Money and Liquidity in Financial Markets1
Kjell G. Nyborg
Per Östberg
University of Zürich,
University of Zürich
Swiss Finance Institute,
and Swiss Finance Institute
and CEPR
June 2011
First draft: November 2009
1 We
thank Zexi Wang for research assistance. We are also grateful for comments from seminar
participants at the Bank for International Settlements, Bank of Finland, Copenhagen Business
School, ESSEC, European Central Bank, ISCTE, Norges Bank, Norwegian School of Management, Stockholm School of Economics, Warwick Business School, and the Universities of Aberdeen, Alabama, Amsterdam, Gothenburg, Porto, South Carolina, and Zürich as well as participants at: European Finance Association Annual Meetings, Frankfurt, Germany, August 2010;
European Central Bank conference on The Role of Financial Market Liquidity in Periods of Turbulence: Theory, Empirical Evidence and Implications for Policy, Frankfurt, Germany, October
2010; FRIAS-CEPR conference on Information, Liquidity and Trust in Incomplete Financial
Markets, Freiburg, Germany, October 2010; Swiss Finance Institute Annual Meetings, Zürich
November 2011; FINRISK Research Day, Gerzensee, Switzerland, June 2011; WU Gutmann
Symposium on Liquidity and Asset Management, Vienna, Austria, June 2011. We also thank
Philipp Halbherr, Henrik Hasseltoft, Wolfgang Lemke, Loriana Pellizon, and members of the
external board of the Department of Banking and Finance, University of Zürich, for comments.
This research has been supported by grant 189355 of the Norwegian Research Council. We also
thank NCCR-FINRISK for financial support.
Nyborg: Department of Banking and Finance, University of Zürich, Plattenstrasse 14, 8032
Zürich, Switzerland. Email: [email protected]. Östberg: Department of Banking
and Finance, University of Zürich, Plattenstrasse 14, 8032 Zürich, Switzerland. Email:
[email protected].
Abstract
Money and Liquidity in Financial Markets
We argue that there is a connection between the interbank market for liquidity and the
broader financial markets, which has its basis in demand for liquidity by banks. Tightness
in the interbank market for liquidity leads banks to engage in what we term “liquidity
pull-back,” which involves selling financial assets either by banks directly or by levered
investors. In particular, tighter interbank markets should lead to relatively more volume in
more liquid assets. Empirical tests on the stock market support this. While our data covers
part of the recent crisis period, our results are not driven by the crisis. Our general point
is that money matters in financial markets. Different financial assets have different degrees
of moneyness (liquidity) and, as a result, there are systematic cross-sectional variations in
trading activity as the price of liquidity, or the level of tightness, in the interbank market
fluctuates. Our tests control for a variety of factors, including market-wide uncertainty
which also affects volume cross-sectionally.
Keywords: money, liquidity, interbank and financial markets, liquidity pull-back, volume,
portfolio rebalancing
JEL: G12, G21, G11, E41, E44, E51
– All the rivers run into the sea; yet the sea is not full: unto the place from whence the
rivers come, thither they return again.
Ecclesiastes 1:7
1
Introduction
We study the connection between the interbank market for liquidity and the broader financial markets. That such a connection exists is suggested, for example, by the experience
of the recent financial crisis, which saw both a breakdown in the interbank market and
a collapse in the prices of financial assets. However, our focus is not on the crisis, but
rather on the day-to-day interaction between the interbank market for liquidity and financial market activity. The paper makes three contributions. First, it advances what we
call the liquidity pull-back hypothesis, which addresses how demand for liquidity by banks
impacts on financial market activity. Second, we test and find supportive evidence for
this hypothesis. Third, as a byproduct of controlling for market-wide uncertainty in the
testing of the liquidity pull-back hypothesis, we document a relation between uncertainty,
stock liquidity, and volume, which may help shed light on how agents rebalance portfolios
in response to fluctuations in market-wide uncertainty. In broad terms, the paper bridges
two different concepts of liquidity; namely the finance idea that liquidity is a property of
an asset and the central banking and monetary economics concept of liquidity simply as
high powered money.
There is evidence in the extant literature that financial markets are affected by monetary phenomena. For example, returns in bond and equity markets appear to be influenced
by monetary shocks (Fleming and Remolona 1997, Fair 2002, Piazzesi 2005) and fund flows
(Edelen and Warner 2001, Boyer and Zheng 2009, Goetzmann and Massa 2002), as are
measures of liquidity in these markets (Chordia, Sarkar, and Subrahmanyam, 2005). However, we are not aware of research that explicitly documents a link between the interbank
market and the stock markets, as we do in this paper.
Our line of reasoning has its basis in a money and banking perspective on financial
market activity. Banks need liquidity, or central bank money, to satisfy reserve require1
ments, allow depositor withdrawals, etc. The central bank determines the quantity of
liquidity via its operations and then the interbank market (re)allocates it. However, if
the price of liquidity in the interbank market is high, alternative sources of liquidity may
be more attractive. Banks that have exhausted credit limits, must look for alternative
sources. But to paraphrase Friedman (1970), “One bank can increase its money balances
only by persuading another one to decrease its balances.”1 And as emphasized by Tobin
(1980), “The nominal supply of money is something to which the economy must adapt,
not a variable that adapts itself to the economy – unless the policy authorities want it to.”
So what alternatives to the interbank market are there?
Banks have, in fact, several alternatives. They can go to the discount window, but this
is expensive and a last resort. They can try to induce more deposits, but this is unlikely to
be effective within a short time span. Rather, the alternative that we wish to emphasize
here is pulling back liquidity from the financial markets. This can be done in several ways.
The most obvious one is through selling financial assets.2 This could occur through the
mechanism of a banks’ internal liquidity management system feeding into trading desks’
limits. Alternatively, a bank can increase margins to investors, which in turn may lead to
asset sales as investors seek to meet margin requirements. Increasing haircuts in repos has
a similar effect. These actions do not increase the quantity of liquidity in the system, but
they can increase the selling bank’s liquidity balances, as long as the buying counterparty
banks with another bank. One can think of liquidity pull-back as a bank dipping its ladle
into the “ocean” of financial assets, recovering for itself liquidity granted to a counterparty
some time in the past and stored all the while in the financial asset that now is being sold.
Thus, we argue that there is a connection between the interbank market for liquidity and
the broader financial markets arising from (the possibility of) liquidity pull-back.
Liquidity pull-back trading is arguably most likely to occur if the interbank market is
not allocatively efficient. The crisis is an example of it being so; volume in the interbank
market fell (Cassola, Holthausen, and Lo Duca, 2008) while central banks around the
1
The original Friedman quote is: “One man can reduce his nominal money balances only by persuading
someone else to increase his. The community as a whole cannot in general spend more than it receives.”
2
Kashyap and Stein (2000) document that many banks hold substantial amounts of securities.
2
world injected vast amounts of liquidity to counteract banks’ unwillingness to lend to each
other. In addition, Bindseil, Nyborg, and Strebulaev (2009) find evidence that there is a
degree of allocational inefficiency in the interbank market even during what we think of
as times of normalcy, and Fecht, Nyborg, and Rocholl (2010) find evidence that interbank
liquidity networks, which are intended to overcome imperfections in the interbank market,
are not always effective. We therefore expect increased tightening in the interbank market
to give rise to an increased volume of liquidity pull-back trading.
To test the liquidity pull-back hypothesis, we focus on the cross-sectional implications.
In particular, the liquidity pull-back effect on volume should be felt differentially across
assets, depending on their degree of liquidity in the financial economics sense of the word.
By definition, trade in a highly liquid asset involves lower price impact, or transaction costs,
on average than an equivalent trade in a less liquid asset (Black 1971, Kyle 1985). The
implication, and our central hypothesis, is thus that increased tightness in the interbank
market for liquidity is associated with an increase in the volume of more liquid assets
relative to that of less liquid assets.
While our main focus is on volume, the liquidity pull-back hypothesis also has implications with respect to returns. A higher price of liquidity, ceteris paribus, should be
associated with offsetting drops in asset prices. This is so as to equalize, insofar as possible, the cost of acquiring liquidity directly in the interbank market versus acquiring it
indirectly by engaging in liquidity pull-back. Put in a different way, selling a financial asset
can be thought of as an act of converting low powered money (financial assets) into higher
powered money (liquidity). When the price of liquidity goes up, the price of conversion
also rises and asset prices therefore fall. However, with respect to this price effect, we
expect no differential impact across assets of different liquidity levels, since in equilibrium
there should be an equalization across assets of the marginal costs of converting into higher
powered money.
We test the implications of the liquidity pull-back hypothesis on the CRSP universe
of stocks using the three month Libor-OIS and TED spreads as measures of tightness in
the interbank market.3 The Libor-OIS spread may be a more precise measure of the state
3
Libor is London Interbank Offered Rate, OIS is overnight index swap, TED spread is 3 month Libor
3
of the interbank market, since it is the difference between two interbank rates, rather
than an interbank and a treasury rate. However, we have a longer time series of the TED
spread. The in-sample correlation between the two spreads is 0.92. While it is possible that
liquidity-pull back is prevalent in the Treasury security or broader fixed income markets,
testing our hypotheses using the CRSP universe of stocks offers several advantages. First,
the data is reliable and of high quality. Second, there are thousands of stocks, with a wide
range of liquidity levels. Third, there is homogeneity in trading infrastructure.
The empirical design involves forming portfolios of stocks based on the Amihud (2002)
price impact measure of liquidity (or illiquidity). Our predictions are confirmed in the
data: First, the market share of volume of highly liquid stocks is increasing in either
spread; second, the difference in portfolio returns between high and low spread days is
negative, but the size of the difference does not depend on the degree of liquidity of the
portfolio.
In testing the cross-sectional volume implications of the liquidity pull-back hypothesis,
we control for market-wide uncertainty, as measured by the VIX, and other factors.4
Thus, we control for the alternative hypothesis that our findings are the result of portfolio
rebalancing by investors who seek to reduce equity exposures as volatility increases (see,
e.g., Ritter (1988), Hau and Rey (2004), Calvet, Campbell, and Sodini (2009) for evidence
on portfolio rebalancing.) We find that the market share of volume of more liquid stocks
is also increasing in that part of the Libor-OIS spread that cannot be explained by the
VIX. This is strong evidence in support of the liquidity pull-back hypothesis.
The market share of volume of more liquid stocks is also increasing in the VIX itself
as well as that part of it that cannot be explained by the Libor-OIS spread. This is
supportive of a “flight to safety” effect, whereby increased volatility leads to a relative
increase in the sale of more liquid stocks as the price impact per unit is smaller for these
stocks (as outlined above). In short, our evidence suggests that liquidity pull-back and
portfolio rebalancing exist side-by-side.
While we use it as a general measure of tightness in the interbank market, the Liborless the 3 month treasury bill rate.
4
For information about the VIX, see http://www.cboe.com/micro/VIX/vixintro.aspx.
4
OIS spread can be viewed more specifically as a measure of the price of liquidity. A “Libor
transaction” gives the borrower a fixed quantity of liquidity for a fixed period of time at
a fixed rate. The alternative (in the unsecured end of the market) is borrowing overnight
and hedging the interest rate risk using the OIS. But this entails quantity risk; a bank
cannot be sure that it will get the desired quantity of liquidity every day over the next
three months, say.5 While the spread thus captures the extra cost of having the liquidity
for sure, we believe it may also reflect quantity constraints. The drop in interbank activity
during the crisis (especially at the longer end) supports this view. In addition, Gorton
and Metrick (2009) find that high Libor-OIS spreads coincide with increased haircuts in
repos. From a theoretical perspective, standard Akerlof (1970) adverse selection reasoning
yields a positive relation between the price of liquidity and unsatiated demand.6 Thus,
the Libor-OIS spread may be viewed as a general measure of interbank tightness as well
as a specific measure of the price of liquidity.
The empirical analysis in this paper is motivated by the theoretical framework sketched
above. A less bank-centric line of reasoning that is consistent with our findings is as
follows: Higher spreads imply higher funding costs for investors, as banks pass on their
own borrowing costs. As a result, stock prices fall. In turn, this leads to margin calls and
portfolio rebalancing, as already described. This still implies a connection between the
interbank market for liquidity and the broader financial markets, but the role of banks
is deemphasized. Our perspective differs from that of Grossman and Miller (1988) and
Brunnermeier and Pedersen (2009), where selling pressure originates in the asset market
rather than the money market. Brunnermeier and Pedersen in particular emphasize how
a liquidity event in the asset market can lead to dramatic falls in prices, as providers of
funding liquidity may tighten margins too much if they are uninformed about the cause
of the liquidity event. In our framework, a severe decline in stock prices could potentially
5
There is also some interest rate risk, since a bank’s overnight borrowing costs will not necessarily be
equal to or perfectly correlated with the floating rate that inputs into the OIS contract, for example since
overnight rates may vary across banks.
6
Adverse selection may also lead to credit rationing along the lines of Jaffee and Russell (1976) or
Stiglitz and Weiss (1981).
5
be triggered by unrest in the interbank market, for example arising from extreme adverse
selection in that market. Both of these perspectives may well be relevant for understanding
the collapse in the stock markets during the crisis.
However, it bears emphasis that the liquidity pull-back hypothesis is not about the
crisis. Indeed, we find stronger evidence for the presence of liquidity pull-back over the
pre-crisis period than over the crisis period. While at first glance this may seem surprising,
it may well be the result of loose monetary policy during the crisis. TAF (term auction
facility) and other tools were introduced to help banks get liquidity, thus reducing the
need for banks to engage in liquidity pull-back.7
This paper is related to several other literatures. Liquidity pull-back is a potential contributor to the commonality in liquidity found by Chordia, Subrahmanyam, and
Roll (2000), Hasbrouck and Seppi (2001), and Huberman and Halka (2001). We also
contribute to the literature on trading volume (e.g. Ying 1966, Karpoff 1987, Lo and
Wang 2000) by documenting that the Libor-OIS and TED spreads are associated with
cross-sectional variations in volume.
The rest of this paper is organized as follows. Section 2 describes the data, provides
descriptive statistics, and defines the volume measures that we subsequently use in our
tests. Section 3 studies the liquidity pull-back hypothesis by examining volume crosssectionally, by liquidity, on extreme spread days. Section 4 runs time-series regressions,
controlling for market volatility and other factors. This section also contains a variety
of robustness analyses. Section 5 concludes. An appendix provides an overview of key
variables.
2
Data, variable definitions, and descriptive statistics
2.1
Money market spreads
For both the Libor-OIS and TED spreads, we focus on the widely used three month
maturity versions of these spreads. The Libor-OIS spread thus refers to the difference
7
For information on TAF, see http://www.federalreserve.gov/monetarypolicy/taf.htm.
6
between 3 month USD Libor and the 3 month USD overnight index swap rate.8 Libor is
collected daily over the period January 2, 1986 to December 31, 2008. This yields a total
of 5,817 Libor observations. The OIS data are also daily, but only cover the period from
December 4, 2001 to November 11, 2008. Thus, we have 1,716 daily observations of the
Libor-OIS spread.
The TED spread is the difference between the 3 month USD Libor and the 3 month
T-Bill rate.9 The T-Bill data is available for the entire period for which we have Libor
data. But on some days we only have one or the other rate, for example because bank
holidays in the UK may fall on different days than US holidays. We have 5,648 daily
observations of the TED spread.
The Libor-OIS and TED spreads are highly correlated; the correlation in our sample
is 0.92. For this paper, the advantage of using the Libor-OIS spread is that it is a pure
interbank spread and therefore captures the price of liquidity better than the TED spread.
On the other hand, we have a longer time series of the TED spread. Thus, we make use
of both spreads in our tests.
Graphs of the two spreads over their respective sample periods are in Figures 1 and
2. As seen in Figure 1, the two spreads reacted similarly to the recent crisis, increasing
sharply in August 2007 and again in the wake of the Lehman Brothers bankruptcy in
September 2008. Figure 2 confirms that the TED spread tends to spike in times of turmoil
when financial markets often are depressed, such as when the stock market crashed in
October 1987. This is suggestive, but nothing more, of there being a connection between
the state of the money markets and the broader financial markets.10 Our focus, however, is
not on crisis periods, but on the relation between the day-to-day fluctuations in the price
of liquidity in the interbank market and the relative volume of high versus low liquidity
8
Libor is downloaded from http://www.bbalibor.com. The OIS rate is obtained from Reuters.
The T-Bill data is obtained from http://www.federalreserve.gov/releases/h15/data.htm.
10
For example, the recent crisis saw not only high spreads, but also declining asset prices. From 1
9
August 2007 to 31 March 2009, the S&P 500 fell by 46% and, in local currency, the DAX, NIKKEI, and
the FTSE 100 fell by 45%, 52%, and 37%, respectively. Similarly, in the crash of October 1987 (Black
Monday), Roll (1988) notes that 19 out of 23 markets declined by more than 20 percent that month.
Ferson and Harvey (1993) find evidence of a relation between the TED spread and stock returns.
7
stocks. The graphs show that even outside of crises, there is a fair amount of day-to-day
variation in the spreads.
Table 1 reports summary statistics of the Libor-OIS and TED spreads. The table
also breaks out the data into non-crisis versus crisis periods (see the notes to Figure 2
for a list of crisis dates) and monthly high and low spread days. For the Libor-OIS and
TED spreads, respectively, the unconditional means (standard deviations) are 25.80 basis
points [bps] (41.74 bps) and 55.80 bps (43.85 bps), respectively. For non-crisis periods, the
corresponding numbers are 11.06 bps (3.59 bps) and 47.95 bps (31.82 bps). This confirms
that there is substantial volatility in these spreads.
This can also be seen in the means of the high versus low spread days, which is also
reported in Table 1. For each spread (Libor-OIS and TED), the high spread sample
contains one observation for each month, namely the equally-weighted average of the
month’s two highest spreads.11 The low spread series are created in the analogous way.
Because there is only one observation per month in each series, this does not put more
weight on crisis periods than non-crisis periods. For the Libor-OIS spread, the mean spread
on high and low days are 33.03 bps and 21.69 bps, respectively. For the TED spread, the
corresponding means are are 70.11 bps and 43.14 bps. Thus, there is substantial intramonth variation in the spreads.12
2.2
Stock market data
The stock market data is from the Center for Research in Security Prices (CRSP). We
consider stocks that are listed on the NYSE, NASDAQ and AMEX over the period 1986 to
11
If there are several days within a month with the highest value for the spread, then the observation
for that month is simply the value of this highest spread. If there is only one day within a month with
the highest spread but several with the second highest, the value on the high spread day is averaged with
the common (single) value on the second highest spread days.
12
Chordia, Roll and Subrahmanyam (2001) document that there are day of the week and holiday effects
in aggregate dollar volume. However, we have seen no evidence of such effects in the Libor-OIS and TED
spreads (details are available upon request). So our findings below which relate variations in these spreads
to the relative volume of stocks with different levels of liquidity are not driven by systematic day of the
week effects.
8
2008 with CRSP share codes 10 or 11. Thus, we exclude ADRs, closed-end funds, REITs,
and shares of firms incorporated outside of the US. We also exclude financials by removing
firms with Standard Industrial Classification (SIC) codes between 6000 and 6999. Stocks
that meet any of the following criteria at any time within a given year are also removed
for that year: the stock price exceeds $999 or the firm changes ticker, cusip, or exchange.
This leaves us with an average of 4,506 individual stocks per day. The change in exchange
removal criterion is used to minimize the impact that any market microstructure changes
might have on the stock. In a later section, we examine the robustness of our findings by
removing all NASDAQ stocks.
2.3
Measuring stock liquidity
To test the liquidity pull-back hypothesis, we need a measure of the liquidity of stocks.
Goyenko, Holden and Trczinka (2009) show that low-frequency measures of liquidity based
on daily data perform well as compared with high frequency intraday measures. Low
frequency measures have the advantage of being computable for a larger range of stocks
over longer time horizons than measures based on intraday data. Among low frequency
price impact measures, Goyenko et al. finds that the best performer is Amihud’s (2002)
ILLIQ, which we therefore employ in this paper.
We measure ILLIQ on a monthly basis. For stock i and month j, ILLIQ is defined as:
ILLIQij = Average
t∈monthj
|rt |
Volume t
(1)
where, |rt | and Volumet are the absolute value of the stock’s rate of return and dollar
volume, respectively, on day t. Thus, a large ILLIQ is indicative of a highly illiquid
stock, since price impact per unit of volume is large. The average in (1) is taken across
observations for stock i in month j for days when recorded volume is positive. We exclude
observations with no recorded closing price on either day t or day t − 1 and a zero return
on day t, as this is highly suggestive of stale prices and spurious volume.13 This exclusion
13
In the CRSP database, all days without a recorded closing price are given a “closing” price of the
bid/ask average and this situation is flagged by the use of negative numbers. There are occurrences in
9
results in a loss of less than 2% of monthly ILLIQ observations.
Throughout the paper, we work with portfolios sorted by ILLIQ on a monthly basis.
Each month stocks are sorted into 10 groups based on their ILLIQ for that month. Group
1 consists of the month’s 10% most liquid stocks, i.e., the stocks with the lowest ILLIQ,
Group 2 consists of the next 10% most liquid stocks, etc. Table 2 examines the month-tomonth stability of the groups, by providing month-by-month transition frequencies from
one group to another. Not surprisingly, the more liquid and illiquid groups are more
stable than the groups in the middle. For groups 1, 5, and 10, the proportion of stocks
that remain in the group the next month is 89.88%, 48.46%, and 72.90%, respectively.
In all of our analysis, for example when examining the impact of the Libor-OIS spread
on volume across different liquidity groups, we work with liquidity groups based on the
previous month’s ILLIQ s.
Table 3, Panel A, provides descriptive statistics of individual stock’s monthly ILLIQ.
The pooled sample mean is 17.632 (median 0.134). These numbers refer to the plain (or
raw) ILLIQ multiplied by 1,000,000. That is to say, volume is measured in millions. By
way of comparison, Goyenko et al. (2009) report a mean of 6.31 (median 0.104). Thus,
the stocks in our sample are more illiquid than those in Goyenko et al.’s sample.14 Both
our and their estimates can be viewed as being large; a volume of 1 million dollars implies
a price change of 13.4% for the median firm in our sample and 10.4% in Goyenko et al.’s
sample. However, for the most part, we only use stocks’ ILLIQ values to classify them into
groups. For our purposes, it is the ordinal, not cardinal, accuracy of ILLIQ that matters.15
Panel B of Table 3 provides descriptive statistics of the average ILLIQ within groups.
the database where there is no recorded closing price, thus suggesting an absence of trade, yet there is
positive recorded volume.
14
Potential reasons for this difference include: (i) Goyenko et al. require their sample firms to be present
in the TAQ master file (because of their objective to compare the performance of high and low frequency
measures), which we do not. (ii) We consider different time periods. (iii) They use a random sample of
400 firms, while we use the entire CRSP database.
15
If cardinal accuracy is important, it may be advisable to use an Acharya and Pedersen (2005) style
truncation of ILLIQ in order to reduce the impact of extreme observations. In this paper, this is an issue
only for the regressions in Table 6. Footnote 19 discusses this further.
10
There is substantial variation across groups. The average ILLIQ is 0.001 for Group 1 and
162.601 for Group 10.
2.4
Liquidity group volume measures
Our analysis of volume in this paper revolves around four measures of normalized and
relative volume.
1. For each liquidity group G and day t, we calculate:
Volume Group G
t
Normalized share volume Group Gt = Average volume for G over the previous
five days ,
where volume is the number of shares traded.
2. Normalized dollar volume Group Gt.
This is defined the same way as normalized share volume, but with volume now being
the dollar value of trades. The two normalized volume measures capture abnormal
volume on a given day in a simple way.
3. For each liquidity group G, we calculate its market share of volume on day t as:
Volume Group Gt
Market share of volume Group Gt = Aggregate volume across all stocks
in the sample ,
t
where volume is measured in dollars.
4. For each pair of liquidity groups, G and H, G > H we calculate their relative volume
on day t as:
Volume Group G
Relative volume of Group G to Group H t = Volume Group Ht ,
t
where volume is measured in dollars.
To get a cursory sense of the magnitudes of these variables, over the period January 2,
1986 to December 31, 2008, the average (standard deviation) normalized share and dollar
volumes for the market as a whole are 1.007 (0.162) and 1.009 (0.183), respectively.16 In
16
These are calculated as follows: First, for each day we calculate the normalized share volume for each
individual stock and then take their equally weighted average, thus obtaining a series of daily observations
of the market normalized share volume. The market normalized dollar volume is calculated analogously.
11
other words, for a typical stock, share (dollar) volume increases by approximately 0.7%
(0.9%) per day. Over the same period, the average (standard deviation) of the Market
share of volume Group 1 and the Relative volume of Group 10 to Group 1 are 78.6%
(4.7%) and 0.043% (0.0847%), respectively.
2.5
Correlations
The analysis in this paper is primarily concerned with examining the relation between
interbank spreads and cross-sectional variations in volume for stocks with different liquidity
levels. According to the liquidity pull-back hypothesis, we expect to see relatively more
volume in more liquid stocks as the price of liquidity increases. The correlation between
the Libor-OIS and TED spreads and the Relative volume of Group 10 to Group 1 are,
respectively, -0.11 and -0.09 – yet, the correlations with aggregate market volume are
approximately zero. This is simple evidence in support of the liquidity pull-back idea.
The correlations between the Libor-OIS and TED spreads and the equally weighted market
return are -0.17 and -0.10, which are also consistent with the liquidity pull-back hypothesis.
In the next two sections we will test the hypothesis more fully.
3
Volume and returns on high versus low spread days
In this section, we carry out tests of the liquidity pull-back hypothesis using the normalized
share and dollar volume measures. We examine volume and returns of the ten liquidity
groups on high versus low spread days, by first conducting univariate analysis on our
portfolio sorts and second running Fama-MacBeth regressions. The idea is that if there is
a liquidity pull-back effect, it should be visible on extreme spread days.
We proceed, separately for the Libor-OIS and TED spreads, as follows: For each month
we: (i) select the two days with the highest and the two days with the lowest spreads;
and (ii) for the two high, as well as for the two low, spread days, we average the values of
the spread and the following three variables: each liquidity group’s normalized share and
12
dollar volumes and equally weighted return.17 In this way, for each variable of interest we
generate two time series with monthly observations, corresponding to the variable’s withinmonth average value on high and low spread days, respectively.18 Since each month in the
sample period contributes equally to any statistic we calculate, this procedure controls for
changes in the level of the spreads over time (e.g. crisis versus non-crisis periods).
3.1
Differences in means
Table 4, Panel A, reports average values of the selected variables on high and low Libor-OIS
spread days for each liquidity group. The normalized volume of liquid and illiquid stocks
are seen to move in opposite directions as we go from high to low spread days. On high
Libor-OIS spread days, the most liquid stocks (Group 1) have a normalized share (dollar)
volume of 1.058 (1.054) versus 0.933 (0.922) for the most illiquid stocks (Group 10). Put
differently, the most liquid stocks have an abnormally large share (dollar) volume of 5.8%
(5.4%) on high Libor-OIS spread days, while the most illiquid stocks have an abnormal
share (dollar) volume of -6.7% (-7.8%). The difference of 12.5% (13.2%) is significant, both
economically and statistically, with the t-statistic being 3.76 (2.95). On low spread days,
there is a reversal. The most liquid stocks (Group 1) have a normalized share (dollar)
volume of 0.999 (1.006) versus 1.090 (1.154) for the most illiquid stocks (Group 10). Put
differently, the most illiquid stocks have an abnormally large share (dollar) volume of 9.1%
(14.8%) on low spread days, while no effect is seen for the most liquid stocks. Consistent
17
The groups are formed a month in advance. That is, the volume and return variables are estimated
using data from the current month, while groups are based on the previous month’s ILLIQs. The volume
variables are calculated for the group as a whole (i.e. not averaged across the stocks in the group).
18
For a given month and spread, if there are more than two highest spread days, then all of those days
are weighted equally when calculating the monthly values of the variables we are looking at. If there is
a single highest spread day but several second highest spread days, then the latter are weighted equally.
For example, if three days have the second highest spread for a particular month then, for that month,
each of these three days represent one third of a high spread day. For a given variable (e.g. normalized
share volume), the monthly observation of the high spread day is then 0.5 times the variable’s value on the
unique high spread day plus 0.5 times the average value on the second highest spread days. We proceed
in the analogous way for low spread days.
13
with the liquidity pull-back hypothesis, this shows that volume is abnormally high (low)
for highly liquid (illiquid) stocks on days when the price of liquidity is especially large.
In terms of returns, we see that illiquid stocks offer higher returns overall (Amihud and
Mendelson 1986, Amihud 2002). More importantly with respect to the liquidity pull-back
hypothesis, we also see that returns are uniformly smaller across groups on high spread
days relative to low spread days.
Panel B of Table 4 reports on a similar exercise for the TED spread. The results
parallel those for the Libor-OIS spread.
Table 5 reports on the differences in normalized volume and returns between high and
low spread days for each liquidity group. Panel A is based on the Libor-OIS and Panel B
on the TED spread. Consistent with the liquidity pull-back hypothesis, for both volume
measures and for each spread, the difference in volume between high and low spread
days is decreasing in illiquidity, albeit not monotonically. For example, the differences
in normalized share volume between high and low Libor-OIS spread days is 5.9% for
Group 1 and -15.7% for Group 10. Thus, in terms of the difference in differences, as
we go from low to high spread days, the abnormal volume for the most liquid stocks
increases by a statistically significant 21.6% relative to that of the most illiquid stock. For
the normalized dollar volume, the difference in differences is even larger. These findings
support the liquidity pull-back hypothesis.
Our theoretical framework also predicts that returns should be lower on high spread
days, as agents pull-back liquidity from the markets. This is confirmed in Table 5. For
each liquidity group, returns are lower on high spread days. There also seems to be an
equalizing effect, in the sense that the difference in returns as we go from low to high
spread days is of the same magnitude across liquidity groups. For example, there is no
statistically significant difference (in the high minus low returns) between groups 1 and
10. This is so whether we base our analysis on the Libor-OIS or the TED spreads.
These initial results support the view that high spreads are associated with an increase
in volume for liquid stocks and a decrease for illiquid stocks. Our interpretation is that
when the market for liquidity is tight, banks or investors choose to sell assets for which
the price impact would be the least. That it is selling pressure that is behind the volume
14
effect we have documented is corroborated by the negative returns on high spread days.
3.2
Fama-MacBeth regressions
In this subsection, we run Fama-MacBeth regressions to test the hypothesis that the
difference in normalized volume between high and low spread days is decreasing in ILLIQ.
We proceed in two steps as follows:
First, we run the following cross-sectional regression for each month, m:
HSV OLG,m − LSV OLG,m = αm + βm × ILLIQG,m−1 + εG,m
(2)
where HSV OLG,m −LSV OLG,m is the difference in normalized share volume between high
and low spread days in month m for liquidity group G, and ILLIQG,m−1 is the average
ILLIQ across stocks in the group in month m − 1. Second, we average the coefficients
from the monthly regressions. These averages are reported in Table 6, with Newey-West
(1987) (3 lags) t-statistics.
This two-step procedure is run separately for the Libor-OIS and TED spreads. To
examine whether our findings are driven by the recent crisis, we run the procedure not
only over the entire Libor-OIS and TED spread sample periods, but also separately over
the pre-crisis periods. For the TED spread, we also run the procedure separately over the
Libor-OIS sample period.
Table 6 shows that, consistent with the liquidity pull-back hypothesis, the coefficient
on ILLIQG is negative regardless of which specification we look at.19 Furthermore, for the
Libor-OIS spread, the results are stronger for the pre-crisis period than for the sample
as a whole. For the pre-crisis period, the average coefficient on ILLIQG is -0.017 with a
t-statistic of -2.79, whereas the corresponding numbers for the whole Libor-OIS sample
period is -0.015 and -1.80. For the TED spread, the coefficient on ILLIQG is -0.004,
whether we use the sample as a whole or only the pre-crisis period, with the respective
t-statistics being -2.28 and -2.05. For the regression based on the monthly high and low
19
We have also run the regressions in Table 6 using a truncated ILLIQ measure, along the lines of
Acharya and Pedersen (2005), as follows: ILLIQij,trunc = min(0.25+0.3×ILLIQij , 30). This strengthens
the results both in terms of statistical and economic significance.
15
TED spread days over the Libor-OIS sample period, the coefficient on ILLIQG is -0.013,
with a t-statistic of -2.21.
These findings are supportive of liquidity pull-back being a feature of the markets. Our
results are not driven by the crisis. During normal non-crisis times, we find that abnormal
trading volume is decreasing in stock illiquidity as the interbank price of liquidity increases.
Below, we run further tests, examining volume across the different liquidity groups as
the spreads fluctuate day-to-day. In these regressions, we consider different time periods
and include a number of controls to examine the robustness of our findings.
4
Regressions using daily observations
This section contains the paper’s main regression analysis. We examine three issues.
• First, we test whether there is support for the liquidity pull-back hypothesis in the
data on a day-to-day basis. Thus, we use the full samples rather than just the
extreme spread days.
• Second, we examine whether our findings are driven by the crisis. Thus, we run
separate regressions for the pre-crisis and crisis periods.
• Third, we examine the alternative hypothesis that our results are driven by portfolio
rebalancing due to market-wide uncertainty rather than liquidity pull-back, as such.
Thus, we include the VIX as a control variable to proxy for market-wide risk. This
also means that we examine cross-sectional variations in volume, across stocks of different liquidity levels, in response to changes in market-wide uncertainty. Additional
control variables (see below) are also included in these regressions.
The main regressions are run with the Market share of volume of the different groups
as the dependent variable. To examine whether the results are driven by the most liquid
stocks, we rerun a number of the regressions with Relative volume (for all group pairs) as
the dependent variable. Because we use the Market share of volume and Relative volume
variables, the analysis in this section relies only on the ordinal, rather than the cardinal,
accuracy of Amihud’s ILLIQ measure.
16
4.1
Simple regressions: Market share of volume on the spreads
As a baseline, we start the analysis by running simple regressions on the full sample periods.
For each group G, we run the following time-series regression using daily observations:
Market Share of Volume Group Gt
= α + β × spreadt + εt
mean(Market Share of Volume Group G )
(3)
where spreadt is either the Libor-OIS or TED spread on date t and, as always, liquidity
groups are formed based on individual stocks’ ILLIQ values the previous month. We run
separate regressions for each spread. To correct for autocorrelation, the regressions are run
using the Cochrane-Orcutt procedure.20 So as to facilitate comparisons across liquidity
groups, we have normalized the market share of volume for each group by its time series
average. Based on the liquidity pull-back hypothesis, we expect to see the coefficient on
the spread to be positive for the group consisting of the most liquid stocks (Group 1) and
negative for the most illiquid stocks (Group 10). More generally, we expect the regression
coefficient to decrease as we go from Group 1 to Group 10.
The regression results are reported in Table 7. For both spreads, the coefficient on
the spread is positive for the most liquid group (Group 1) and negative for all other
groups. With one exception, all coefficients are statistically significant. Furthermore,
the coefficients on either spread are decreasing (almost monotonically) in the illiquidity
ranking of the groups. In terms of economic significance, a one standard deviation increase
in the Libor-OIS spread leads to an increase (decrease) in the Market Share of Volume of
Group 1 (Group 10) of 1.14% (28.27%) relative to the group’s unconditional mean. These
results support the liquidity pull-back hypothesis.
20
We have also run OLS with Newey-West (5 lags) standard errors. In the majority of cases, this yields
smaller standard errors and results that are more supportive of our hypothesis than the results using the
Cochrane-Orcutt procedure. We have also run unit root tests on the Libor-OIS and TED spreads. Using
the Augmented Dickey-Fuller test we reject that the Libor-OIS is a unit root at the 5% and that the TED
spread is a unit root at the 1% level. We also reject that the Libor-OIS and TED spreads follows a unit
root at the 1% level with the Zivot-Andrews (1992) test that allows for a structural break. This tests
identifies a structural break in August 2007, which is also when visual inspection reveals a sharp increase
in this spread. Details are available upon request.
17
4.2
Multivariate regressions, different time periods, and alternative hypotheses
In this subsection, we examine whether the results in the previous section (i) are driven by
the crisis, and (ii) stand up to the inclusion of control variables. With respect to point (i),
we break the sample period for the Libor-OIS spread up into a pre-crisis period, December
4, 2001 to June 30, 2007, and a post-TAF (term auction facility) period, December 17,
2007 to November 11, 2008. We drop July 2007 from the pre-crisis period sample in order
to avoid any contamination from the beginning of the crisis. For the crisis-period, we focus
on the post-TAF period, since the introduction of the term auction facility represents a
loosening of monetary policy. The large quantity of additional liquidity injected through
TAF and subsequent programs during the crisis may have weakened the need for liquidity
pull-back. Our regressions allow us to study this and thus comment on the effectiveness
of TAF. To minimize the number of tables, in this subsection and for the remainder of the
paper, we focus on the Libor-OIS spread only.21
With respect to point (ii), while we include several control variables, our main concern
is to control for the alternative hypothesis that our finding in the previous subsection is
the result of portfolio-rebalancing arising from market-wide uncertainty that also affects
the Libor-OIS spread. The idea is that an increase in market-wide uncertainty may lead
to an increase in the Libor-OIS spread, for example due to increased credit risk, and at the
same time to portfolio rebalancing whereby (some) agents in the economy seek to reduce
their equity exposures. Moreover, this portfolio rebalancing is such that it gives rise to
the cross-sectional pattern in volume we observed above. As our measure of market-wide
uncertainty, we use the VIX.
That the alternative portfolio rebalancing hypothesis should be taken seriously is underscored by (i) the large correlation between the VIX and the Libor-OIS spread (0.67
over the sample period) and (ii) the finding by Ang, Gorovyy, and Inwegen (2010) that
hedge fund leverage is decreasing in market-wide uncertainty, as measured by the VIX.
21
As discussed in the Introduction, the Libor-OIS spread is a more accurate measure of the price of
liquidity than the TED spread.
18
This supports the view that some agents shift out of riskier assets, such as equities, as
volatility increases. Of course, even if some agents reduce equity exposures as volatility
increases, it is not clear that this would impact on the market share of volume of different
liquidity groups. An increase in volatility and an accompanying shift out of equities may
involve larger volume in more liquid stocks, since, as we have explained above, liquidating
large stock positions while minimizing total price impact would have to involve relatively
more volume in more liquid stocks. But the flip side of this is that a fall in volatility
should see a shift into equities, which, to minimize price impact, should also be associated with relatively more volume in more liquid stocks. There is evidence, however, that
more illiquid stocks also have larger liquidity risk (Amihud, Mendelson, and Wood 1990,
Acharya and Pedersen 2005).22 So more volatile markets may involve more illiquid stocks
becoming relatively more illiquid. This could give rise to an asymmetric reaction between
liquid and illiquid stocks to the level of volatility. Given the high correlation between
the Libor-OIS spread and the VIX, it is difficult to distinguish between the liquidity pullback and the portfolio rebalancing hypotheses. Moreover, re-running the regression in the
previous subsection with the VIX in place of the Libor-OIS spread, we get statistically significant coefficients that exhibit the same decreasing pattern as we saw for the Libor-OIS
spread. Thus, we cannot say whether the results from the previous subsection is evidence
of liquidity pull-back, portfolio rebalancing, or both.
To distinguish between the two hypotheses, we run a two-step procedure where we
use only that part of the Libor-OIS spread that is orthogonal to the VIX (for a strong
test of the liquidity pull-back hypothesis) and vice versa (for a strong test of the portfolio
rebalancing hypothesis). We run this procedure separately over both the pre-crisis and
post-TAF periods. Specifically, we proceed as follows:
Step 1. Regress, using OLS, the Libor-OIS spread on the VIX (or vice versa):
Zt = α + γXt + εt
(4)
where Zt and Xt are the Libor-OIS spread and the VIX, respectively (or vice versa).
Step 2. Include the residuals, ResidualZ|X , from Step 1 among the regressors in the
22
See also Pastor and Stambaugh (2003) for a discussion of liquidity risk.
19
regression of interest:
Market Share of V olume Gt
= α + β1 Xt + β2ResidualZt|Xt + Γ0Wt + ηt , (5)
mean(Market Share of V olume G)
where W is a vector with control variables (discussed below) and Γ is a vector with the
corresponding regression coefficients. Estimation is performed using the Cochrane-Orcutt
procedure.23 By toggling between the Libor-OIS spread and the VIX as X or Z, we get
two sets of results for each subperiod. If both liquidity pull-back and portfolio rebalancing
are present in the data, we should see the same decreasing pattern for the coefficients on
X as we got in the simple regression in the previous subsection, regardless of which of the
Libor-OIS spread and the VIX is used as X or Z. For stronger tests, we should observe
the decreasing pattern also on ResidualZ|X .
To make this more concrete, suppose X is the Libor-OIS spread and Z is the VIX. In
this case, a decreasing pattern on the coefficient of the Libor-OIS spread, as we go from
more liquid to less liquid groups, would be support for the liquidity pull-back hypothesis.
However, the support is only weak, in the sense that we could not exclude the possibility
that the decreasing pattern is due to the VIX, since the second step regression here is only
controlling for that part of the VIX that cannot be explained by the Libor-OIS spread. A
much stronger test of the liquidity pull-back hypothesis would be to reverse the roles of the
VIX and the Libor-OIS spread (in terms of being X or Z) and examine the coefficients on
ResidualLibor−OIS|V IX . If there is liquidity pull-back taking place, we would expect these
coefficients to be statistically significant and exhibit a decreasing pattern. Conversely, our
strongest test of the portfolio rebalancing hypothesis is to let the VIX be Z and the LiborOIS spread be X and require the coefficients on ResidualV IX|Libor−OIS to be decreasing in
the group number. These are strong tests in part because of the high correlation between
the Libor-OIS spread and the VIX and in part because of the presence of control variables.
We use three control variables, measured with daily frequency. (i) Lagmarket: the rate
of return on the value weighted CRSP market portfolio from the previous day. Past returns
23
We have also run the second step with OLS and Newey-West standard errors. This yields somewhat
stronger results (in support of the liquidity pull-back hypothesis) both economically and statistically.
Details are available upon request.
20
have been shown by Gallant, Rossi and Tauchen (1992) to affect aggregate volume. (ii)
Relative bid-ask spread (for a liquidity group): On each day, for each stock in the group, we
calculate its bid-ask spread as a fraction of its reported closing price.24 We then calculate
the equally weighted average of these (fractional) bid-ask spreads for each group, which
we finally express as a fraction of the equally weighted average across the ten liquidity
groups to get the group’s Relative bid-ask spread. This variable may pick up differences in
the liquidity of stocks not captured by ILLIQ. Benston and Hagerman (1974) document
a relation between the bid-ask spread and volume. (iii) Normalized market dollar volume.
This is a straight control for the aggregate volume in the market. To the extent that Group
1 stocks are larger than other stocks, this may soak up much of the variation in a group’s
Market share of volume and thus make it that much “harder” for the Libor-OIS spread or
the ResidualLibor−OIS|V IX to have statistically significant Step 2 regression coefficients.
Summary statistics of the VIX and the three other control variables are in Table 8,
broken down into the pre-crisis and post-TAF periods. The table also reports on the
correlations between the Libor-OIS and the VIX in these two periods. Pre-crisis, the
correlation is 0.61 and post-TAF it is 0.89.
The results are in Table 9, which reports the second step regression output from the
four constellations described above. Panel A is for the pre-crisis period with the LiborOIS spread as X and the VIX as Z. This thus contains our weakest test of the liquidity
pull-back hypothesis and our strongest test of the portfolio rebalancing hypothesis, for
the pre-crisis period. Panel B reverses the roles of the Libor-OIS spread and the VIX.
It therefore contains our weakest test of the portfolio rebalancing hypothesis and our
strongest test of the liquidity pull-back hypothesis. Panels C and D repeat the exercise
for the post-TAF period.
The results for the liquidity pull-back hypothesis are as follows: Panel A shows the
same decreasing trend in the coefficient (going from Group 1 to Group 10) on the LiborOIS spread as in the previous subsection. The coefficient (t-statistic) is 0.787 (18.60)
24
In those rare instances where CRSP reports a greater bid than ask for a given stock on a given day,
we exclude that observation. On the upside, we winsorize the 99th percentile. In particular, for individual
stocks with a bid-ask spread above 22% on a given day, the stock’s bid-ask spread is set equal to 22%.
21
and -6.896 (-1.90) for Group 1 and 10, respectively. In terms of economic significance,
one standard deviation (based on the pre-crisis period) increase in the Libor-OIS spread
leads to an increase (decrease) in the Market Share of Volume of Group 1 (Group 10) of
2.83% (24.83%) relative to the pre-crisis period mean. This is supportive evidence for the
liquidity pull-back hypothesis.
The stronger test in Panel B shows that the coefficient on ResidualLibor−OIS|V IX is
statistically significant (at the 1% to 10% levels) for Groups 1 to 6 (inclusive) and is
decreasing over these groups. We interpret this as strong evidence for the presence of
liquidity pull-back during normal, non-crisis times.
With respect to the portfolio rebalancing hypothesis, Panel B reports statistically significant coefficients on the VIX that are decreasing in the group number. The coefficient
(t-statistic) is 0.006 (21.79) and -0.045 (-1.89) for Group 1 and 10, respectively. In terms
of economic significance, one standard deviation (based on the pre-crisis period) increase
in the VIX leads to an increase (decrease) in the Market Share of Volume of Group 1
(Group 10) of 4.08% (31.21%) relative to the pre-crisis period mean. This is supportive
evidence for the portfolio rebalancing hypothesis.
For a stronger test, Panel A shows that the coefficients on ResidualV IX|Libor−OIS are
highly statistically significant for all groups and exhibit a decreasing pattern. We interpret
this as strong evidence in support of the portfolio rebalancing hypothesis. Indeed, while the
evidence thus shows that both liquidity pull-back and portfolio rebalancing are features of
the markets over the non-crisis period, portfolio rebalancing appears to be stronger effect
(in the sense that the coefficients on ResidualV IX|Libor−OIS have larger t-statistics than
those on ResidualLibor−OIS|V IX ). But this does not take away from the positive results on
the liquidity pull-back hypothesis.
Panels C and D repeat the exercise for the post-TAF period. Panel C reports that the
coefficients on the Libor-OIS still exhibits the same decreasing trend as earlier; but, as
seen in Panel D, the coefficients on ResidualLibor−OIS|V IX are not statistically significant
over the post-TAF period for any group (with the exception of Group 8). Thus, there is
only weak, if any, evidence of liquidity pull-back over this period. This is consistent with
the view that the loose monetary policy over this period obviated the need for liquidity
22
pull-back. In this sense, TAF can be said to have worked.
In contrast, there is strong evidence for portfolio rebalancing over the post-TAF period.
Panel D reports that the coefficients on the VIX are statistically significant over the postTAF period and are decreasing in the liquidity group number. Panel C shows that the
coefficients on ResidualV IX|Libor−OIS are statistically significant up to Group 7 (inclusive)
and exhibit a decreasing trend. Thus, basic portfolio rebalancing in response to marketwide uncertainty is unaffected by the looser monetary policy in the post-TAF period, while
liquidity pull-back more or less disappears.
In conclusion, the results in Table 9 show that (i) there is evidence of liquidity pull-back
on a day-to-day basis in the pre-crisis period, (ii) TAF seems to have been successful in
the sense that it appears to have eliminated liquidity pull-back, (iii) the relative volume
of more liquid stocks is increasing in market-wide uncertainty, which we interpret as being
evidence of portfolio rebalancing (for some market participants). Our results establish
that stock market activity reacts to that part of the Libor-OIS spread which is orthogonal
to variations in the VIX (and vice versa). This is strong evidence that there is a monetary
effect (liquidity pull-back) alongside a more standard effect due to uncertainty that affects
trading activity in stock markets.
4.3
Multivariate regressions with Relative volume
In this subsection, we examine whether our results are driven by Group 1 by repeating
the two step procedure from the previous subsection [equations (4) and (5)], but now
with Relative volume as the dependent variable. Thus, for each specification (and time
period), we run 45 separate regressions. To minimize the number of panels, we report the
results only for the pre-crisis period, which is also where our previous findings indicated
the strongest support for the liquidity pull-back hypothesis.
Table 10, Panel A (Panel B) exhibits the regression coefficients on the Libor-OIS spread
(VIX). With only one exception, regardless of which pairing we look at, the coefficients
on the Libor-OIS spread and the VIX are negative.25 The coefficients on the Libor-OIS
25
The exception is the Relative volume of Group 7 to Group 8. Here, the coefficients are positive, but
23
spread are statistically significant at the 1% level for all pairs with H ≤ 5 and G ≤ 8
(with the exception of one coefficient which is significant at the 5% level), where H (G)
is the liquidity group in the denominator (numerator) and H < G. For the VIX, the
results are similar. Recalling that Relative volume is always measured in terms of the
volume of the less liquid group as a fraction of the volume of the more liquid group, this
is strong evidence that relatively more liquid stocks experience a relatively higher volume
as market-wide uncertainty or the price of liquidity in the interbank market increases.
Our findings here show that the cross-sectional variation in volume and its relation to the
Libor-OIS spread and the VIX we found in the previous subsection is not driven by the
most liquid group (Group 1).
4.4
Robustness: exclusion of NASDAQ stocks
To maximize the variation in liquidity across stocks, and thus hopefully the power of
our tests, we have included stocks listed on all the major exchanges, NYSE, AMEX and
NASDAQ. Sixty three percent of our sample is comprised of NASDAQ stocks. However,
as noted by Atkins and Dyl (1997) and Anderson and Dyl (2007), the volume of a stock
that switches from NASDAQ to NYSE or AMEX often falls. This is due to the dealer
structure on NASDAQ which implies a significant amount of transactions between dealers
that is recorded as trading volume.
In this subsection, we examine whether our results are driven by NASDAQ stocks. To
do so, we exclude all NASDAQ stocks from the analysis. Thus, we construct new liquidity
groups for each month, comprised only of NYSE and AMEX stocks, and rerun the two-step
procedure described in Subsection 4.2.
Table 11 reports the results. The table is laid out in four panels in the exact same
way as our main results in Table 9. With respect to the liquidity pull-back hypothesis
over the pre-crisis period, the results are stronger than in Table 9, when NASDAQ stocks
were included. We see the same decreasing pattern in the coefficients on the Libor-OIS
spread and on ResidualLibor−OIS|V IX , but for the latter the coefficients are now statistically
not significantly so.
24
significant up to Group 7, inclusive. The portfolio rebalancing hypothesis also sees strong
support, as before. For the post-TAF period, we once again get only weak support for the
presence of liquidity pull-back, but strong support for portfolio rebalancing. In conclusion,
our results are robust to the exclusion of NASDAQ stocks.
5
Conclusion
We have argued that there is a connection between the interbank market for liquidity
and the broader financial markets, which has its basis in demand for liquidity by banks.
Tightness in the interbank market for liquidity leads banks to engage in what we term
liquidity pull-back, which involves selling financial assets either by banks directly or by
levered investors. This does not increase the stock of money in aggregate, but can increase
the money balances of an individual bank. This line of reasoning has several implications,
and the body of the paper is devoted to testing these. The implications are verified in the
data.
In our empirical analysis, we capture the tightness of the interbank markets by the
price of liquidity, as measured by the Libor-OIS spread. In some of our analysis, we
have also run parallel tests using the TED spread. The baseline finding is that over the
whole sample period (which overlaps with the crisis) an increase in the price of liquidity is
associated with an increase in the volume of highly liquid stocks and a decline in that of
highly illiquid stocks. We have also done a cursory analysis on returns and found evidence
that there is an equalizing effect across stocks of different liquidity levels with respect to
how they react to changes in the price of liquidity. These findings are consistent with the
liquidity pull-back hypothesis.
To examine the robustness of our baseline finding, we have studied the pre-crisis and
post-TAF periods separately, included the VIX and other factors as control variables,
examined the extent to which our results are driven by the most liquid stocks, and excluded
NASDAQ stocks from the analysis. Interestingly, we find strong evidence for liquidity pullback over the pre-crisis period, but only weak evidence over the post-TAF period. We think
this makes sense in light of the loose monetary policy post-TAF.
25
By using the VIX as a control variable, we have also examined the alternative hypothesis that it is portfolio rebalancing in response to changes in volatility that drives the
relation we have uncovered between the price of liquidity in the interbank market and the
market share of volume of stocks of different liquidity levels. Our evidence suggests that
both such portfolio rebalancing and liquidity pull-back are going on. Unlike liquidity pullback, however, the evidence for portfolio rebalancing is also strong post-TAF. Our other
robustness tests shows that our results are not driven by the most liquid stocks, nor are
they driven by NASDAQ stocks. Indeed, our findings are stronger without those stocks.
The perspective we advance in this paper is that an important function of financial
markets is to act as a liquidity storage facility that players can dip into when they need
liquidity (in the monetary sense). In many ways, this is not new. It is present or implicit
in models of intertemporal saving. It is also explicit in the idea of “liquidity traders” in
the literatures on noisy rational expectations equilibria and market microstructure. In
simplistic terms, our point is that banks can act as liquidity traders as well, or force
liquidity trading by levered investors through increasing margins or haircuts. A bank may
engage in this when it needs higher powered money but the price it faces in the interbank
market is too high or its interbank credit limits are exhausted. Selling a financial asset
can be thought of as an act of converting low powered money (financial assets) into higher
powered money (liquidity). When the price of liquidity goes up, the price of conversion
also rises and asset prices therefore fall. Cross-sectional volume effects arise because stocks
have different liquidity levels.
The framework for thinking about money and liquidity in financial markets that we
have outlined in this paper may be relevant for a number of liquidity phenomena. As an
example, consider the phenomenon of increased correlations during crisis times (King and
Wadhwani, 1990) and “flight to quality” (Sundaresan, 2009 p. 343). While much of these
effects may be due to increased uncertainty, there may also be a liquidity pull-back effect
present. The liquidity pull-back interpretation of the phenomenon of increased correlations
in crisis is that these are times when liquidity is extremely dear or difficult to get in the
interbank market. Banks therefore engage in liquidity pull-back. Put differently, there
is a (financial market) credit contraction. The conjecture is that the worse conditions
26
are in the interbank market, the larger are asset cross-correlations and the stronger will
flight to quality appear to be. We have presented some cursory evidence on asset returns.
Investigating this more thoroughly is an important avenue for future research.
27
A
Appendix
Variable
Libor-OIS spread
Description
The three month USD London Interbank Offered Rate (Libor) less the USD three month
Overnight Index Swap (OIS) rate.
T ed spread
The three month USD London Interbank Offered Rate (Libor) less the three month T-bill rate.
VIX
The VIX is calculated using the methodology introduced in 2003.
See http://www.cboe.com/micro/VIX/vixintro.aspx.
Normalized share volume
Share volume on day t divided by the average daily share volume over the previous five day period.
Normalized dollar volume
Dollar volume on day t divided by the average daily dollar volume over the previous five day
period.
28
ILLIQij = Avg
|rt|
V olumet
rt is the return and Volumet is the dollar trading volume on day t of stock i. ILLIQij is
the average of |rt | /Volumet over all days in month j. Throughout the paper we report
ILLIQij × 106 (i.e., volume is measured in millions).
ILLIQG
Equally-weighted average of ILLIQij across all stocks that belong to liquidity group G.
Market share of volume of Group Gt
Dollar volume of all stocks in liquidity group G on day t divided by the dollar volume of all stocks
in our sample.
Relative volume of Group G to Group Ht
Dollar volume of all stocks in liquidity group G divided by the dollar volume of all stocks in
liquidity group H on day t.
Lagmarkett
CRSP value-weighted return on day t − 1
Group relative bid-askt
Equally weighted average relative bid-ask (closing ask less the closing bid divided by the price) of all
stocks in the liquidity group divided by the equal-weighted relative bid-ask of all stocks in the
sample on day t.
Normalized market dollar volumet
Equally weighted average normalized dollar volume of all stocks on day t.
References
Acharya, Viral and Lasse H. Pedersen, 2005, Asset pricing with liquidity risk, Journal of
Financial Economics 77, 375-410.
Akerlof, George, A., 1970, The market for ’lemons’: Quality uncertainty and the market
mechanism, Quarterly Journal of Economics 84, 488-500.
Amihud, Yakov, 2002, Illiquidity and stock returns: Cross section and time-series effects,
Journal of Financial Markets 5, 31-56.
Amihud, Yakov and Haim Mendelson, 1986, Asset pricing and the bid-ask spread, Journal
of Financial Economics 17, 223-249.
Amihud, Yakov, Haim Mendelson, and R. Wood, 1990, Liquidity and the 1987 stock
market crash, Journal of Portfolio Management 16, 6569.
Anderson, Anne M., and Edward A. Dyl, 2007, Trading Volume: NASDAQ and the NYSE,
Financial Analysts Journal 63, 115-131.
Ang, Andrew, Sergiy Gorovyy, and Gregory B. van Inwegen, 2010, Hedge fund leverage,
working paper, Columbia University.
Atkins, Allen and Edward A. Dyl, 1997, Market Structure and Reported Trading Volume:
Nasdaq versus the NYSE, Journal of Financial Research 20, 291-304.
Benston, George J. and Robert L. Hagerman, 1974, Determinants of bid-ask spreads in
the over-the-counter market, Journal of Financial Economics 1, 353-364.
Bindseil, Ulrich, Kjell G. Nyborg and Ilya Strebulaev, 2009, Repo auctions and the market
for liquidity, Journal of Money Credit and Banking 41, 1391-1421.
Black, Fischer, 1971, Towards a fully automated exchange, Part I, Financial Analysts
Journal 27, 29-34.
Boyer, Brian and Lu Zheng, 2009, Investor flows and stock market returns, Journal of
Empirical Finance 16, 87-100.
Brunnermeier, Markus, K., and Lasse H. Pedersen, 2009, Market liquidity and funding
liquidity, Review of Financial Studies 22, 2201-2238.
Calvet, Laurent E., John Y. Campbell, and Paolo Sodini, 2009, Fight or flight? Portfolio
rebalancing by individual investors, Quarterly Journal of Economics 124, 301-348.
Cassola, Nuno, Cornelia Holthausen, and Marco Lo Duca, 2008, The 2007/2008 turmoil:
a challenge for the integration of the euro area money market?, European Central Bank
Working Paper.
Chordia, Tarun, Richard Roll, and Avanidhar Subrahmanyam, 2000, Commonality in
liquidity, Journal of Financial Economics 56, 3-28.
Chordia, Tarun, Richard Roll, and Avanidhar Subrahmanyam, 2001, Market liquidity and
29
trading activity, Journal of Finance 56, 501-530.
Chordia, Tarun, Asani Sarkar, and Avanidhar Subrahmanyam, 2005, An empirical analysis
of stocks and bond market liquidity, Review of Financial Studies 18, 85-130.
Edelen, Roger, M., and Jerold B. Warner, 2001, Aggregate price effects of institutional
trading: A study of mutual fund flow and market returns, Journal of Financial Economics
59, 195-220.
Fair, Ray, C., 2002, Events that shock the market, Journal of Business 75, 713-731.
Fecht, Falko, Kjell G. Nyborg, and Jörg Rocholl, 2010, The price of liquidity: the effects of
market conditions and bank characteristics, Journal of Financial Economics, forthcoming.
Ferson, Wayne E. and Campbell R. Harvey, 1993, The risk and predictability of international equity returns, Review of Financial Studies 6, 527-566.
Fleming, Michael, J., and Eli M. Remolona, 1997, What moves the bond market?, FRBNY
Economic Policy Review 3, 31-50.
Friedman, Milton, 1970, A framework for monetary analysis, Journal of Political Economy
78, 193-238.
Gallant, A. Ronald, Rossi, Peter E., and George Tauchen, 1992, Stock prices and volume,
Review of Financial Studies 5, 199-242.
Goetzmann, William, N., and Massimo Massa, 2002, Daily momentum and contrarian
behavior of index fund investors, Journal of Financial and Quantitative Analysis 37, 375389.
Goyenko, Ruslan, Craig W. Holden, and Charles Trzcinka, 2009, Do liquidity measures
measure liquidity? Journal of Financial Economics 92, 153-181.
Gorton, Gary, B., and Andrew Metrick, 2009, Securitized banking and the run on repo,
Yale ICF Working Paper 09-14.
Grossman, Sanford, J., and Merton Miller, 1988, Liquidity and market structure, Journal
of Finance 43, 617-637.
Hasbrouck, Joel, and Duane J. Seppi, 2001, Common Factors in Prices, Order Flow, and
Liquidity, Journal of Financial Economics 59, 383-411.
Hau, Harald and Rey, Helene, 2004, Can portfolio returns explain the dynamics of equity
returns, equity flows, and exchange rates?, American Economic Review 94, 126-133.
Huberman, Gur, and Dominika Halka, 2001, Systematic liquidity, Journal of Financial
Research 24, 161-178.
Jaffee, Dwight, M., and Thomas Russell, 1976, Imperfect information, uncertainty, and
credit rationing, Quarterly Journal of Economics 90, 651-666.
Kashyap, Anil, and Jeremy C. Stein, 2000, What do a million observations on banks say
30
about the transmission of monetary policy?, American Economic Review 90, 407-428.
Karpoff, Jonathan, M., 1987, The relation between price changes and trading volume: A
survey, Journal of Financial and Quantitative Analysis 22, 109-126.
King, Mervyn, A., and Sushil Wadhwani, 1990, Transmission of volatility between stock
markets, Review of Financial Studies 3, 5-33.
Kyle, Albert, 1985, Continuous auctions and insider trading, Econometrica 53, 1315-1335.
Lo, Andrew, W., and Jiang Wang, 2000, Trading volume: Definitions, data analysis, and
implications of portfolio theory, Review of Financial Studies 13, 257-300.
Newey, Whitney K., and Kenneth D. West, 1987, A Simple, Positive Semi-definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix, Econometrica 55, 703–
708.
Pastor, Lubos and Robert Stambaugh, 2003, Liquidity risk and expected stock returns,
Journal of Political Economy 111, 642685.
Piazzezi, Monika, 2005, Bond yields and the Federal Reserve, Journal of Political Economy
113, 311-344.
Ritter, Jay R., 1988, The buying and selling behavior of individual investors at the turn
of the year, Journal of Finance 43, 701-717.
Roll, Richard, 1988, The international crash of october 1987, Financial Analysts Journal
19-35.
Stiglitz, Joseph, E., and Andrew Weiss, 1981, Credit rationing in markets with imperfect
information, American Economic Review 71, 393-410.
Sundaresan, Suresh, 2009, Fixed income markets and their derivatives, Academic Press.
Tobin, James, 1980, Redefining the aggregates: Comments on the exercise, in Measuring
the Money Aggregates: Compendium of views prepared by the subcommittee on domestic monetary policy of the House committee on banking, finance, and urban affairs, 96th
Congress, 2nd Session, Washington DC: US Government Printing Office.
Ying, Charles, C., 1966, Stock market prices and volumes of sales, Econometrica 34, 676685.
Zivot, Eric and Donald W. K. Andrews, 1992, Further Evidence on the Great Crash,
the Oil Price Shock and the Unit Root Hypothesis, Journal of Business and Economic
Statistics 10, 251-270.
31
Basis Points
Figure 1: Ted and Libor‐OIS spreads
600
500
400
TED
Libor‐OIS
Start of Crisis: 9 Aug 2007
300
200
100
0
Figure 2: TED Spread and Crisis Periods
Basis Points
500
450
400
TED Crisis
350
TED Non‐Crisis
300
Black Monday
Financial Crisis
250
200
Iraq invades Kuwait
Russian crisis and LTCM
150
End of dot com "bubble"
100
50
0
Asian crisis
Notes to Figure 2: Crisis periods
Crisis
Dates
Black Monday / the 1987 Crash October 15, 1987 –October 30, 1987
First Gulf War
August 2, 1990 – February 1, 1991
The Asian crisis
May 12, 1997 – August 19, 1997
Russian crisis and LTCM
August 13, 1998 – October 30, 1998
End of dot com ”bubble”
March 13, 2000 – April 28, 2000
Financial crisis
August 9, 2007 – December 31, 2008
Note: The sample period for the figure is January 2, 1986 to December 31, 2008.
1
Table 1
Libor-OIS and Ted spreads: descriptive statistics
The Libor-OIS and TED spread sample periods are, respectively, December 4, 2001 to November 11, 2008 and January 2, 1986
to December 31, 2008. Descriptive statistics are reported for the full sample (All Days) and four subsamples: (i) Non-crisis and
(ii) Crisis, which are defined as in the notes to Figure 2. (iii) High and (iv) Low spread days, which are defined as follows: For
each spread and for each month we calculate the average of the two highest, respectively lowest, values of the spread within the
month, thus for each spread yielding a sample of monthly high, respectively low, spreads. bps: basis points.
2
Variable
All Days
Libor-OIS
Ted
Non-Crisis Libor-OIS
Ted
Crisis
Libor-OIS
Ted
High
Libor-OIS
Ted
Ted (Libor-OIS period)
Low
Libor-OIS
Ted
Ted (Libor-OIS period)
Units Mean
bps
25.80
bps
55.80
bps
11.06
bps
47.95
bps
91.62
bps
117.7
bps
33.03
bps
70.11
bps
62.31
bps
21.69
bps
43.14
bps
38.88
Std Error Std Dev Median
1.01
41.74
11.75
0.58
43.85
41.77
0.10
3.59
10.52
0.45
31.82
38.44
3.64
64.58
72.83
2.75
69.25
99.50
5.87
53.81
13.80
3.22
53.55
54.39
8.02
73.48
34.06
3.85
35.31
10.05
2.01
33.33
32.72
4.74
43.42
21.05
Min
2.51
2.63
2.51
2.63
30.18
26.00
7.75
15.28
15.28
3.56
6.31
8.16
Max
366.33
456.88
28.40
258.00
366.33
456.88
363.24
444.94
444.94
243.69
252.31
252.31
N
1716
5648
1402
5013
314
635
84
276
84
84
276
84
Table 2
Liquidity group transition matrix
Each month, stocks are sorted into liquidity deciles (1 liquid and 10 illiquid), based on Amihud’s (2002)
ILLIQ [see equation (1) for the definition]. The table reports the percentage of stocks that are in group G one
month and H the next; G, H = 1, . . . , 10.
month t
Group
1
2
3
4
5
6
7
8
9
10
1
2
89.88
9.76
9.84 72.69
0.23 15.61
0.05 1.00
0.02 0.20
0.01 0.06
0.00 0.02
0.01 0.00
0.01 0.00
0.01 0.00
month t + 1
3
4
0.19 0.08
16.41
0.85
61.07 20.36
18.35 53.43
2.49 19.82
0.50 3.61
0.14 0.88
0.04 0.24
0.02 0.06
0.00 0.02
3
5
0.04
0.14
2.21
22.72
48.46
19.90
4.58
1.18
0.27
0.06
6
0.02
0.05
0.39
3.53
23.12
45.88
20.14
5.25
1.22
0.26
7
0.01
0.02
0.09
0.73
4.68
23.52
44.32
20.39
5.33
0.93
8
0.01
0.01
0.02
0.15
0.98
5.33
23.54
44.20
21.58
4.39
9
0.01
0.00
0.01
0.03
0.19
1.02
5.59
24.33
48.14
21.43
10
0.01
0.00
0.00
0.01
0.04
0.18
0.80
4.35
23.36
72.90
Table 3
ILLIQ : descriptive statistics
Panel A displays descriptive statistics of the pooled sample of Amihuds’ (2002)
ILLIQ across all stocks and months in the sample period. Panel B does the same for each
liquidity group (see Table 2). For each stock, ILLIQ is calculated for each month according
to equation (1), with volume measured in millions of dollars. In calculating each stock’s
monthly ILLIQ, we only consider positive volume days and exclude days with no recorded
closing price on either day t or t−1 and a zero return on day t. The sample period is January
2, 1986 to December 31, 2008.
Mean Median Std Err
Panel A: Pooled sample
17.632
0.134
0.566
Panel B: Groupwise
1
0.001
0.001
0.000
2
0.008
0.005
0.000
3
0.026
0.015
0.000
4
0.077
0.039
0.000
5
0.201
0.093
0.001
6
0.493
0.232
0.002
7
1.190
0.612
0.004
8
2.987
1.766
0.009
9
8.858
5.981
0.026
10 162.601 34.435
5.648
Std Dev
Min
Max
N
631.266
0.000
291,059.300
1,243,739
0.002
0.009
0.032
0.099
0.259
0.621
1.424
3.322
9.212
1991.112
0.000
0.000
0.000
0.001
0.002
0.004
0.010
0.032
0.100
0.468
0.016
0.075
0.277
0.813
2.113
4.905
11.688
26.653
77.213
291,059.300
124,254
124,381
124,417
124,390
124,360
124,436
124,424
124,383
124,415
124,279
4
Table 4
Volume and returns by liquidity group on high versus low spread days
This table presents sorts capturing the effect of spreads on stocks of different levels of liquidity. In the month prior to the observation month, all stocks are
sorted into ten groups based on ILLIQ (see Table 2), with Group 1 (10) consisting of the most (least) liquid stocks. The table reports average values of the selected
variables on month-by-month high and low spread days for each liquidity group. In particular, for each spread and for each month, we select the two days with
the highest spreads and average the selected variables over these two days for each liquidity group. We do the same for the two lowest spread days. (If, for a given
month and spread, there are more than two highest (lowest) spread days, then we average across all those days. If there is a single highest (lowest) spread day and
n second highest (lowest) spread days, we take a weighted average where the highest (lowest) spread day has a weight of 0.5 and the second highest (lowest) spread
days have a weight of 0.5/n.) For each spread, each variable, and each liquidity group, this yields two time series with monthly observations on high and low spread
days. The table reports the averages of these time series for each liquidity group. Panel A (B) classifies high and low spread days according to the Libor-OIS (TED)
spread. The Libor-OIS and TED spread sample periods are, respectively, December 4, 2001 to November 11, 2008 and January 2, 1986 to December 31, 2008. a,
b, and c denote statistical significance (two-tailed) at the 1%, 5%, and 10% level respectively. For all columns except Diff 1-10, statistical significance refers to the
difference between the variable on high and low spread days. In column Diff 1-10, a, b, and c indicate whether there is a statistically significant difference between
Groups 1 and 10. Variables are defined in the appendix. Returns are daily and in percent.
5
1
Panel A: Libor-OIS
HIGH
Normalized share volume 1.058b
Normalized dollar volume 1.054b
Return (equally weighted) -0.048
LOW
Normalized share volume
0.999
Normalized dollar volume
1.006
Return (equally weighted) 0.176
Panel B: Ted
HIGH
Normalized share volume 1.052a
Normalized dollar volume 1.050a
Return (equally weighted) 0.004
LOW
Normalized share volume
1.004
Normalized dollar volume
1.005
Return (equally weighted) 0.107
Liquidity Group
4
5
6
2
3
1.052
1.049
-0.041
1.056c
1.045
-0.073c
1.010
1.010
0.204
1.012
1.010
0.237
1.028
1.024
0.240
1.030
1.028
0.242
1.053
1.019
0.267
1.035
1.020
0.244
1.041b
1.040b
0.002
1.034c
1.029
-0.006
1.032
1.031
0.001
1.038
1.035
-0.013
1.034
1.035
-0.014
1.012
1.013
0.126
1.009
1.010
0.132
1.014
1.014
0.135
1.016
1.018
0.133
1.024
1.019
0.138
1.044
1.058
1.034
1.041
c
-0.047 -0.081c
1.093
1.071
-0.036c
7
8
9
10
Diff 1-10
t-statistic
1.031
1.087
0.019c
0.933a
0.922b
0.176c
0.125a
0.132a
-0.225c
3.76
2.95
-1.83
1.037
1.012
0.203
1.060
1.039
0.234
1.090
1.154
0.328
-0.091b
-0.148
-0.152
-2.34
-1.64
-1.26
1.04
1.044
0.016
1.028
1.052c
0.049
1.036
1.036
0.089
0.997c
1.004
0.359
0.056a
0.046b
-0.355a
3.22
2.29
-4.42
1.013
1.008
0.109
1.021
1.010
0.120
1.024
1.022
0.163
1.033
1.021
0.369
-0.030c
-0.016
-0.262a
-1.72
-0.77
-4.14
1.128
1.037
1.091
1.046
0.016 -0.016c
Table 5
Differences in volume and returns between high and low spread days
For each listed variable, this table reports the difference in its average value on high and low spread days, as reported in Table 4, for each liquidity group.
Panel A (B) classifies high and low spread days according to the Libor-OIS (TED) spread. The Libor-OIS and TED spread sample periods are, respectively,
December 4, 2001 to November 11, 2008 and January 2, 1986 to December 31, 2008. a, b, and c denote statistical significance (two-tailed) at the 1%, 5%, and
10% level respectively. For all columns except Diff 1-10, statistical significance refers to the difference between the variable on high and low spread days. In
column Diff 1-10, a, b, and c indicate whether there is a statistically significant difference between Groups 1 and 10. Variables are defined in the appendix.
Returns are daily and in percent.
6
1
Panel A: Libor-OIS
Normalized share volume 0.059b
Normalized dollar volume 0.048b
Return (equally weighted) -0.224
Panel B: Ted
Normalized share volume 0.049a
Normalized dollar volume 0.046a
Return (equally weighted) -0.103
2
3
0.042
0.039
-0.245
0.044c
0.035
-0.310c
0.029b
0.027b
-0.124
0.024c
0.019
-0.138
4
Liquidity Group
5
6
0.016
0.027
0.010
0.012
-0.287c -0.323c
0.017
0.018
-0.133
0.022
0.017
-0.146
7
8
9
10
Diff 1-10
t-statistic
0.039
0.052
-0.304c
0.093
0.071
-0.228
-0.001
0.034
-0.219c
-0.029
0.047
-0.215c
-0.157a
-0.232b
-0.152c
0.216a
0.280a
-0.072
4.04
2.81
-0.39
0.011
0.016
-0.152
0.027
0.036
-0.093
0.007
0.042c
-0.071
0.012
0.014
-0.074
-0.037
-0.017
-0.01
0.086a
0.062b
-0.093
3.28
2.03
-0.97
Table 6
Fama-MacBeth regressions: Difference in normalized share volume on high versus low
spread days regressed on ILLIQ
Each month, m, we run the cross-sectional regression
HSVOLG,m − LSVOLG,m = αm + βm × ILLIQG,m−1 + εG,m .
We then report the average of all the cross-sectional coefficient estimates (α̂m , β̂m ) with corresponding t-statistics. Standard errors are corrected using the Newey-West procedure with three lags.
HSVOLG,m − LSVOLG,m is the difference in month m between the normalized share volumes on
high versus low spread days for liquidity group G (see Tables 2 and 4). ILLIQG,m−1 is the equally
weighted average ILLIQ across stocks in liquidity group G in month m − 1. In specification (1) we
use the Libor-OIS spread to identify high and low spread days. Specification (2) uses the Ted spread
to identify high and low spread days. Specification (3) uses the Ted spread, but considers the time
period for which the Libor-OIS spread is available. Specifications (4) and (5) use the Libor-OIS and
Ted spread, respectively, but consider the period prior to the financial crisis (pre 07/2007). The
Libor-OIS and TED spread sample periods are, respectively, December 4, 2001 to November 11, 2008
and January 2, 1986 to December 31, 2008. a, b, and c denote statistical significance (two-tailed)
at the 1%, 5%, and 10% level respectively. Variables are defined in the appendix.
(1)
(2)
(3)
(4)
(5)
Spread / Period
Libor-OIS
Ted
Ted (OIS-period)
Libor-OIS (pre 07/2007)
Ted (pre 07/2007)
Intercept
0.034
0.023
0.019
0.026
0.019
t-stat ILLIQG
(1.21) -0.015c
(1.65) -0.004b
(0.66) -0.013b
(0.83) -0.017a
(1.50) -0.004b
7
t-stat
(-1.80)
(-2.28)
(-2.21)
(-2.79)
(-2.05)
Adj. R2
18.07%
14.28%
20.30%
16.49%
13.70%
N
83
275
83
66
257
Table 7
Regressions of Market share of volume on spreads
Each column represents a separate regression (using the Cochrane-Orcutt procedure), one for each liquidity group, G (see Table 2):
Market share of volume Gt
= α + β × spreadt + εt .
mean(Market share of volume G)
Daily data (t denotes an individual day). The Market share of volume is divided by its time series average to facilitate comparisons among groups. In
Panel A (B), the Libor-OIS (TED) spread is used. The Libor-OIS and TED spread sample periods are, respectively, December 4, 2001 to November 11,
2008 and January 2, 1986 to December 31, 2008. t-statistics are in brackets. a, b, and c denote statistical significance (two-tailed) at the 1%, 5%, and
10% level, respectively. Variables are defined in the appendix.
Panel A
Constant
Libor-OIS spread (%)
8
N
Adj R2
Panel B
Constant
Ted spread (%)
N
Adj R2
Liquidity Group
3
4
5
1
2
0.992a
(145.24)
0.027b
(2.55)
1715
0.003
1.007a
(92.14)
-0.024
(-1.16)
1715
0.000
1.025a 1.044a
(55.39) (34.00)
-0.093a -0.161a
(-2.79) (-3.16)
1715
1715
0.004
0.005
0.988a
(233.61)
0.021a
(5.47)
5647
0.005
1.036a
(96.36)
-0.064a
(-5.11)
5647
0.004
1.050a 1.054a
(60.07) (42.71)
-0.089a -0.097a
(-4.96) (-4.23)
5647
5647
0.004
0.003
6
7
8
9
10
1.04a
(25.36)
-0.146b
(-2.15)
1715
0.002
1.064a
(27.21)
-0.240a
(-3.30)
1715
0.006
1.084a
(25.66)
-0.320a
(-3.97)
1715
0.009
1.106a
(22.83)
-0.405a
(-4.41)
1715
0.011
1.114a
(12.34)
-0.438b
(-2.47)
1715
0.003
1.177a
(6.74)
-0.677b
(-1.96)
1715
0.002
1.070a
(41.53)
-0.125a
(-4.38)
5647
0.003
1.084a
(41.38)
-0.151a
(-4.67)
5647
0.004
1.102a
(38.02)
-0.183a
(-5.07)
5647
0.004
1.093a
(31.76)
-0.169a
(-3.93)
5647
0.003
1.089a
(23.17)
-0.160a
(-2.60)
5647
0.001
1.149a
(15.36)
-0.268a
(-2.64)
5647
0.001
Table 8
Libor-OIS, VIX, and control variables: descriptive statistics
Panel A is for the pre-crisis period, December 4, 2001 to July 29, 2007. Panel B is for the post TAF period, December 17, 2007 to November 11, 2008.
Daily data. Variables are defined in the appendix. TAF: term auction facility. bps: basis points.
Units
9
Panel A: Pre-crisis period
Libor-OIS
VIX
Group 1 Relative bid-ask
Group 2 Relative bid-ask
Group 3 Relative bid-ask
Group 4 Relative bid-ask
Group 5 Relative bid-ask
Group 6 Relative bid-ask
Group 7 Relative bid-ask
Group 8 Relative bid-ask
Group 9 Relative bid-ask
Group 10 Relative bid-ask
Lagmarket
Normalized market dollar volume
Correlation between VIX and Libor-OIS
Panel B: Post-TAF period
Libor-OIS
VIX
Group 1 Relative bid-ask
Group 2 Relative bid-ask
Group 3 Relative bid-ask
Group 4 Relative bid-ask
Group 5 Relative bid-ask
Group 6 Relative bid-ask
Group 7 Relative bid-ask
Group 8 Relative bid-ask
Group 9 Relative bid-ask
Group 10 Relative bid-ask
Lagmarket
Normalized market dollar volume
Correlation between VIX and Libor-OIS
Mean
Std Error
Std Dev
Median
Min
Max
N
bps
%
%
%
%
%
%
%
%
%
%
%
%
%
0.61
11.09
17.73
12.36
16.83
21.26
27.51
36.82
53.81
84.26
134.81
216.05
396.30
0.04
101.00
0.10
0.19
0.16
0.17
0.16
0.16
0.20
0.26
0.30
0.25
0.31
1.19
0.03
0.43
3.60
6.92
5.78
6.32
5.98
6.12
7.42
9.69
10.98
9.12
11.58
44.18
0.97
15.96
10.60
15.50
10.58
14.91
19.44
25.74
35.60
52.76
82.05
134.09
216.55
408.97
0.08
99.79
2.51
9.89
4.95
7.72
11.58
16.07
20.43
32.78
63.99
106.24
173.99
280.52
-3.82
27.87
28.40
45.08
39.18
44.32
45.91
51.02
59.11
82.47
113.70
165.03
265.59
485.78
5.32
226.78
1375
1375
1375
1375
1375
1375
1375
1375
1375
1375
1375
1375
1375
1375
bps
%
%
%
%
%
%
%
%
%
%
%
%
%
0.89
100.48
28.67
8.78
11.74
13.44
16.61
22.13
34.11
65.33
128.16
228.54
471.17
-0.21
100.70
4.91
0.90
0.19
0.19
0.20
0.22
0.29
0.38
0.59
0.88
1.07
2.46
0.15
1.17
73.68
13.50
2.79
2.91
2.93
3.34
4.36
5.68
8.89
13.13
16.07
36.97
2.24
17.58
73.93
23.82
8.49
11.50
13.36
16.35
21.82
33.71
63.70
126.44
225.76
479.32
-0.06
98.85
30.18
16.30
4.07
6.79
8.39
11.40
14.06
23.32
50.45
105.20
190.79
225.88
-8.99
34.31
366.33
80.06
27.78
32.50
36.63
47.48
62.22
89.36
134.35
175.45
274.56
540.98
11.52
179.91
225
225
225
225
225
225
225
225
225
225
225
225
225
225
Table 9
Regressions of Market share of volume on the Libor-OIS spread, VIX, and control variables
Each column represents a separate regression (run using the Cochrane-Orcutt procedure), one for each liquidity group, G (see Table 2):
Market share of volume Gt
= α + β1 × Xt + β2 × ResidualZt |Xt + Γ0 Wt + ηt .
mean(Market share of volume G)
Daily data (t denotes an individual day). We divide the market share volume of each liquidity group by its time-series average to facilitate comparison
between groups. ResidualZt |Xt is the residual (εbt ) from the following first step regression:
Zt = α + γ × Xt + εt ,
where Zt and Xt are either the VIX or the Libor-OIS spread. W is a vector of control variables (see below) and Γ is a vector of coefficients. Panel
A: X is the Libor-OIS spread, Z is the VIX, and we consider the pre-crisis period (December 4, 2001 to June 29, 2007). Panel B: X is the VIX, Z is
the Libor-OIS spread, and we consider the pre-crisis period. Panel C: X is the Libor-OIS spread, Z is the VIX, and we consider the post TAF period
(December 17, 2007 to November 11, 2008). Panel D: X is the VIX, Z is the Libor-OIS spread, and we consider the post TAF period. t-statistics are in
brackets. a, b, and c denote statistical significance (two-tailed) at the 1%, 5%, and 10% level, respectively. Variables are defined in the appendix.
Liquidity Group
3
4
10
1
2
Panel A: Pre-crisis period. X is the Libor-OIS spread.
Constant
0.874a
1.256a
1.473a
1.635a
(135.07) (66.13) (60.78) (46.22)
Libor-OIS (%)
0.787a -1.420a -2.747a -3.882a
(18.60) (-11.91) (-20.35) (-18.84)
ResidualVIX|Libor-OIS (%)
0.005a -0.009a -0.017a -0.027a
(18.48) (-11.49) (-19.08) (-18.94)
Lagmarket
-0.111b
0.182
0.413c
0.279
(-2.15)
(1.19)
(1.89)
(1.04)
Group Relative bid-ask
0.163a -0.313a -0.521a -0.498a
(6.77) (-4.86)
(-6.91)
(-5.68)
a
a
a
Norm. market dollar volume
0.018
-0.045
-0.057
-0.067a
(4.79) (-4.06)
(-3.69)
(-3.46)
N
1374
1374
1374
1374
Adj R2
0.334
0.186
0.411
0.341
Panel B: Pre-crisis period. X is the VIX.
Constant
0.857a
1.283a
1.515a
1.720a
(132.86) (68.66) (64.89) (49.42)
VIX (%)
0.006a -0.010a -0.020a -0.029a
(21.79) (-13.85) (-24.29) (-22.53)
ResidualLibor-OIS|VIX (%)
0.149a -0.314a -0.714a -0.748a
(3.61) (-2.64)
(-4.91)
(-3.61)
Control variables, their coefficients and t-statistics, N, and Adj R2 are
5
6
7
8
9
10
1.786a
(34.42)
-4.495a
(-15.03)
-0.034a
(-16.92)
0.778b
(2.16)
-0.611a
(-5.69)
-0.061b
(-2.36)
1374
0.281
1.945a
(26.57)
-5.492a
(-14.53)
-0.039a
(-16.28)
0.865
(1.63)
-0.562a
(-4.52)
-0.033
(-0.87)
1374
0.310
2.056a
(16.17)
-5.486a
(-10.63)
-0.043a
(-12.72)
0.380
(0.57)
-0.287c
(-1.86)
-0.203a
(-4.24)
1374
0.190
2.067a
(11.26)
-5.607a
(-9.35)
-0.042a
(-9.95)
1.001
(1.42)
-0.033
(-0.26)
-0.396a
(-7.71)
1374
0.125
2.417a
(3.98)
-5.829a
(-3.57)
-0.056a
(-4.95)
0.156
(0.08)
-0.027
(-0.10)
-0.704a
(-4.96)
1374
0.033
1.743
(1.25)
-6.896c
(-1.90)
-0.037
(-1.48)
4.691
(1.19)
0.199
(0.70)
-0.764a
(-2.67)
1374
0.009
1.919a
2.079a
2.239a
2.223a 2.722a
(37.00) (28.71) (17.99) (12.22) (4.50)
-0.036a -0.042a -0.045a -0.044a -0.054a
(-18.91) (-18.64) (-13.89) (-11.34) (-5.14)
-0.492c -0.900b
-0.383
-0.724
0.765
(-1.70)
(-2.38)
(-0.77)
(-1.25) (0.48)
by construction identical to those in Panel A.
1.778
(1.25)
-0.045c
(-1.89)
-2.547
(-0.79)
Table 9 – continued
11
Liquidity Group
1
2
3
4
5
6
7
8
9
Panel C: Post-TAF period. X is the Libor-OIS spread.
Constant
0.967a
1.082a
1.092a
1.100a
1.088a
0.836a
0.996a
0.950a
1.91a
(107.84) (38.42) (28.08) (18.88) (12.60)
(6.22)
(3.90)
(2.79) (3.49)
a
a
a
a
a
a
a
Libor-OIS (%)
0.024
-0.054
-0.085
-0.127
-0.136
-0.208
-0.259
-0.323a -0.238b
(6.89) (-4.89)
(-5.63)
(-6.08)
(-3.82)
(-5.33)
(-4.57)
(-4.06) (-2.05)
ResidualVIX|Libor-OIS (%)
0.001a -0.004a -0.005a -0.007a -0.009a -0.010a -0.014b
-0.007
0.001
(4.51) (-4.02)
(-3.71)
(-3.61)
(-3.17)
(-2.64)
(-2.44)
(-1.04) (0.18)
Lagmarket
0.009
0.057
-0.037
-0.158
-0.415
-0.581
-1.174
-0.553
0.039
(0.19) (-0.40)
(-0.21)
(-0.58)
(-1.15)
(-0.98)
(-1.11)
(-0.67) (0.04)
Group Relative bid-ask
0.032
0.014
0.151
0.120
0.335
0.857a
0.502
0.511b -0.049
(0.73)
(0.11)
(0.87)
(0.53)
(1.37)
(2.95)
(1.42)
(2.06) (-0.23)
Norm. market dollar volume
0.006
-0.029
-0.026
0.008
-0.029
0.077
-0.068 -0.281b -0.562a
(0.78) (-1.23)
(-0.90)
(0.18)
(-0.49)
(0.82)
(-0.43)
(-2.11) (-3.55)
N
224
224
224
224
224
224
224
224
224
2
Adj R
0.233
0.143
0.156
0.169
0.087
0.155
0.085
0.072
0.057
Panel D: Post-TAF period. X is the VIX.
Constant
0.949a
1.127a
1.154a
1.19a
1.196a
0.977a
1.183a
1.113a 1.991a
(98.08) (37.07) (27.49) (19.14) (12.86)
(6.99)
(4.70)
(3.34) (3.61)
VIX (%)
0.001a -0.003a -0.005a -0.008a -0.009a -0.012a -0.016a -0.017a -0.011c
(-8.00)
(-6.02)
(-6.59)
(-6.99)
(-4.68)
(-5.94)
(-4.98)
(-4.05) (-1.90)
ResidualLibor-OIS|VIX (%)
0.001
0.011
-0.008
-0.018
0.01
-0.048
-0.033 -0.218c
-0.260
(0.18)
(0.56)
(-0.32)
(-0.48)
(0.17)
(-0.66)
(-0.31)
(-1.79) (-1.54)
Control variables, their coefficients and t-statistics, N, and Adj R2 are by construction identical to those in Panel C.
10
2.639a
(4.33)
-0.345a
(-4.05)
-0.014c
(-1.81)
-1.708
(-1.54)
-0.173
(-1.54)
-0.485a
(-2.77)
224
0.084
2.856a
(4.44)
-0.020a
(-4.23)
-0.122
(-0.87)
Table 10
Regressions of Relative volume on Libor-OIS spread, VIX, and control variables over the pre-crisis
period
This repeats the regressions from Table 9, but with Relative volume in place of Market share of volume, as
the dependent variable. Each cell represents a separate regression. In particular, it reports the estimate of the
coefficient β1 (with t-statistics in brackets) from the regression (run using the Cochrane-Orcutt procedure):
Relative volume of Group G to Group Ht
= α + β1 × Xt + β2 × ResidualZt |Xt + Γ0 Wt + ηt ,
mean(Relative volume of Group G to Group H)
where G and H refer to liquidity groups, G > H. G is given by the column number and H by the the row
number. Daily data (t denotes an individual day). We divide Relative volume by its time-series average to facilitate
comparisons of β̂1 across specifications (different G’s and H’s). W is a vector of control variables (Lagmarket,
Group Relative bid-ask, and Normalized market dollar volume) and Γ is a vector of coefficients. ResidualZt |Xt is
the residual (b
εt ) from the first step regression:
Zt = α + γ × Xt + εt .
Panel A (B): X is the Libor-OIS spread (VIX) and Z is the VIX (Libor-OIS spread). The sample period is the
pre-crisis period (December 4, 2001 to June 29, 2007). a, b and c denote statistical significance (two-tailed) at the
1%, 5% and 10% level, respectively. Variables are defined in the appendix.
Panel A: Pre-crisis period. X is the Libor-OIS spread.
Group
2
3
4
5
6
a
a
a
a
1
-2.112
-3.411
-4.446
-5.059
-5.968a
(-13.36) (-19.76) (-18.35) (-15.17) (-14.08)
2
-1.442a -2.910a -3.745a -4.782a
(-12.38) (-17.22) (-13.77) (-13.77)
3
-1.636a -2.770a -3.760a
(-11.06) (-11.43) (-11.88)
4
-1.742a -2.744a
(-10.37) (-10.81)
5
-1.003a
(-4.64)
6
7
8
9
12
7
-5.956a
(-10.65)
-4.885a
(-9.90)
-3.895a
(-8.27)
-2.902a
(-7.27)
-1.214a
(-3.24)
-0.329
(-0.92)
8
9
10
a
a
-6.125 -6.144 -7.102c
(-10.04) (-3.70) (-1.92)
-4.939a -5.347a -6.575c
(-7.96) (-3.09) (-1.82)
-3.830a -4.456a -6.080c
(-6.57) (-2.81) (-1.70)
-2.756a -3.364b -5.209
(-5.17) (-2.31) (-1.53)
-1.326b -2.339c -4.843
(-2.56) (-1.83) (-1.49)
-0.035 -1.311 -3.469
(-0.07) (-1.07) (-1.18)
0.526 -0.331 -2.880
(1.04) (-0.30) (-0.98)
-0.749 -3.166
(-0.94) (-1.18)
-2.253
(-0.73)
Table 10 – continued
Panel B: Pre-crisis period. X is the VIX.
Group
2
3
4
5
a
a
a
1
-0.015
-0.024
-0.033
-0.039a
(-15.46) (-23.29) (-21.79) (-19.15)
2
-0.011a -0.023a -0.030a
(-15.59) (-22.47) (-17.85)
3
-0.014a -0.023a
(-15.59) (-15.40)
4
-0.014a
(-14.73)
5
6
7
8
9
13
6
-0.044a
(-18.05)
-0.038a
(-17.97)
-0.031a
(-15.77)
-0.022a
(-14.68)
-0.008a
(-6.63)
7
-0.047a
(-13.80)
-0.041a
(-13.36)
-0.035a
(-11.64)
-0.027a
(-11.17)
-0.015a
(-6.46)
-0.009a
(-3.95)
8
-0.048a
(-12.33)
-0.038a
(-9.52)
-0.031a
(-8.14)
-0.024a
(-7.04)
-0.012a
(-3.74)
-0.004
(-1.22)
0.003
(1.04)
9
10
a
-0.056 -0.046c
(-5.28) (-1.88)
-0.050a -0.044c
(-4.51) (-1.86)
-0.044a -0.042c
(-4.26) (-1.80)
-0.036a -0.036
(-3.87) (-1.60)
-0.026a -0.032
(-3.16) (-1.50)
-0.018b -0.025
(-2.23) (-1.29)
-0.01 -0.019
(-1.35) (-1.00)
-0.011b -0.020
(-2.24) (-1.14)
-0.010
(-0.47)
Table 11
Regressions of Market share of volume on the Libor-OIS spread, VIX, and control variables: NASDAQ stocks excluded
This table runs the same regressions as in Table 9, Panels A and B, but with NASDAQ stocks excluded from the whole analysis. Each column
represents a separate regression (run using the Cochrane-Orcutt procedure), one for each liquidity group, G (see Table 2):
Market share of volume Group Gt
= α + β1 × Xt + β2 × ResidualZt |Xt + Γ0 Wt + ηt .
mean(Market share of volume Group G)
Daily data (t denotes an individual day). We divide the market share volume of each liquidity group by its time-series average to facilitate comparison
between groups. ResidualZt |Xt is the residual (εbt ) from the following first step regression:
Zt = α + γ × Xt + εt ,
where Zt and Xt are either the VIX or the Libor-OIS spread. W is a column vector of control variables and Γ is a row vector of coefficients. Panel A:
X is the Libor-OIS spread, Z is the VIX, and we consider the pre-crisis period (December 4, 2001 to June 29, 2007). Panel B: X is the VIX, Z is the
Libor-OIS spread, and we consider the pre-crisis period. Panel C: X is the Libor-OIS spread, Z is the VIX, we consider the post TAF period (December
17, 2007 to November 11, 2008). Panel D: X is the VIX, Z is the Libor-OIS spread, and we consider the post TAF period. t-statistics are in brackets.
a, b, and c denote statistical significance (two-tailed) at the 1%, 5%, and 10% level, respectively. Variables are defined in the appendix.
14
Liquidity Group
1
2
3
4
5
6
7
Panel A: Pre-crisis period. X is the Libor-OIS spread.
Constant
0.844a 1.119a
1.299a
1.399a
1.511a
1.582a
1.805a
(113.28) (79.99) (63.15) (57.37) (47.07) (35.55) (30.15)
Libor-OIS (%)
0.905a -0.693a -1.525a -2.461a -3.067a -4.335a -5.363a
(18.62) (-8.40) (-11.62) (-15.88) (-15.26) (-15.02) (-16.47)
ResidualVIX|Libor-OIS (%)
0.005a -0.003a -0.008a -0.016a -0.021a -0.026a -0.034a
(16.28) (-6.35)
(-9.19) (-15.12) (-15.14) (-13.04) (-15.45)
Lagmarket
-0.056 -0.079
0.309
0.180
0.221
0.342
0.407
(-0.81) (-0.51)
(1.59)
(0.79)
(0.79)
(1.05)
(0.94)
Group Relative bid-ask
0.163a -0.010 -0.248a -0.189a -0.178a -0.067c -0.145a
(6.66) (-0.32)
(-5.73)
(-4.77)
(-4.19)
(-1.72)
(-2.84)
a
a
a
a
a
a
Norm. market dollar volume
0.035 -0.040
-0.072
-0.069
-0.103
-0.064
-0.089a
(7.36) (-3.88)
(-5.41)
(-4.46)
(-5.34)
(-2.83)
(-2.97)
N
1374
1374
1374
1374
1374
1374
1374
2
Adj R
0.346
0.083
0.196
0.278
0.248
0.175
0.237
Panel B: Pre-crisis period. X is the VIX.
Constant
0.833a 1.120a
1.308a
1.443a
1.574a
1.638a
1.894a
(113.64) (83.13) (65.39) (60.58) (49.60) (35.82) (31.66)
VIX (%)
0.006a -0.004a -0.010a -0.018a -0.023a -0.030a -0.039a
(20.64) (-9.03) (-12.27) (-18.56) (-18.18) (-15.90) (-18.85)
ResidualLibor-OIS|VIX (%)
0.268a -0.283a -0.549a -0.577a -0.625a -1.247a -1.320a
(5.45) (-3.15)
(-4.09)
(-3.63)
(-3.07)
(-4.72)
(-4.07)
2
Control variables, their coefficients and t-statistics, N, and Adj R are by construction identical to
8
9
10
1.725a
2.018a
(19.07) (15.66)
-5.015a -4.338a
(-12.73)
(-8.99)
a
-0.04
-0.038a
(-14.78) (-11.44)
0.210 -1.118c
(0.39)
(-1.65)
-0.006
-0.053
(-0.11)
(-1.10)
a
-0.158
-0.421a
(-4.25)
(-9.04)
1374
1374
0.176
0.146
1.480a
(3.70)
-2.816b
(-2.20)
-0.037a
(-4.31)
1.922
(1.10)
0.137c
(1.83)
-0.703a
(-5.86)
1374
0.048
1.898a
2.203a
(20.80) (17.23)
-0.041a -0.038a
(-16.67) (-12.53)
-0.294
0.125
(-0.73)
(0.25)
those in Panel A.
1.738a
(4.27)
-0.032a
(-3.99)
1.544
(1.25)
Table 11 – continued
15
Liquidity Group
1
2
3
4
5
6
7
8
9
Panel C: Post-TAF period. X is the Libor-OIS spread.
Constant
0.939a 1.082a
1.107a
1.107a
1.139a
1.174a
1.105a
1.271a
1.965a
(86.82) (52.38) (32.19) (29.38) (20.62) (16.13) (12.21)
(5.21)
(4.32)
a
a
a
a
a
a
a
a
Libor-OIS (%)
0.036 -0.045
-0.053
-0.061
-0.106
-0.142
-0.18
-0.194
-0.110
(9.40) (-6.00)
(-5.34)
(-4.60)
(-5.90)
(-4.99)
(-5.17)
(-3.95)
(-1.52)
ResidualVIX|Libor-OIS (%)
0.002a -0.001 -0.004a -0.006a -0.005a -0.007a -0.011a
-0.006
-0.002
(4.80) (-1.28)
(-4.18)
(-4.51)
(-3.02)
(-2.92)
(-3.93)
(-1.51)
(-0.24)
Lagmarket
0.002
0.029
0.028
-0.008
0.032
-0.421
-0.424
-0.697
0.982
(0.03)
(0.26)
(0.15)
(-0.04)
(0.12)
(-1.38)
(-1.17)
(-0.95)
(0.89)
Group Relative Bid-Ask
-0.014
0.059
-0.018
-0.028
0.146
-0.066
0.149
-0.109 -0.308c
(-0.34)
(0.69)
(-0.13)
(-0.23)
(1.03)
(-0.52)
(1.27) (-0.67)
(-1.76)
Norm. market dollar volume
0.026a -0.045b -0.051c
-0.040
-0.065
-0.008
-0.027
0.070
-0.130
(2.78) (-2.47)
(-1.79)
(-1.35)
(-1.53)
(-0.15)
(-0.45)
(0.62)
(-0.75)
N
224
224
224
224
224
224
224
224
224
Adj R2
0.339
0.146
0.166
0.150
0.152
0.124
0.144
0.099
0.012
Panel D: Post-TAF period. X is the VIX.
Constant
0.914a 1.105a
1.153a
1.166a
1.212a
1.271a
1.241a
1.385a
2.017a
(78.69) (49.74) (31.88) (29.10) (21.05) (16.49) (13.75)
(5.81)
(4.44)
a
a
a
a
a
a
a
a
VIX (%)
0.002 -0.002
-0.003
-0.004
-0.006
-0.008
-0.011
-0.011
-0.006
(10.4) (-5.93)
(-6.60)
(-5.96)
(-6.53)
(-5.63)
(-5.89)
(-4.12)
(-1.48)
ResidualLibor-OIS|VIX (%)
0.007
-0.03b
0.016
0.029
-0.021
-0.033
-0.006
-0.091
-0.083
(0.97) (-2.22)
(0.84)
(1.22)
(-0.63)
(-0.71)
(-0.11)
(-1.12)
(-0.63)
2
Control variables, their coefficients and t-statistics, N, and Adj R are by construction identical to those in Panel C.
10
2.19c
(1.72)
-0.270
(-0.87)
-0.011
(-0.59)
0.414
(0.22)
-0.194
(-0.84)
-0.044
(-0.14)
224
-0.017
2.362c
(1.70)
-0.015
(-0.96)
-0.091
(-0.23)