiliency measure? In addition, is it possible to aggregate exposure to measures derived
using the three dimensions of liquidity to derive an estimate of the price of liquidity risk?
Second, is it possible for investors who do not care about return sensitivity to liquidity shocks
to invest in a portfolio that is sensitive to liquidity shocks but hedged against other common
sources of systematic risk, using only prior information, and earn a liquidity risk premium?
Finally, does the premium associated with high liquidity beta stocks survive after controlling
for market capitalization, the covariance of a stock's characteristic liquidity with changes in
aggregate liquidity, and the level of the stock's characteristic liquidity? To preview, I find that
estimates of the liquidity risk premium of approximately 2-5% per year are not sensitive to
the approach used for measuring market-wide liquidity, that a feasible investment strategy
earns approximately this return before transaction costs, and that the result survives after
controlling for the three alternatives listed above. I investigate the pricing of alternative
liquidity measures by first calculating two variations of each of three types of aggregate
liquidity measures based on market resiliency, depth, and breadth. The resiliency measure
relies on the principle that order flow induces greater return 4 reversals when market-wide
liquidity is low, as in Pastor and Stambaugh (2002). The second type of liquidity measure
attempts to capture the depth of the market and reflects the average price impact per unit of
trading volume. This measure is closely related to that used by Amihud (2002). The third type
reflects the breadth of the market and is derived from microstructure data on individual stock
bid/ask spreads. Although this measure is available for only part of the sample period (19832001) and is computationally intensive, it is important to understand how breadth measures
derived from transaction level data compare to the two alternative approaches that are
estimated using daily frequency data. The innovations in each time series of market-wide
liquidity measures are highly correlated with each other and reflect periods of especially low
measured liquidity corresponding to commonly accepted low liquidity periods in recent U.S.
12
history. I find that return covariance with shocks to aggregate liquidity is priced for all three
types of liquidity measures. The estimated risk premium is positive, however there is a
significant negative January seasonal in the liquidity premium that is not related to firm size.
A feasible investment strategy constructed to have positive exposure to systemic liquidity
shocks but hedged against other common risk factors earns positive returns on average and
negative returns when there is a shock to liquidity. The relationship between liquidity beta
and return remains after controlling for market capitalization, the covariance between stock
liquidity and market-wide liquidity, and the liquidity level of the individual stocks. In other
words, the higher return earned by stocks with large liquidity betas is not due to these stocks
being small, being themselves illiquid, or becoming particularly illiquid when there is a
market-wide liquidity shock. The rest of this paper is organized as follows. Section II
describes each of the measures of aggregate liquidity examined in this paper. Section III tests
whether liquidity risk as measured by return covariance with shocks to the aggregate liquidity
measures is priced. Section IV investigates the relationship between liquidity betas and the
covariance of individual stock liquidity with aggregate liquidity, stock's characteristic
liquidity, and firm size. Section V concludes.
which to estimate the regression (2.1), it is not the first or last month that the stock
appears on CRSP, and the share price at the end of the previous month is between $5 and
$1000. The market-wide liquidity measure is then constructed from the individual stock
measures by averaging all of the individual measures during the month and inflating by
the ratio of total market capitalization at the end of month t-1 to total market
capitalization at month 0. See Pastor and Stambaugh (2002) for a detailed discussion of
their measure. The rational for inflating the average liquidity measure by the ratio of
market capitalizations may not be clear. Pastor and Stambaugh (2002) argue that , PRV i t
can be viewed as "the liquidity cost, in terms of return reversal, of trading $1 million of
stock i, averaged across all stocks." Since $1 million was a relatively larger trade in the
1960s than in the 1990s, the simple average coefficient will fall through time. Inflating
the coefficient adjusts for this condition. One drawback of the PRV liquidity measure is
the use of dollar volume since equal size trades may have different impact due to
differences in, for example, the number of shares outstanding, differences in float, and
differences in the number and types of shareholders. One possible alternative is to
substitute turnover (dollar volume divided by end of previous month market
capitalization) for dollar volume although, as Pastor and Stambaugh point out, this is
similar to simply value weighting their measure. However even this measure would miss
variation in return impact of order flow due to, for example, differences in float. A
second alternative, unexplored in previous studies, is to substitute turnover scaled by
average daily turnover during the previous month for dollar volume in equation (2.1).
This measure of order flow will capture any unusual volume at the expense of ease of
interpretation but without the need to inflate the average coefficient by total market
capitalization. Figure 1a plots the time series of scaled , PRV i t and Figure 1b plots ,
mPRV i t . The series are very similar with large negative liquidity levels in months
where liquidity is generally considered to be low including October of 1987 (the crash,
which is the largest negative value in both series), November of 1973 (the Arab oil
embargo, 2nd and 12th largest negative levels respectively), September of 1998 (the
Russian debt and LTCM crisis, 4th and 2nd), and October of 1997 (the height of the
Asian financial crisis, 13th and 9th). The overall correlation between the two series is
0.713 . Table 1 reports that both series display significant autocorrelation (0.21 and 0.16
for PRV and mPRV respectively).
Amihud (2002) estimates a liquidity measure based on price impact. Kyle (1985) argues
that spreads are an increasing function of the probability of facing an informed trader,
and since the market-maker cannot distinguish between order flow from informed traders
and order flow from noise traders, she sets prices that are an increasing function of the
order imbalance that may indicate informed trading. This implies an inverse relationship
between price impact and liquidity. Alternatively, price impact measures for a particular
stock may be large for reasons unrelated to asymmetric information issues or liquidity.
For example, when there is a news 3 Omitting October of 1987 from both series reduces
the estimated correlation to 0.67. Although influential, omitting this observation does not
have a large impact on the reported correlations of the levels or innovations of the
liquidity measures. 8 release which impacts firm value but about which there is little
disagreement, price change can be large and volume small resulting in a large estimated
price impact. My use of the price impact measure follows the spirit of Amihud but is
different from the individual stock or characteristic liquidity approach. When marketwide liquidity is low, price concessions required from Grossman-Miller market makers
are larger per unit of volume than when market-wide liquidity is high. By averaging price
impact measures across all stocks the idiosyncratic effects should diversify leaving only
systematic liquidity. Whether this is measurable in practice is an empirical issue.
The measure is defined as the negative of the daily average so that large negative values
signify 'low liquidity' consistent with the interpretation of the PRV and mPRV measures.
The marketwide measure is the simple average of the individual stock measures. The
resulting time series is then inflated by the ratio of total market capitalization at the end
of month t-1 to total market capitalization at the end of month 0. The same criticisms that
apply to the use of dollar volume in the PRV measure also apply here; therefore I also
examine a modified version of the price impact measure: ( ) , , , 1 , , 1 1 Dn mPI idt i t d
n idt r D STO = = + (2.4) The aggregate measure is the simple average of the
individual stock measures and is not rescaled by market capitalization. 9 Figure 1c plots
the time series of scaled , PI i t and Figure 1d plots , mPI i t . Scaled , PI i t is highly
serially correlated (first order serial correlation, 1=0.89) with periods of especially low
liquidity in the early 1970s and again in 1999-2000. , mPI i t also shows similar periods
of illiquidity although not as severe as , PI i t . The correlation between the two
measures is 0.54. The correlation matrix for all of the aggregate liquidity measures is
shown in Table 1. All of the measures are positively correlated with each other and we
15
can reject the null that each correlation is zero although the correlation between scaled ,
PRV i t and scaled , PI i t is only 0.09. C. Measures Based on Bid/Ask Spread
Amihud and Mendelson (1986), Chordia, Roll, and Subrahmanyam (2000), Hasbrouck
and Seppi (2001), Huberman and Halka (2001), Jones (2001), Baker and Stein (2002)
and many others examine the bid-ask spread as a measure of the characteristic liquidity
of individual stocks. An investor wishing to trade immediately may always sell (buy) at
the quoted bid (ask) price that includes a concession (premium) for immediate execution.
Therefore the spread between the bid and the ask prices, which is the sum of the
concession and premium, divided by the midpoint of the spread, is a natural measure of
liquidity. Using ISSM data from 1983-1992 and TAC data from 1993-2001, I calculate
aggregate liquidity measures using all NYSE/AMEX stocks as follows.4 First, define
RQSi,d,t as the daily average relative quoted spread for stock i on day d in month t.
RQSi,d,t is the average of every best bid and offer (BBO) eligible quote from the open
until just prior to the market close divided by the quote midpoint. RESi,d,t is defined as
the daily average relative effective spread and is the average of the absolute value of the
difference between each transaction price and the midpoint of the most recent quote,
which is at least five seconds prior to the trade, divided by the quote midpoint. The
aggregate liquidity level during month t is: Dt i,d,t 1 d=1 t 1 RQS D = = RQS Nt
t i Nt (2.5) 4 I am grateful to M. Nimalendran for providing the quote and effective
spread data. 10 Dt i,d,t 1 d=1 t 1 RES D = = RES Nt t i Nt (2.6) where Nt is the
number of firms in month t and Dt is the number of days in month t. Increasing spreads
are associated with decreasing liquidity, therefore the leading negative sign is added so
that smaller values of are associated with lower liquidity, consistent with the other
measures. Figure 1e plots the time series of RQS t and Figure 1f plots RES t for the
period 1983-2001. Both plots show an upward trend reflecting the falling quoted and
effective spreads during the period. There are also large negative changes in the liquidity
measure in October of 1987 and September of 1998. Consistent with the positive time
trend, both series are strongly positively serially correlated. D. Innovations in Aggregate
Liquidity For asset pricing purposes it is the covariance of asset returns with innovations
in the aggregate liquidity measure that is important. This is in contrast to characteristic
liquidity where the difference in liquidity levels implies differences in transaction costs
that must be compensated with expected return. To estimate innovations from levels, I
calculate the first difference of each liquidity measure as: ( ) , ,1 1 1 . = =
Nt X X XX t t it it i t M N (2.7) where X Mt = (mt-1 /m0), the ratio of total market
16
capitalization at time t-1 and time zero, for X = PRV and X=PI and Mt X =1 for all
others. I then regress t on its lag as well as the lagged value of the scaled series: 1 1 ,
1 . X X XX X t t j it t a b cM u = + + + (2.8) Thus, the predicted change
depends on the lag level and the lag change. The innovation in aggregate liquidity is X t
u . To ease comparison of results between liquidity measures in later sections, I rescale X
t u so that the standard deviation of the innovations is of the same order of magnitude for
each measure. 11 100 0.10 10 100 100 PRV PRV mPRV mPRV PI PI tt t t t t mPI mPI
RQS RQS RES RES t tt t t t Lu L u L u L uL u L u = = = = = = (2.9) Table 2 shows
that (2.8) yields innovations that are serially uncorrelated for all measures in the full
sample and in both subperiods. If the three dimensions of market-wide liquidity are
related, then we might expect the innovations to be correlated. Panel A of Table 2 shows
the correlation matrix of the innovations over the full sample period. All of the
innovations are significantly positively correlated, both in the full sample and in each
subperiod. The only exception is the correlation between the PRV and PI measures in the
second subperiod, which is a statistically insignificant 0.06. The significant correlations
for the later subperiod in Panel B between the breadth measures estimated from
microstructure data and the resiliency and depth measures estimated from daily data are
particularly encouraging. If market-wide liquidity measures can be estimated using low
frequency data then the cost of estimation is greatly reduced and the measures can be
estimated over much longer time periods and for markets for which transaction level data
is not available. Figure 2 plots the time series of the innovations in aggregate liquidity.
All show large negative values on similar dates, October of 1987 in particular, although
the magnitude of these shocks varies. Panel B of Table 2 shows us that the LmPI measure
is highly correlated with each of the microstructure based measures with an estimated
correlation of 0.74 with each. The LPRV, LmPRV, and LPI measures are also
significantly correlated with LRQS and LRES. The correlation matrices in Table 2
suggest a similarity among proxies but do not by themselves imply that market liquidity
is a priced state variable. E. Empirical Features of the Liquidity Measures Pastor and
Stambaugh (2002) describe a "flight to quality" effect when their measure of market
liquidity is low. Months in which liquidity is exceptionally low tend to be months in
which stock returns and bond returns move in opposite directions. Table 3 reports the
correlation between the value-weighted CRSP index of NYSE-AMEX stocks and three
fixed income variables: minus the change in the rate on one-month Treasury bills, the
return on the thirty year 12 government bond, and the return on a portfolio of long term
17
corporate bonds5 . Over the full sample period 1962-2001, the correlation between minus
the change in the rate on one-month Treasury bills and the market return is near zero and
between the market return and the bond returns is positive. In months of low liquidity,
defined as a liquidity shock more than two standard deviations below the mean, the
correlation between the market return and both minus the treasury bill return and the
government bond return is negative, regardless of the liquidity measure used to identify
months of low liquidity. The correlation between the corporate bond return and the
market return is near zero when low liquidity is defined using LPRV or LmPRV and
negative when using LPI or LmPI. Panel B reports similar figures for the 1983-2001
period and include the spread based (breadth) measures of liquidity. The results are very
similar to those in Panel A. Also shown in Table 3 is the correlation between the market
return and the equally weighted average percentage change in monthly dollar volume for
NYSE-AMEX stocks. The unconditional correlation between volume changes and
market returns is positive; however, regardless of the measure used to identify months of
low liquidity, when liquidity is low, market returns and changes in volume are negatively
correlated. Table 4 reports correlations between innovations in each liquidity measure
and the valueweighted CRSP index, the equal-weighted CRSP index, and the Fama
French factors SMB and HML. Each measure is positively correlated with both CRSP
indices; however the correlation is driven by months in which the market falls. For
example, the LPRV measure has a correlation of 0.29 with the value-weighted index, but
the correlation is 0.02 in months in which the index return is positive and 0.44 when
negative. Each of the measures is positively correlated with SMB and negatively
correlated with HML. When liquidity is low, large stocks outperform small stocks and
value outperforms growth. The correlations are larger in magnitude and significance for
the price impact and spread based measures than for the reversal-based measures. It is
remarkable is that the six liquidity measures that address the three separate dimensions of
liquidity appear so similar. Months of low liquidity are months in which stock market
returns fall, large stocks outperform small stocks, and value outperforms growth. The
next step is to 5 The corporate bond data is from Ibbotson Associates. 13 examine the
pricing implications of a stock's return covariance with each of these measures,
controlling for other commonly used sources of risk. III. The Liquidity Risk Premium
This section investigates whether a stock's expected return is related to the covariance of
its return with innovations in each of the liquidity measures after controlling for other
variables that have been found to be important in asset pricing. To accomplish this I use a
18
portfolio-based approach where the portfolios are formed on the basis of predicted
sensitivity to liquidity shocks. Each month, the universe of available stocks is sorted into
ten portfolios by predicted liquidity beta and held for one month. The portfolio returns
are linked through time to form a single return series for each decile portfolio. These post
formation returns are then regressed on return based factors that are commonly used in
empirical asset pricing studies. To the extent that the intercepts are different from zero,
liquidity sensitivity explains a component of returns not captured by exposure to other
factors. Specifically, for each month t, I regress the excess stock return on the liquidity
innovation, LX, in a regression that also includes the Fama and French (1993) factors: 0 ,
M S H XX it i i t i t i t i t t r RMRF SMB HML L =+ + + + + (3.1) where X Lt
is the innovation calculated using one of the six methods described above. For every
month t between December 1965 and December 2001, the regression is run for every
stock whose end of month price at month t-1 is between $5 and $1000 and which has
valid return data in at least 36 months between t-1 and t-60. Although it seems natural to
use the estimated regression coefficient l X i to sort stocks into portfolios, it is well
known that sorting on regression coefficients in this manner is problematic, especially
when the standard errors of the regression coefficients are large. This is of particular
concern for the regression (3.1) since the standard errors of l X i for individual stocks
are very large and therefore sorting on l X i , in effect, leads to sorting on estimation
errors. 14 To mitigate this problem I use a Bayesian approach to form the estimates of X
i then sort into portfolios based on these estimates. The Bayesian estimates of X i are
only used to sort stocks into portfolios, all point estimates reported in the tables are the
result of classical econometric techniques. Specifically, I estimate (3.1) for every stock at
month t, then treat each estimate of the vector = [0 , M, S , H, X] as a draw from
a multivariate normal distribution with estimated covariance . The Bayesian estimate of
, lbi , is then estimated as: l ( ) ( ) ( ) l ( ) = + + 1 1 1 2 2 ' ' i i b XX
XX i s s i (3.2) where ( ) 1 2 ' s i X X is the estimated covariance matrix of l i . The
Bayesian estimate of is a weighted average of the OLS estimate of i and the average
of across all stocks at time t where the weight on i is the inverse of the covariance
matrix of i. This estimator "shrinks" the estimate of the coefficient vector for each stock
towards the population average with the amount of shrinkage an inverse function of the
precision of the estimate for the individual stock. Pastor and Stambaugh (PS) use a
similar approach to infer that their liquidity measure is priced, although they use a
different method of sorting stocks into portfolios. They model the time variation in X i
19
explicitly using the full sample up to time t to estimate the parameters. I prefer my
method for sorting into portfolios for three reasons. First, although PS model time
variation in the liquidity beta, they assume the other factor loadings and the parameters
of the model for time variation in liquidity beta do not change over a sample period of up
to 35 years. Second, the model of time variation proposed by PS captures very little of
the variation in liquidity betas as measured by R2 , and the coefficients are unstable
through time. Third, the insample loadings on innovations in liquidity are not as the
model predicts. Appendix A discusses the PS methodology, elaborates on the above
points, and compares it to the method used in this paper. A. Asset Pricing Tests I test for
the existence of a liquidity risk premium in two ways. First I estimate the abnormal
return to each predicted liquidity beta sorted decile portfolio using the three-factor model
of Fama and French and examine the intercepts. The difference in abnormal return
between the 15 extreme deciles provides information about a component of expected
returns not captured by the three-factor model. The second test uses the information in all
ten decile return series to estimate the liquidity risk premium directly. 1. Fama French
Alphas The time series returns for each liquidity beta decile portfolio are regressed on the
three Fama-French factors that are commonly used in empirical asset pricing studies6 .
To the extent that the regression intercepts, or alphas, differ from zero, X explains a
component of expected returns not captured by exposure to the other factors. Table 5,
panel A shows the alphas from Fama-French regressions of the excess return on each
equal-weighted decile portfolio for each liquidity measure. The intercepts are generally
negative for the portfolio of those stocks with the least sensitivity to liquidity shocks and
increasing as we move to portfolios with a greater sensitivity. The spread in intercepts
between the most and least sensitive portfolios is positive, ranging from 0.32 to 3.11
percent per year. None of the spreads differs significantly from zero. Panel B reports
results from value-weighting the decile portfolios. The pattern in intercepts in Panel A
repeats although the pattern is less apparent. Five of the six spreads in intercept are
positive and none of the spreads are statistically significant. Since there is a size
component to liquidity, smaller firms are concentrated in the extreme deciles, it seems
appropriate to check for a January seasonal in order to verify that the "liquidity
premium" is not a rediscovery of the size/January effect7 . To isolate the seasonal effect
in the intercepts I estimate (3.1) and (3.2) separately for Januaries and for all other
months. The portfolio returns are the same as those above. Table 6 reports the results.
When the portfolios are equal-weighted, the spread in annualized intercepts for non20
January months ranges from 0.93 to 2.67 percent. Few of the individual portfolio
intercepts differ significantly from zero, but both price reversal measures and one of the
two price impact 6 For each liquidity measure, the ten equations are stacked and
estimated using GMM. The point estimates will be identical to those from equation by
equation OLS but the standard errors are corrected for autocorrelation and conditional
heteroskedasticity. This method makes tests of cross-equation restrictions simple. 7
Unreported tests fail to identify a January seasonal in any of the six market-wide
liquidity measures. 16 measures' portfolio intercept spreads are statistically significant.
Although the spread in intercepts for the microstructure based measures are of similar
magnitude (annualized spreads of 2.65 and 1.87), the shorter time series results in an
inability to reject that the spreads are zero. Interestingly, the spreads in January are
negative for five of the six measures. Only the RQS measure, which is based on quoted
spreads and does not include transaction prices, has a positive spread in January. Panel B
of Table 6 reports value-weighted results similar to the equalweighted results. Five of six
measures have positive intercept spreads in non-January months although none are
statistically significant. Only the RQS measure is associated with a positive intercept
spread in January. The difference in intercept spreads in Januaries vs. Non-Januaries
presents an interesting puzzle. Even after including the SMB factor, the model of Fama
and French does not fully explain the January effect. Since small firms are concentrated
in the lowest and highest liquidity beta sorted deciles, we might be tempted to argue that
the inability of the three factor model to explain the January effect in small stock returns
is confounding any liquidity effects. However, as I will show later, the negative January
seasonal in the liquidity premium is not confined to small stocks but exists across all size
quintiles. 2. Direct Estimation of the Liquidity Risk Premium The previous section infers
the existence of a liquidity risk premium from the spread in abnormal returns between
the highest and lowest decile of predicted liquidity sensitivity. It is also possible to
estimate the liquidity risk premium directly using information from all ten portfolios. For
each measure of aggregate liquidity X, define the time series regression: 0 X X t t tt r BF
L e =+ + + (3.3) where rt is a 10x1 vector of excess returns on the decile portfolios,
Ft is a 3x1 vector containing the realizations of the Fama-French factors RMRF, SMB,
and HML, B is a 10x3 matrix of factor loadings, X is a 10x1 vector of liquidity betas,
and X Lt is the innovation in aggregate liquidity measure X. Assume the portfolios are
priced by: ( ) , F XX Er B t = + (3.4) 17 where E ( )i denotes the unconditional
expectation, and i is the risk premium for factor i. Since F are returns on portfolios, let
21
the results. When the Fama-French factor-neutral portfolios are equalweighted, the
liquidity trading strategy earns from 14 to 42 basis points (1.68% and 5.04% annualized)
per month. The profits from portfolios formed on the four non-microstructure data based
measures that are available for the full sample period are all statistically significant. Five
of the six portfolios have negative returns in months when liquidity is low, and all six
have smaller returns in low liquidity months than in other months. The last three columns
drop Januaries from the sample with little effect. When the feasible Fama-French factorneutral portfolios are value weighted, all six earn positive returns, four of six are negative
in low liquidity months and five of six earn lower returns on average when liquidity is
low than in other months. Statistical significance is generally lower than when portfolios
are equal-weighted. C. Combining Liquidity Measures If the three dimensions of
liquidity are related, then it should be possible to combine the information contained in
each variable to improve our estimates of liquidity risk factor exposures and liquidity
premia. To this end I form a new set of decile portfolios based on the sum of the 19
portfolio assignments from each of the six individual liquidity measures. For each stock
that has a portfolio assignment for each of the six measures (four prior to 1987), I sum
the portfolio numbers to which each is assigned then sort this summary statistic into
deciles. I then repeat the experiment outlined in Section A using the summary deciles as
test assets. Table 9 presents the results. The spread in annualized intercepts is a
statistically significant 2.64% when portfolios are equal-weighted and 2.11% when
portfolios are value weighted. We are unable to reject that the spread in intercepts is zero
when portfolios are value-weighted. Both point estimates are within the range of those
estimated in Table 5 using the individual liquidity measures. An inspection of the tstatistics associated with the individual intercepts shows that the estimation error
associated with the intercepts is much smaller with the aggregate measure than with
individual measures. The annualized spread in intercepts is a statistically significant
3.13% when portfolios are equalweighted and Januaries are omitted and 2.65% when
value-weighted. Direct estimation of the liquidity risk premia using the method of Table
8 is difficult since the portfolios have been formed based on information contained in all
six liquidity measures. However, it is possible to estimate the returns to an investment
strategy long decile 10 and short decile 1 formed using the aggregate measure and
hedged against any exposure to the FamaFrench risk factors. The results to such a
strategy are reported in Panel B of Table 9. The strategy using equal-weighted portfolios
earns a statistically significant return of 43 basis points per month (5.16% annualized),
23
larger than the return earned by any of the six individual liquidity measure based
strategies in Table 8. When the portfolio is value-weighted, the investment strategy earns
a statistically significant 50 basis points per month, again larger than that earned by the
feasible investment strategy based on any of the six individual liquidity measures. If a
"low liquidity" month is defined as a month in which all available liquidity measures are
greater than two standard deviations below their mean, the equal weighted strategy earns
an average 242 basis points in low liquidity months and the value weighted strategy an
average of +33 basis points per month. Although the average monthly return for the
value weighted strategy is positive, it is the average of only three observations. The
median return of these three observations is -324 basis points and the average return is
below that of the other months. If 20 "low liquidity" is defined as any liquidity measure
being more than two standard deviations below is average, then both equal and valueweighted strategies earn negative returns in low liquidity months and significantly
positive returns in other months. IV. Individual Stock Liquidity The previous sections ask
whether stocks whose return covaries with each of several marketwide liquidity measures
earn higher returns. This section asks whether the covariance between stock return and
market-wide liquidity shocks (liquidity return beta) is a proxy for the covariance between
changes in a stock's characteristic liquidity and market-wide liquidity shocks (liquidity
spread betas), or a proxy for the stock's characteristic liquidity level, or simply a proxy
for market capitalization. A. Liquidity Spread Beta Amihud and Mendelson (1986)
develop a model in which expected returns are an increasing function of the bid/ask
spread. Because illiquid stocks are more expensive to trade, investors must be
compensated with higher expected returns. The question examined here is similar to that
of Amihud and Mendelson, stocks that become relatively more illiquid when aggregate
liquidity falls are particularly unattractive members of a portfolio. To see why this might
be the case recall that aggregate liquidity, regardless of which measure is used, and
market returns covary strongly when returns are negative, therefore a mutual fund
manager selling to meet redemption requests or an investor selling to meet a margin call
would incur a particularly large transaction cost associated with liquidity for holding
stocks which become particularly illiquid when market liquidity falls. This was of
particular importance during the Long Term Capital Management (LTCM) experience of
1998. (See Lowenstein (2000) for a description of events surrounding the takeover of
LTCM by a consortium orchestrated by the New York Federal Reserve.) The hedge fund
was very highly levered in often very illiquid securities. When the Russian debt crisis
24
precipitated a fall in market liquidity, the value of the fund's portfolio value dropped
triggering a need to liquidate positions to meet margin calls. The anticipation of LTCM's
need to liquidate further eroded the value of the fund's positions. Prior to 1998, did
LTCM earn a liquidity premium for 21 holding illiquid securities9 , holding securities
whose returns were sensitive to liquidity shocks, or both? To investigate whether the
covariance of individual stock liquidity with aggregate liquidity is priced, each month I
regress changes in individual stock liquidity measures on lag changes in individual stock
liquidity measures, the lag level of the individual stock's liquidity, and shocks to
aggregate liquidity: , ,1 ,1 , . X X X XX it it it t it a b c dL u = + + + +
(4.1) where , X i t : The change in characteristic liquidity of stock i from month t-1
to t using liquidity measure X. , 1 X i t : The characteristic liquidity of stock i using
liquidity measure X. X Lt : The shock to market-wide liquidity at time t using liquidity
measure X. and X corresponds to one of the six liquidity measures: PRV, mPRV, PI, mPI,
RQS, or RES. The coefficient vector =[a b c d] is adjusted using the Bayesian technique
described in section III. I then sort the stocks into deciles by the Bayesian estimate of the
coefficient on aggregate liquidity shocks, LX. I refer to this coefficient as a "liquidity
spread beta" and refer to the liquidity beta discussed in the first three sections as a
"liquidity return beta". The resulting decile portfolios are then regressed on the FamaFrench risk factors and the difference in annualized intercepts between deciles 10 and 1
is examined for evidence of variation in alpha across the decile portfolios in a manner
similar to Tables 5 and 6. Table 10 reports the results. For brevity the individual decile
intercepts have been omitted and only the annualized difference in the extreme deciles is
reported. The first three columns represent the difference in alphas for equal-weighted
portfolios for the full sample and for the sample split by January vs. Non-January.
Examining the Non-January column there is some evidence, particularly for the mPI
(modified price impact) measure and the RQS (relative quoted 9 This argument applies
to long positions. Many of LTCM's trading strategies involved the simultaneous purchase
of long and short positions in similar securities whose prices were expected to converge.
22 spread) measure that stocks which become relatively more illiquid when the market
becomes more illiquid, conditional on the previous months change in liquidity and
liquidity level, earn negative abnormal returns, precisely the opposite of what we might
expect. These results must be interpreted with caution. Sorting on the sensitivity of
changes in individual stock liquidity to changes in aggregate liquidity is also a sort on
market capitalization. For example, the ratio of the average market capitalization of
25
decile 1 to decile 10 for the mPI measure reported in the first six columns of Table 10 is
4.20 and market capitalization decreases nearly monotonically between decile 1 and
decile 10. The size relative for the RQS measure is much larger (38.70) and market
capitalization is also monotonically decreasing across deciles10. B. Characteristic
Liquidity To investigate the role of the level of characteristic liquidity, each month I sort
all stocks with available data into ten portfolios by the average of their characteristic
liquidity over months t-2 to t-4. I skip month t-1 to avoid issues associated with bid/ask
bounce. The deciles are then linked through time, regressed on the Fama-French risk
factors, and the difference in intercepts between the extreme portfolios examined. The
results are reported in the right six columns of Table 10. There is some evidence that the
most liquid stocks in decile 10 earn higher risk adjusted returns than the less liquid stocks
in decile 1, especially when the mPI (modified price impact ) or the RES (relative
effective spread) measures are used as a measure of liquidity. Again, we must interpret
the results with caution since sorting on characteristic liquidity is similar to a sort on
market capitalization with the most illiquid stocks in decile 1 also being the smallest
stocks. Table 10 provides weak evidence that more liquid stocks have higher risk
adjusted returns than illiquid stocks. Although this result contradicts Amihud and
Mendelson (1986), it is consistent with Eleswarapu and Reinganum (1993). In particular,
Eleswarapu and Reinganum find stocks that are particularly illiquid as measured by
relative quoted bid/ask spread earn higher size-adjusted average returns in January and
lower size-adjusted returns in non-Januaries, consistent with Table 10. 10 This is in
contrast to the liquidity return beta deciles from Section III that have smaller firms
concentrated in the lower and higher deciles with larger firms in the middle deciles. 23 C.
Two-Way Sorts To disentangle liquidity effects from pure size effects, I first sort all
stocks into size quintiles using NYSE derived breakpoints, then within each size quintile
I sort into quintiles by liquidity measure, either liquidity return beta, liquidity spread
beta, or characteristic liquidity, to form 25 portfolios. These portfolios are linked through
time and regressed on the Fama-French risk factors and, as before, the difference in
alphas is examined for evidence of abnormal return associated with liquidity while
controlling for market capitalization. I also examine the relationship between liquidity
return betas and both liquidity spread betas and characteristic liquidity by first sorting
into quintiles by either spread beta or characteristic liquidity and then sorting on liquidity
return beta within each quintile. In the interest of brevity, the results reported in Table 11
are only for equal-weighted portfolios and only report the difference in alpha by control
26
variable quintile. Examining the spread in alphas from liquidity return betas while
controlling for market capitalization, there is no obvious relationship between size
quintile and spread in abnormal return. The quintile with the largest spread in intercepts
varies by measure with the quintile of the largest stocks having the biggest spread in
intercepts (in non-Januaries) for three of six measures. The most surprising result comes
from the January months. The negative spread in abnormal returns is not confined to the
smallest stocks, indeed the biggest negative spread in intercepts is always in one of the
biggest three of the five quintiles. While the classic January effect is closely related to
market capitalization, the relative underperformance of high liquidity return beta stocks
in January is not limited to smaller firms. Consistent with Table 10, when using the
reversal-based liquidity measures PRV and mPRV, there is no evidence that liquidity
spread betas or characteristic liquidity is priced after controlling for market
capitalization. The price impact measure mPI's significant negative spread in alphas
when sorted by liquidity spread beta continues when controlling for size although the
effect is larger for the smaller quintiles. The similar results using the spread based
measure RQS also persists across size deciles. The significant positive spread in alphas
between a portfolio of stocks with low characteristic liquidity as measured by mPI and
high characteristic liquidity is largest among the smallest stocks and virtually disappears
by the largest quintile 24 because there is little variability in the measure for larger
stocks. The significant positive spread using the spread-based RQS as a measure of
characteristic liquidity almost disappears after controlling for size. To verify that liquidity
return betas are not proxies for characteristic liquidity or liquidity spread betas, I use the
latter variables as control variables and sort stocks into quintiles by the control variable
before sorting by liquidity return beta. For each liquidity measure, sorting first by
liquidity spread beta or characteristic liquidity then by liquidity return beta has little
effect. The point estimates by quintile are similar to those of the univariate sort in Table 5
and Table 6. In summary, there is weak evidence that liquidity spread beta and
characteristic liquidity are priced, but the risk premia do not have the expected sign.
Stocks with a high covariance between changes in individual liquidity and shocks to
market-wide liquidity (after controlling for the lagged level and lagged change in
characteristic liquidity) earn lower risk adjusted returns than those with a low covariance.
There is also weak evidence that stocks which are relatively liquid earn higher average
risk adjusted returns than stocks which are relatively illiquid in non-January months.
There is strong evidence that illiquid stocks (as measured by characteristic liquidity) do
27
earn larger abnormal returns in January, but the effect is concentrated in the smallest two
size quintiles. Most important, there does not appear to be any relationship between the
higher abnormal returns earned by high liquidity return beta stocks and either liquidity
spread betas or the stock's characteristic liquidity. V. Conclusion Numerous authors have
found that illiquid stocks earn higher average returns, presumably to compensate for the
higher costs of transacting. This paper addresses the related but separate issue of marketwide liquidity. If the ability to transact a given volume with minimal price concession
varies through time and there exist cross sectional differences in a stocks return
covariance with measures of market-wide liquidity, then this covariance should carry
with it a higher expected return. Measures of market-wide liquidity designed to capture
three related, but separate dimensions of liquidity yield similar results, namely that
covariance with market-wide measures 25 of liquidity carries a risk premium. This risk
premium varies from approximately 2 to 5% per year depending on the measure. This
estimate of the liquidity risk premium associated with covariance with market-wide
liquidity shocks is much lower than that estimated by Pastor and Stambaugh (2002) but
is still statistically significant. Four of the six measures used do not require transaction
level data and thus can be constructed over much long time spans and in markets for
which transaction level data is unavailable. Covariance of return with market-wide
liquidity does not appear related to the covariance between changes in a stocks
characteristic liquidity and market-wide liquidity. In other words, stocks whose price is
sensitive to market-wide liquidity shocks do not necessarily become themselves more
illiquid when market liquidity is low. This is consistent with fund managers selling more
liquid stocks to meet margin calls or redemption requests when market-wide liquidity is
low. Several surprising results provide avenues for future research. First, there is a strong
January seasonal in the liquidity risk premia but not in the liquidity measures themselves.
Although high liquidity beta stocks earn higher risk-adjusted returns on average, they
earn lower returns in January. This January effect in liquidity is not related to firm size,
rather it exists across all size quintiles. Second, liquidity spread betas, or the covariance
between changes in a stock's own characteristic liquidity and shocks to market-wide
liquidity, is associated with lower average returns. In other words, stocks that become
particularly illiquid when markets become illiquid earn below average returns, a
counterintuitive result. The negative average return associated with liquidity spread betas
is particularly surprising because one might have expected that an individual stock's
liquidity is particularly important when the market is less liquid overall. If aggregate
28
liquidity falls when market returns are large and negative, then investors who must sell
will sell those investments that have the best individual characteristic liquidity so as to
minimize the transaction costs associated with liquidity. This reasoning is consistent with
the financing of margin investors by uninformed outside lenders who react to losses by
cutting lending (see Shleifer and Vishny (1997) and Xiong (1999)). This would result in
a high covariance between the return of stocks whose individual liquidity is high and the
aggregate liquidity state variable. Why then should investors demand a 26 premium for
holding these stocks? Why do investors not require a premium for holding stocks whose
characteristic liquidity worsens when aggregate liquidity falls? These questions provide
an interesting avenue for further research.
29
CHAPTER - 4
30
31
CHAPTER - 5
32
33
CHAPTER - 6
34
35
Bibliography
36
Annexure
1. Amihud, Yakov, 2002, "Illiquidity and Stock Returns: Cross-Section and Time-Series
Effects," Journal of Financial Markets, 5, 31-56.
2. Amihud, Yakov and Haim Mendelson, 1986, "Asset Pricing and the Bid-Ask Spread,"
Journal of Financial Economics, 17 223-249.
3. Andrews, D. W. K., 1991, Heteroskedasticity and Autocorrelation Consistent
Covariance Matrix Estimation, Econometrica, 59, 817-858.
4. Baker, Malcolm and Jeremy C. Stein, 2002, "Market Liquidity as a Sentiment
Indicator," Harvard Institute of Economic Research Discussion Paper Number 1977.
5. Brennan, Michael J. and Avanidhar Subrahmanyam, 1996, "Market Microstructure
and Asset Pricing: On the Compensation for Illiquidity in Stock Returns," Journal of
Financial Economics, 41, 441-464.
6. Campbell, John Y., Sanford J. Grossman and Jiang Wang, 1993, "Trading Volume and
Serial Correlation in Stock Returns," The Quarterly Journal of Economics, 108, 905939.
7. Chordia, Tarun, Richard Roll, and Avanidhar Subrahmanyam, 2000, "Commonality in
Liquidity," Journal of Financial Economics, 56, 3-28.
8. Cochrane, John H., 2001, Asset Pricing, Princeton University Press, Princeton N.J.
Datar, Vinay T., Narayan Y. Naik, and Robert Radcliffe, 1998, "Liquidity and Stock
Returns: An Alternative Test" Journal of Financial Markets, 1, 203-219.
9. Eleswarapu, Venkat R., and Marc R. Reinganum, 1993, "The Seasonal Behavior of
the Liquidity Risk Premium in Asset Pricing", Journal of Financial Economics, 34,
373-386.
10. Fama, Eugene F. and Kenneth R. French, 1993, "Common Risk Factors in the Returns
on Stocks and Bonds," Journal of Financial Economics, 33, 3-56.
11. Fama, Eugene F. and James D. MacBeth, 1973, "Risk, Return, and Equilibrium:
Empirical Tests," Journal of Political Economy, 81, 607-636.
37
12. Fernandez, Frank A., 1999, "Liquidity Risk: New Approaches to Measurement and
Monitoring," Securities Industry Association Working Paper. Grossman, Sanford J.
and Merton H. Miller, 1988, "Liquidity and Market Structure," The Journal of
Finance, 43, 617-633.
13. Hansen, L. P., 1982, Large Sample Properties of Generalized Method of Moments
Estimators, Econometrica, 50, 1029-1054. 28
14. Hasbrouck, Joel and Duane J. Seppi, 2001m, "Common Factors in Prices, Order
Flows, and Liquidity," Journal of Financial Economics, 59, 383-411.
15. Huberman, Gur and Dominika Halka, 2001, "Systematic Liquidity," The Journal of
Financial Research, 24, 161-178.
16. Jones, Charles M., 2001, "A Century of Stock Market Liquidity and Trading Costs,"
Columbia University Working Paper. Kyle, Albert S., 1985, "Continuous Auctions
and Insider Trading," Econometrica, 53 1315-1335.
17. Lowenstein, Roger, 2000, When Genius Failed, New York, Random House. Pastor,
Lubos and Robert F. Stambaugh, 2002, "Liquidity Risk and Expected Stock Returns,"
Journal of Political Economy, forthcoming. Shleifer, Andrei and Robert Vishny, 1997,
"Limits of Arbitrage," The Journal of Finance, 52, 35-55.
18. Shanken, Jay, 1990, "Intertemporal Asset Pricing: An Empirical Investigation,"
Journal of Econometrics, 45, 99-120.
19. Shanken, Jay, 1992, "On the Estimation of Beta Pricing Models", Review of Financial
Studies 5, 1-24. Xiong, Wei, 2001, "Convergence Trading with Wealth Effects: An
Amplification Mechanism in Financial Markets," Journal of Financial Economics, 62
247-292.
38
39