6ZLVV)LQDQFH,QVWLWXWH

5HVHDUFK3DSHU6HULHV1ƒ²

0RQH\DQG/LTXLGLW\LQ)LQDQFLDO
0DUNHWV


.MHOO*1<%25*

8QLYHUVLW\RI=XULFK6ZLVV)LQDQFH,QVWLWXWHDQG&(35


3HUg67%(5*

8QLYHUVLW\RI=XULFKDQG6ZLVV)LQDQFH,QVWLWXWH
Established at the initiative of the Swiss Bankers' Association, the Swiss
Finance Institute is a private foundation funded by the Swiss banks and
Swiss Stock Exchange. It merges 3 existing foundations: the International
Center FAME, the Swiss Banking School and the Stiftung "Banking and
Finance" in Zurich. With its university partners, the Swiss Finance Institute
pursues the objective of forming a competence center in banking and
finance commensurate to the importance of the Swiss financial center. It will
be active in research, doctoral training and executive education while also
proposing activities fostering interactions between academia and the
industry. The Swiss Finance Institute supports and promotes promising
research projects in selected subject areas. It develops its activity in
complete symbiosis with the NCCR FinRisk.

The National Centre of Competence in Research “Financial Valuation and

Risk Management” (FinRisk) was launched in 2001 by the Swiss National
Science Foundation (SNSF). FinRisk constitutes an academic forum that
fosters cutting-edge finance research, education of highly qualified finance
specialists at the doctoral level and knowledge transfer between finance
academics and practitioners. It is managed from the University of Zurich and
includes various academic institutions from Geneva, Lausanne, Lugano,
St.Gallen and Zurich. For more information see www.nccr-finrisk.ch .

This paper can be downloaded without charge from the Swiss Finance
Institute Research Paper Series hosted on the Social Science Research
Network electronic library at:

http://ssrn.com/abstract=
 Money and Liquidity in Financial Markets1

Kjell G. Nyborg Per Östberg

University of Zurich, University of Zurich
Swiss Finance Institute, and Swiss Finance Institute
and CEPR

June 2013
First draft: November 2009

1 We thank Zexi Wang for research assistance. We are also grateful for comments from semi-
nar participants at the Bank for International Settlements, Bank of Finland, Copenhagen Busi-
ness School, ESSEC, European Central Bank, ISCTE, Norges Bank, Norwegian School of Man-
agement, Stockholm School of Economics, Warwick Business School, and the Universities of
Aberdeen, Alabama, Amsterdam, Gothenburg, Porto, South Carolina, and Zurich 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 Fi-
nancial Markets, Freiburg, Germany, October 2010; Swiss Finance Institute Annual Meetings,
Zurich November 2011; FINRISK Research Day, Gerzensee, Switzerland, June 2011; WU Gut-
mann Symposium on Liquidity and Asset Management, Vienna, Austria, June 2011; Rothschild
Caesarea Center Conference, IDC, Israel, May 2012; 30th SUERF colloquium, States, Banks,
and the Financing of the Economy, Zurich, September 2012. We also thank Yakov Amihud,
Philipp Halbherr, Henrik Hasseltoft, Wolfgang Lemke, Loriana Pellizon, Cornelia Rösler, and
members of the external board of the Department of Banking and Finance, University of Zurich,
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 Zurich, Plattenstrasse 14, 8032
Zurich, Switzerland. Email: kjell.nyborg@bf.uzh.ch. Östberg: Department of Banking
and Finance, University of Zurich, Plattenstrasse 14, 8032 Zurich, Switzerland. Email:
per.oestberg@bf.uzh.ch.
 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. Empirical tests on the stock market are supportive. Tighter interbank markets
are associated with relatively more volume in more liquid stocks; selling pressure, espe-
cially in more liquid stocks; and transitory negative returns. We control for market-wide
uncertainty and in the process also contribute to the literature on portfolio rebalancing.
Our general point is that money matters in financial markets.

Keywords: money, liquidity, interbank and financial markets, liquidity pull-back, volume,
order imbalance, returns, portfolio rebalancing, Libor-OIS spread, VIX
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 finan-
cial 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 fi-
nancial 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 by examining stock market volume, order imbalance, and returns. Third, as a
byproduct of controlling for market-wide uncertainty in the testing of the liquidity pull-
back hypothesis, we document relations between uncertainty, stock liquidity, and trading
activity that may help shed light on how agents rebalance portfolios in response to fluctua-
tions 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 mone-
tary 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 posits and 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 require-

1
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 world
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
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 (2011) find evidence that interbank liquidity
networks, which are intended to overcome imperfections in the interbank market, are not
always effective. We expect tighter interbank markets to be associated with a higher level
of liquidity pull-back activity. This has implications for volume, order imbalance, and
returns.
The implication of liquidity pull-back on volume is cross-sectional in nature. In par-
ticular, 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 rela-
tive to that of less liquid assets.3 We expect this relation because when interbank markets
function well, banks that hold excess liquidity place this with banks that are short; whereas
when interbank markets work less well, banks that are short make use of the more indirect
route of liquidity pull-back to obtain the liquidity they need.
With respect to order imbalance, an immediate implication of the liquidity pull-back
hypothesis is that increased tightness in the interbank market puts negative pressure on the
aggregate order flow of financial assets. Since highly illiquid assets may not be suitable for
liquidity pull-back, for example due to discontinuities and a lack of depth in the order book,
we would expect to observe selling pressure especially in more liquid assets in response to
increased tightness in the interbank market.
Tighter interbank markets, ceteris paribus, also should be associated with offsetting
3
This is related to the observation made by Scholes (2000), writing on the topic of crises and risk
management and drawing on his experience from LTCM, that “[i]n an unfolding crisis, most market
participants respond by liquidating their most liquid investments first. . . ” because “[more] liquid markets
tend to be large and can handle large trading volumes relatively quickly.”

3
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. Or put in more standard terms, selling pressure will have a
negative effect on returns. The initial price impact from liquidity pull-back would be
expected to reverse, as the need for pull-back subsides. This differentiates the pull-back
hypothesis, with respect to returns, from an information based story. Cross-sectionally,
liquidity pull-back has an equalizing effect on returns. The logic is that in equilibrium the
marginal costs of converting assets into higher powered money should be equalized, as far
as possible, across assets.
We test the implications of the liquidity pull-back hypothesis on the CRSP universe of
stocks using the three month Libor-OIS spread as our main measure of tightness in the
interbank market. Some tests are also run with the TED spread, for which we have a
longer time series.4 The Libor-OIS spread is arguably a more precise measure of the state
of the interbank market, since it is the difference between two interbank rates, rather than
an interbank and a treasury rate. 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 largely confirmed in
the data: First, the market share of volume of highly liquid stocks is increasing in either
spread. Second and third, an increase in the Libor-OIS spread is associated with (i) an
increase in selling pressure, especially for more liquid stocks, as measured by changes in
4
Libor is London Interbank Offered Rate, OIS is overnight index swap, TED spread is 3 month Libor
less the 3 month treasury bill rate.

4
order imbalance, and (ii) negative returns that partially reverse within a few days.5 Fourth,
for the portfolios consisting of the 40-60% most liquid stocks, which comprise roughly 97-
99% of daily volume on average, the magnitudes of the initial price reactions and reversals
are similar. These results are consistent with the liquidity pull-back hypothesis, especially
for more liquid stocks.
In testing the implications of the liquidity pull-back hypothesis, we control for market-
wide uncertainty, as measured by the VIX, and other factors.6 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; and Calvet, Campbell, and Sodini, 2009; for evidence on portfolio rebalancing.)
We find that the market share of volume of more liquid stocks is 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.
This conclusion is supported by our findings on order imbalance and returns. An
increase in the VIX is associated with increased selling pressure and negative returns, just
as for an increase in the price of liquidity. We do not find the return reversal effect with
the VIX that we see with the Libor-OIS spread. The impact of the VIX is instantaneous.
This supports the view that tightness in the interbank market is fundamentally a distinct
phenomenon from market-wide uncertainty.
While we use it as a general measure of tightness in the interbank market, the Libor-
OIS spread can be viewed more specifically as a measure of the price of liquidity. A “Libor
5
The order imbalance and return analysis is carried out using the Libor-OIS spread only, in part to
limit the number of tables and in part because the Libor-OIS spread is arguably a better measure of
interbank market tightness than the TED spread.
6
For information about the VIX, see http://www.cboe.com/micro/VIX/vixintro.aspx.

5
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.7 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 (2012) 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.8 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
be triggered by unrest in the interbank market, for example arising from extreme adverse
7
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.
8
Adverse selection may also lead to credit rationing along the lines of Jaffee and Russell (1976) or
Stiglitz and Weiss (1981).

6
selection in that market. Both of these perspectives may well be relevant for understanding
the collapse in the stock markets during the crisis.
It bears emphasis, however, that the liquidity pull-back hypothesis is not fundamentally
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.9
This paper is related to several other literatures. Liquidity pull-back is a poten-
tial contributor to the commonality in liquidity found by Chordia, Roll, and Subrah-
manyam (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 and to the literature on the associations between trad-
ing activity, liquidity, and stock returns (e.g. Chordia, Roll, and Subrahmanyam, 2002).
The rest of this paper is organized as follows. Section 2 describes the main data,
provides descriptive statistics, and defines the volume measures that we subsequently
use in our tests. Section 3 studies volume, testing the liquidity pull-back and portfolio
rebalancing hypotheses. Section 4 introduces the data used to measure order imbalance
and provides an analysis of order imbalance and returns. Section 5 concludes.

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
between 3 month USD Libor and the 3 month USD overnight index swap rate.10 Libor is
9
For information on TAF, see http://www.federalreserve.gov/monetarypolicy/taf.htm.
10
Libor is downloaded from http://www.bbalibor.com. The OIS rate is obtained from Reuters.

7
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.11 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.
Graphs of the Libor-OIS and TED spreads over their respective sample periods are
in Figures 1 and 2. As seen in Figure 1, the two spreads reacted similarly to the cri-
sis, increasing sharply in August 2007 and again in the wake of the Lehman Brothers
bankruptcy in September 2008. The correlation between the two spreads in our sample is
0.92. Figure 2 confirms that the TED spread tends to spike in times of turmoil when finan-
cial markets often are depressed, such as when the stock market crashed in October 1987.
This is suggestive of there being a connection between the state of the money markets
and the broader financial markets.12 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 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 over the whole
sample periods as well as over crisis and non-crisis dates. 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
11
The T-Bill data is obtained from http://www.federalreserve.gov/releases/h15/data.htm.
12
For example, the recent crisis saw not only high spreads, but also declining asset prices. From 1
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.

8
that there is substantial volatility in these spreads.13

2.2 Stock market data

The main stock market data used in this paper 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 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 change cusip, ticker, or exchange are also excluded for
the year of the change. This leaves us with an average of 4,564 individual stocks per day
over the whole sample period and 3,786 over the Libor-OIS sample period. The change in
exchange removal criterion is used to minimize the impact that any market microstructure
changes might have on the stock. As a robustness check, some of our analysis is carried
out without NASDAQ stocks. In Section 4 we also employ the NYSE’s TAQ database for
the purpose of studying order imbalance.

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. find that the best performer is Amihud’s (2002)
ILLIQ, which we therefore employ in this paper.
13
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.

9
 We measure ILLIQ on a monthly basis. For stock i and month j, ILLIQ is defined as:
 
|rit |
ILLIQij = Average (1)
t∈monthj Volume it

where, |rit | and Volumeit 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.14 This exclusion
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 the previous month.
Group 1 consists of the 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-to-
month stability of the groups, by providing month-by-month transition frequencies from
one group to another over the Libor-OIS sample period. 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 90.90%, 60.28%,
and 79.96%, 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
over the Libor-OIS sample period. The pooled sample has a mean and median of 4.807
and 0.016, respectively. 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, a volume of 1 million dollars
14
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
the database where there is no recorded closing price, thus suggesting an absence of trade, yet there is
positive recorded volume.

10
implies a price change of 1.6% for the median firm in our sample over the Libor-OIS
period and 10.4% in Goyenko et al.’s sample.15 There is substantial variation across
groups. From Group 1 to 9, the average ranges from 0.000 to 2.357, with Group 10 having
a mean ILLIQ of 44.911.16 For the most part, we only use stocks’ ILLIQ values to classify
them into groups. In this paper, it is therefore mainly the ordinal, not cardinal, accuracy
of ILLIQ that matters.17

2.4 Liquidity group volume measures

To study variations in relative volume across different groups, we define the following two
group level volume measures:

1. 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.

2. 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.

Summary statistics for different liquidity groups’ market share of volume are provided
in Table 3, Panel B, for the Libor-OIS sample period. The average market share of volume
of Group 1 is 75.171% and the four largest groups typically account for 97.244 % of volume
15
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) They use a random sample of 400 firms, while we use the entire CRSP
database. (iii) We consider different time periods. Over the TED spread sample period, we get a mean
and median ILLIQ of 17.632 and 0.134, respectively.
16
Using six decimal places, the mean ILLIQ of Group 1 is 0.000206 with a standard error of 0.000001.
17
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 9. Footnote 29 discusses this further.

11
on a given day. While the four most liquid groups are thus substantially larger in terms
of volume than the other groups, there is a fair amount of variation from day to day; the
standard deviation of the market share of volume of Group 1 is 4.194%.

2.5 Correlations

The main analysis in this paper is 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 CRSP equally weighted market return
are -0.17 and -0.10, respectively, which are also consistent with the liquidity pull-back
hypothesis. In the next two sections we will test the hypothesis more fully.

3 Volume
In this section, we test the implications of the liquidity pull-back hypothesis by initially
running simple regressions of the market share of volume on the price of liquidity, as
measured by the Libor-OIS and TED spreads. We then examine the robustness of our
basic findings by expanding the analysis in a variety of ways. First, to examine whether
our findings are driven by the crisis, we run separate regressions over the pre-crisis and
crisis periods. Second, 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. Third, to examine whether

12
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. Fourth, we redo the
main analysis without NASDAQ stocks. Fifth, we run Fama-MacBeth regressions using
within-month variations in volume on high and low spread days. This can be seen as a
robustness check on the time-series methodology employed in the other subsections.

3.1 Simple regressions: Market share of volume on the spreads

For each liquidity group G, we run the following time-series regression using daily obser-
vations over the full sample periods:

Yt = α + β1spreadt + β2Yt−1 + εt (2)

where spreadt is either the Libor-OIS or TED spread on date t, liquidity groups are formed
based on individual stocks’ ILLIQ values the previous month, and

Market Share of Volume Group Gt

Yt = . (3)
mean(Market Share of Volume Group G )

We mean-adjust the market share of volume for each group by its time series average so
as to facilitate comparisons across liquidity groups. We run separate OLS regressions with
Newey-West (1987) standard errors for each spread.18 As outlined in the Introduction,
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
18
Throughout this paper, the number of lags used to calculate Newey-West standard errors is set to the
closest integer to the fourth root of the number of observations, as suggested by Greene (2012) (p. 960).
Most of the volume regressions have also been run with the Cochrane-Orcutt procedure, with very similar
results. Results without the lagged dependent variable are even more supportive of the liquidity pull-back
hypothesis than the results reported here. 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 test identifies a structural break in August 2007, which is also when visual inspection reveals
a sharp increase in these spreads. Details are available upon request.

13
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 4. For both spreads, the coefficient on the
spread is positive (0.003) and statistically significant at the 1% level for the most liquid
group (Group 1). Thus, the market share of volume is larger for the most liquid stocks when
the interbank price of liquidity is high as compared to when it is low. Given that Group 1
accounts for approximately 75% of volume on the average, this is fairly strong evidence
in support of the liquidity pull-back hypothesis.19 For either spread, the coefficients are
negative for all other groups and, with one exception, they are statistically significant
(1% level). Furthermore, these coefficients are decreasing (almost monotonically) in the
illiquidity ranking of the groups, which gives additional support to the liquidity pull-back
hypothesis.

3.2 Multivariate regressions, different time periods, and alter-

native 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 the period after the introduction of TAF (term
auction facility), December 17, 2007 to November 11, 2008, which we will refer to as the
post-TAF period. 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.
19
We have also subdivided Group 1 into three subgroups based on ILLIQ and rerun the regressions in
Table 4 using Group 1 as “the market.” We find the same result; the market share of volume of the most
liquid group is higher when the price of liquidity is higher. See also footnote 26. Details are available
from the authors upon request.

14
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 in most of the remainder of the
paper, we focus on the Libor-OIS spread only.20
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.21
That the alternative portfolio rebalancing hypothesis should be taken seriously is un-
derscored 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.
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, might also be expected
to 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
20
As discussed in the Introduction, the Libor-OIS spread is a more accurate measure of the price of
liquidity than the TED spread.
21
The VIX is calculated using the methodology introduced in 2003 and downloaded from the CBOE.
See http://www.cboe.com/micro/VIX/vixintro.aspx.

15
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 cor-
relation between the Libor-OIS spread and the VIX, it is difficult to distinguish between
the liquidity pull-back 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
regression of interest:

Yt = α + β1 Xt + β2ResidualZ|X,t + β3Yt−1 + Γ0Wt + ηt , (5)

where Y is the mean-adjusted market share of volume, as defined in (3), W is a vector

of control variables (discussed below), and Γ is a vector of the corresponding regression
coefficients. Estimation is performed using OLS, with Newey-West standard errors. 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
22
See also Pastor and Stambaugh (2003) for a discussion of liquidity risk.

16
and the VIX is used as X or Z. In particular, we should see a positive coefficient on X
for Group 1 and a negative coefficient for less liquid groups. 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|VIX. 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 Libor-
OIS spread be X and require the coefficients on ResidualVIX|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) The rate of return on
the value weighted CRSP market portfolio from the previous day. Past returns 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.23 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,
23
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%.

17
defined as the aggregate dollar volume of the market divided by the average aggregate
volume over the previous five days. This is a simple control for deviations in the aggregate
market volume from its short run average. Since Group 1 is much larger in terms of volume
than other groups, this variable 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|VIX to have statistically significant Step 2 regression coefficients.
Summary statistics of the VIX and the three other control variables are in Table 5,
broken down into the pre-crisis and post-TAF periods. The table also reports on the
correlations between the Libor-OIS spread 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 6, which reports the second step regression output from the
four constellations described above. Panel A is for the pre-crisis period with the Libor-
OIS 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 reports that
the coefficient on the Libor-OIS spread for Group 1 is positive; 0.292 with a t-statistic
of 10.26. This supports the liquidity pull-back hypothesis. The coefficients for the other
groups are negative and decreasing in illiquidity, as in the previous subsection. For Group
10, the coefficient (t-statistic) is -2.428 (-3.02). 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 1.05%
(8.75%) 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|VIX is
positive and statistically significant (at the 1% level) for Group 1. The coefficient is
statistically significantly negative for Groups 2-4. Recalling from Table 3 that the top four

18
groups represent 97.24% of volume on a typical day, 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 sig-
nificant coefficients on the VIX that are decreasing in the group number. The coefficient
(t-statistic) is 0.207 (10.61) and -1.988 (-3.08) for Group 1 and 10, respectively. In terms of
economic significance, a 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 1.43% (13.76%) 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 ResidualVIX|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 the stronger
effect (in the sense that the coefficients on ResidualVIX|Libor-OIS have larger t-statistics
than those on ResidualLibor-OIS|VIX ). But this does not diminish the positive results on
the liquidity pull-back hypothesis.24
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, positive
for Group 1 and negative for the other groups. But as seen in Panel D, the coefficients
on ResidualLibor-OIS|VIX are not statistically significant over the post-TAF period for any
group. Thus, there is only weak evidence of liquidity pull-back over this period. This is
consistent with the view that the loose monetary policy over this period reduced the need
for liquidity 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 coefficients on the VIX are statistically significant, Group 9 excepted,
over the post-TAF period and are decreasing in the liquidity group number. Panel C shows
24
We have verified that the results from Panel A and Panel B of Table 6 are not driven by outliers. Coef-
ficients and p-values are quantitatively similar and qualitatively the same when winsorizing the dependent
and independent variables at the 1st and 99th percentiles.

19
that the coefficients on ResidualVIX|Libor-OIS are statistically significant for all groups
except number 9 and exhibit a decreasing trend. Thus, basic portfolio rebalancing in
response to market-wide uncertainty is unaffected by the looser monetary policy in the
post-TAF period, while liquidity pull-back weakens, and possibly disappears.25
In conclusion, the results in Table 6 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 reduced and maybe even eliminated liquidity pull-back,
(iii) the market share of volume of the most 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.26
25
Brown, Carlin, and Lobo (2010) show that it is theoretically possible that an agent who may need to
sell financial assets to obtain liquidity over several periods sells more illiquid assets first, if price impact
is convex in the quantity sold and future liquidity shocks are potentially sufficiently large. This raises the
prospect of an alternative interpretation of the weaker association we observe between the market share
of volume of the most liquid stocks and the Libor-OIS spread during the crisis, namely that agents chose
to liquidate illiquid assets to a larger degree than before as a result of potentially larger liquidity shocks
during the crisis. While the finding that the market share of volume is increasing in the VIX also in the
post-TAF period may speak against this alternative explanation, we have examined it more specifically by
including the volatility of the Libor-OIS spread among the control variables in the regressions in Table 6.
Doing this for the pre-crisis as well as the post-TAF periods, we find essentially nothing, and the results in
Table 6 are not affected. Thus, we do not find support for the idea that the potential for higher liquidity
shocks in the interbank markets imply relatively more pull-back on less liquid stocks.
26
We have also subdivided Group 1 into three subgroups, calculated the market share of volume of
each subgroup as a fraction of Group 1’s total volume, and rerun the regressions in Table 6 with these
subgroup market share of volumes. We see the same pattern; the coefficient on the Libor-OIS spread, the
VIX and the respective residuals from the first step regression are positive (and statistically significant)
for the first subgroup, and negative for the two other subgroups (and always statistically significant for
the third subgroup). Details are available from the authors upon request.

20
3.3 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, 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 indicate the strongest support for the
liquidity pull-back hypothesis. Using Relative volume also provides a straightforward way
to re-examine and test the (near) monotonicity results of the previous subsections. If
stocks are equally subject to liquidity pull-back across liquidity groups, we should see
negative coefficients on the Libor-OIS spread, since 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. Negative coefficients on the VIX would be indicative of portfolio rebalancing.
Table 7, Panel A (Panel B) exhibits the results for the Libor-OIS spread (VIX). With
one exception, the coefficient on the Libor-OIS spread is statistically significantly (5% or
better) negative for all pairs when either the less liquid group, G, is less than or equal
to 6 or the more liquid group, H, is 4 or less. The exception occurs when H = 4 and
G = 9, where we have a negative, but not significant, coefficient. In the 45 complete set
of regressions, the coefficient on the Libor-OIS spread is positive in only six cases, with
only one of these being statistically significant (Group 8 to 6). In the case of the VIX,
all but five coefficients are negative and statistically significant at least at the 5% level or
better. This is strong evidence that relatively more liquid stocks experience a relatively
higher volume as the price of liquidity in the interbank market or market-wide uncertainty
increase. These findings 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). Our results are consistent with liquidity pull-back
being part of the day-to-day activity in the markets, especially for the 40-60% more liquid
stocks, representing approximately 97-99% of daily volume on average.

21
3.4 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 reported volume
of a stock that switches from NASDAQ to NYSE or AMEX often falls. This is due to
the dealer structure on NASDAQ which leads to a potential double counting problem of
trading volume.
To examine whether our results are driven by NASDAQ stocks, we construct new
liquidity groups for each month, comprised only of NYSE and AMEX stocks, and rerun
the two-step procedure described in Subsection 3.2. For the sake of brevity, we only report
the regression output for the pre-crisis period, which is shown in Table 8. As seen, the
results are qualitatively the same as for the full sample in Table 6. So our results are
robust to the exclusion of NASDAQ stocks.27

3.5 Volume on high versus low spread days within months across
liquidity groups

This subsection provides a robustness check on the time-series approach above by running
Fama-MacBeth (1973) regressions of liquidity groups’ within-month differences in volume
on high versus low spread days on their corresponding average ILLIQ (calculated from the
previous month, as always). We proceed, separately for the Libor-OIS and TED spreads,
as follows: First, for each month: (i) We select the two days with the highest and the two
days with the lowest spreads. (ii) For the two high, as well as for the two low, spread days,
we average the values of each liquidity group’s normalized share volume, defined on day t
as
Volume Group Gt
Normalized share volume Group Gt = P5
. (6)
i=1 Volume Group Gt−i /5
27
While not shown here, 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. Details are available from the authors
upon request.

22
In this way, for each group, we generate two time series with monthly observations, cor-
responding to the normalized share volume’s within-month average value on high and low
spread days, respectively.28 Unlike the market share of volume and the relative volume
measures used above, the normalized share volume of a group on a given day is not directly
linked to the volume of other groups. Thus, this measure can be “high” simultaneously
across groups.
Second, we use the Fama-MacBeth procedure. In particular, we run the following
cross-sectional regression for each month, m:

HSV OLG,m − LSV OLG,m = αm + βm ILLIQG,m−1 + εG,m (7)

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 Group G in month m − 1. We then average the coefficients from
the monthly regressions. Standard errors are calculated using the Newey-West procedure.
The Fama-MacBeth procedure is run separately for the Libor-OIS and TED spreads
over their respective sample periods, with and without the crisis. For the TED spread, we
also run the procedure separately over the Libor-OIS sample period.
Table 9 reports the results. Consistent with the liquidity pull-back hypothesis, the
coefficient on ILLIQG is negative regardless of which specification we look at. 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
28
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 normalized share volume on high spread days. 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. The monthly observation of the high
spread day is then 0.5 times the normalized share volume 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. We have also run the analysis in this section with normalized dollar volume (defined the analogous
way to normalized share volume), with similar results, though somewhat weaker for the TED spread.

23
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 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 support the results from our time-series
regressions that there is a connection between the interbank market for liquidity and the
stock market over the pre-crisis period. Furthermore, this connection is consistent with
the liquidity pull-back hypothesis.29

4 Returns and order imbalance

Our main objective in this section is to test the liquidity pull-back hypothesis further by
examining the impact of the price of liquidity on returns and order imbalance. We are
interested in studying both contemporaneous and lagged effects because, if our results on
volume are indeed driven by liquidity pull-back, we would expect to see an increase in the
Libor-OIS spread being associated with negative returns that at least partially reverse.
We should also see temporary selling pressure. As a control, and to further study the
portfolio rebalancing hypothesis, we are also interested in the relation between the VIX
and both returns and order imbalance.
To take account of the interdependence of returns and order imbalance (Chordia, Roll,
and Subrahmanyam, 2002; and Chordia and Subrahmanyam, 2004), we use a vector au-
toregression (VAR) approach as advocated by Chordia, Sarkar, and Subrahmanyam (2005)
in their study of the dynamics of stock and bond market liquidity. We look at the market
as a whole and the same ten liquidity groups as before. Stock returns are from CRSP
and order imbalance is calculated using the NYSE’s Trade and Quote (TAQ) database.
29
We have also run the regressions in Table 9 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. Details are available upon request. The
regressions have not been run separately for the post-TAF period since this only leaves us with 11 data
points in the second step, due to the monthly measurement interval.

24
To minimize the number of tables, the analysis is carried out only over the Libor-OIS
pre-crisis sample period (December 4, 2001 to June 29, 2007).

4.1 Data, variable definitions, and descriptive statistics

Table 10, Panel A presents descriptive statistics of daily value weighted rates of return of
the ten liquidity groups and the market (all groups). Illiquid stocks offer higher returns
(Amihud and Mendelson, 1986; Amihud, 2002). The mean return increases as we go from
Group 1 (0.031%) to Group 10 (0.079%).
Corresponding statistics on daily changes in order imbalance are in Panel B. Before
discussing these, we outline how order imbalance is calculated. We use intraday trades and
quotes from TAQ during normal trading hours (09:30 to 16:00). We only consider trades
that have positive prices and volumes and have not been flagged as corrected; and quotes
for which the bid price, offer price, bid size, and offer size are not missing. Delayed and
bunched trades are excluded; as are quotes that are flagged to have occurred during halted
trading. We draw on Chordia, Roll and Subrahmanyam (2008) in merging the trades with
the quotes. The matching quote is the first quote prior to the trade. We apply the Lee
and Ready (1991) algorithm to the matched trade-quote sample to determine whether
a particular trade was buyer or seller initiated. Finally, from the resulting sample, we
exclude all stock-days with when-issued securities (predominantly relating to stock splits
and reverse splits).30
Having selected and categorized trades using these standard procedures, the order
imbalance for the market as a whole is then calculated as a fraction of all signed orders,
along the lines of Chordia and Subrahmanyam (2004):
buymarket,t − sellmarket,t
OIBmarket,t = (8)
buymarket,t + sell market,t
where buymarket,t and sellmarket,t is the total dollar volume of all stocks present in our
TAQ sample on day t that has been signed as buyer and seller initiated, respectively.
To calculate the order imbalance of the liquidity groups, we merge our TAQ sample
with the liquidity rankings. The resulting sample has 5,228,847 stock-day observations
30
This exclusion restriction amounts to roughly 0.07% of the TAQ sample.

25
over the Libor-OIS pre-crisis sample period. CRSP and TAQ do not overlap perfectly,
meaning that for some stocks (that appear in a liquidity group), there are days without
TAQ data. For Group 1, the average number of stocks per day with TAQ data is 385.00.
For Group 10, it is 315.51. By way of comparison, the average number of stocks per
liquidity group (based on CRSP) per day is 385.67. Across all groups, the minimum
number of stocks with TAQ data on a given day is 189 (Group 10). The order imbalance
for a liquidity group is calculated using (8), but with “market” replaced by the relevant
liquidity group.
As noted by, e.g., Chordia, Roll and Subrahmanyam (2002), order imbalance is persis-
tent. Furthermore, as documented by O’Hara, Yao, and Ye (2012), the exclusion of odd-lot
trades in the TAQ database leads to a time-varying bias. We therefore first difference order
imbalance.
Table 10, Panel B, provides descriptive statistics of ∆OIB, expressed in percent, over
the Libor-OIS pre-crisis period. ∆ is the first difference operator. The mean of the market
is −0.005% and for Groups 1-4 the mean is in the range −0.006% to −0.002%, reflecting
a slight downward drift in order imbalance over the sample period. For Groups 6-10, the
mean ranges from 0.001% to 0.029%. But neither for the market as a whole or any liquidity
group is the mean statistically different from zero.

4.2 Vector autoregression: Specifications and results

The main variables of interest are returns and changes in order imbalance, for the market
as a whole and for the ten liquidity groups. In our VAR specifications, we therefore also
first difference the other variables. In particular, we run eleven VAR systems, each with six
endogenous variables Return G , ∆OIBG , ∆Libor-OIS, ∆VIX, ∆VolumeG , and ∆Bidask G ,
where G refers to the market (M) or a liquidity group (1 through 10). The control variable
Volume G is the dollar volume of Group G, and Bidask G is the dollar volume weighted bid-
ask spread (as a percentage of the closing price) across stocks in G.31
31
We use volume as a control because of the evidence that there is a relation between returns and volume
(e.g. Gallant, Rossi and Tauchen, 1992). Unlike in Section 3, we do not normalize volume by its five day

26
 Lag lengths are set separately for each of the eleven VAR systems, using the Hannan-
Quinn (HQ) criterion. As discussed by Lütkepohl (2007), this criterion is strongly con-
sistent and in simulation appears to strike a balance between too many lags (Akaike’s
criterion) versus too few (Schwarz’ criterion). We discuss the results for the market as a
whole and the individual liquidity groups in turn.

4.2.1 The market

For the market as a whole, the HQ criterion selects four lags. We discuss in turn the
correlations of the innovations in the VAR system, which reflect contemporaneous effects;
Granger causality tests, which reflect lagged effects; and, finally, impulse response functions
(IRFs), which trace out contemporaneous and intertemporal effects.
Table 11, Panel A, contains the VAR innovation correlation matrix. With three excep-
tions, all correlations are significantly different from zero (at the 1% level), indicating that
there are contemporaneous effects among the variables. Shocks to ∆Libor-OIS are signif-
icantly negatively associated with return and ∆OIB innovations. The sign of these two
correlations remains negative if we drop other variables from the system. This suggests
that there is a negative contemporaneous relation between ∆Libor-OIS and both returns
and ∆OIB, which is in line with the liquidity pull-back hypothesis. We find the same for
shocks to ∆VIX, which supports the portfolio rebalancing hypothesis. That both of these
effects seem to be present in the data is consistent with our findings in Section 3.
With respect to lagged effects, Panel B of Table 11 reports on pairwise Granger-
causality tests between the endogenous variables. No variable Granger-causes market
returns. But ∆Libor-OIS has the lowest p-value, 0.26. Relevant for the liquidity pull-
back hypothesis, ∆Libor-OIS Granger-causes ∆OIBM (p-value 0.09). In contrast, ∆VIX
average, since we first difference it instead. We report results using group volume, but the results with
market volume are similar. We use the bid-ask spread as a control variable because there is evidence that
bid-ask spreads and returns are related (e.g., Amihud and Mendelson, 1986). In Section 3.2, we calculated
a group’s bid-ask spread relative to that of other groups because we looked at relative volume measures.
In this section, we use the more standard bid-ask spread as a fraction of the closing price. This slight
modification does not qualitatively affect the results.

27
Granger causes neither returns nor ∆OIBM . The evidence thus far therefore suggests
that changes to the VIX are instantaneously reflected in the market, while those to the
Libor-OIS spread play out over more than one day.
Despite the result in Table 11, Panel B that ∆Libor-OIS does not Granger cause mar-
ket returns, there is other evidence that lagged values of ∆Libor-OIS do affect returns.
First, in the Return M regression (details omitted for brevity, but are available from the
authors upon request), the VAR coefficient on ∆Libor-OISt−3 is statistically significant at
the 10% level; the coefficient is 0.040 with a t-statistic of 1.75 using Newey-West stan-
dard errors with 6 lags. This is robust to dropping other variables from the system or
changing the lag structure. No other lags of ∆Libor-OIS or any of the other variables have
coefficients that are statistically significant at conventional levels. The reason for this
apparent inconsistency between the Granger causality Wald test and the VAR coefficient
t-test may be that the former lacks power due to the noise induced by the inclusion of
all four lags of ∆Libor-OIS. If we were to drop the fourth lag from the VAR, the p-value
of the corresponding Wald test is 0.109. It turns out that in the Group 1 VAR system,
the HQ criterion actually selects three lags. Furthermore, as shown in Table 12 (which
we discuss in more detail in the next sub-subsection), we find that ∆Libor-OIS Granger
causes Return 1 at the 10% level. These findings support the view that there is a significant
relation between lags of ∆Libor-OIS, in particular the third lag, and returns. To take a
closer look at this as well as contemporaneous effects, next we study the IRFs for the
market.
Figure 3 graphs the impulse response functions of Return M (left column) and ∆OIBM
(right column) from a one standard deviation shock to ∆Libor-OIS (upper row) and ∆VIX
(lower row), with 10% confidence bands. The IRFs are computed by orthogonalizing the
covariance matrix of the VAR innovations using the Cholesky decomposition with the
following ordering of the endogenous variables: ∆Libor-OIS, ∆VIX, ∆OIBM , ∆VolumeM ,
∆Bidask M , Return M .32
32
Different orderings yield similar results. For example, if we change the ordering between ∆VIX and
∆Libor-OIS, the effects of a shock to ∆Libor-OIS on Return M and ∆OIBM are qualitatively similar.
The contemporaneous effect on returns is still statistically significantly negative and is followed by a

28
 We focus first on the effect of a shock to ∆Libor-OIS. As seen in Figure 3, a positive
shock to ∆Libor-OIS is associated with a statistically significant contemporaneous nega-
tive market return and ∆OIBM . In other words, an increase in the price of liquidity is
associated with lower asset prices and increased selling pressure. In terms of economic
significance, the contemporaneous impulse response of the market return is −0.122% from
a one standard deviation shock to ∆Libor-OIS. In absolute value terms, this is more than
three times the average daily market return of 0.037% (Table 10). This negative con-
temporaneous effect is consistent with the liquidity pull-back hypothesis. An alternative
hypothesis is that increases in the Libor-OIS spread reflect negative information. How-
ever, the significant partial price reversal (in period 3) seen in the IRF of ∆Libor-OIS on
Return M , points to that the initial price decline from an increase in the Libor-OIS spread
is driven at least in part by liquidity pull-back.
With respect to the IRF of ∆OIBM , the negative contemporaneous reaction from a
shock to ∆Libor-OIS is followed by an equal but opposite reaction in period 1. This is to
say that the level of order imbalance bounces back after one day. Put differently, the selling
pressure induced by the shock to ∆Libor-OIS subsides after a day. The order imbalance
bounce-back and the price reversal seen with changes to the Libor-OIS spread is consistent
with the view that this variable captures temporary fluctuations in liquidity conditions in
the interbank market that spill over into the equity markets.
The lower two graphs in Figure 3 show that a positive shock to ∆VIX has a strong and
significant negative contemporaneous effect on market returns and ∆OIBM . The negative
relation between returns and changes in the VIX is well known in the literature (see,
e.g., Ang, Hodrick, Xing, and Zhang, 2006). That increases in the VIX are associated
with selling pressure and negative returns is consistent with portfolio rebalancing by some
investors in the light of a perception of increased uncertainty in the market. The positive
values of ∆OIBM for periods 1 and 2 show that the level of order imbalance bounces back
within two days. Unlike the case of a shock to ∆Libor-OIS, there is no intertemporal effect
on market returns from a shock to ∆VIX. The impact is instantaneous, consistent with
statistically significant partial price reversal after 3 days.

29
an efficient price response to new information about market-wide uncertainty.33

4.2.2 Liquidity groups

This subsection repeats the above analysis for the ten liquidity groups. In the interest of
paucity, we do not report the full set of results. Our focus is on those results that relate
most directly to the liquidity pull-back hypothesis.
While space concerns prohibit us from showing the ten correlation matrices of the VAR
innovations of the individual group systems, we have found that as for the market, they
point to the existence of contemporaneous effects. For each group, there are always some
correlations that are statistically significant at the 1% level. For example, for Groups 1-4,
8, and 9, the correlations between innovations in ∆Libor-OIS and Return G are statistically
significantly negative at the 1% level, as for the market as a whole.
With respect to lagged effects, Table 12 presents the results of Granger-causality tests
of group returns (Panel A) and order imbalance changes (Panel B). Our focus is on the
effects of ∆Libor-OIS and ∆VIX. Panel A reports that ∆Libor-OIS Granger-causes the
returns of Groups 1 and 10 (at the 10% level). ∆VIX Granger-causes the returns of
Groups 9 and 10. With respect to order imbalance, Panel B reports that ∆Libor-OIS does
not Granger cause ∆OIB at conventional levels, but the p-value of the Wald test is 0.11
for Groups 1, 2, and 5. For ∆VIX, the lowest p-values are reached for the most illiquid
groups, 8-10, where p-values are 0.03 or lower. The pattern that emerges in Table 12 is thus
that ∆Libor-OIS has a relatively stronger lagged effect on more liquid stocks, while ∆VIX
has a relatively stronger lagged effect on less liquid stocks.34 That changes in the price
33
As a robustness check, we have re-run all the analysis in this sub-subsection with all variables win-
sorized at the 1st and 99th percentiles. The qualitative results are unaffected.
34
We have also run the VARs with Return 1 as an endogenous variable, because there is evidence
in the literature that large stocks lead small (Lo and MacKinlay, 1990). Chordia, Sarkar, and Subrah-
manyam (2011) provide evidence that the returns of the largest decile Granger causes that of the smallest
decile. Including Return 1 in our VARs, we get a similar result to that of Chordia et al and also get less
support for ∆VIX Granger causing the least liquid stocks. With respect to Return 1 versus ∆VIX, caution
needs to be exercised with respect to interpretation because Return 1 and ∆VIX are highly correlated.
However, our main interest lies in examining the impact of ∆Libor-OIS, as this speaks to the liquidity

30
of liquidity are especially important for the more liquid stocks is suggestive of liquidity
pull-back mattering more for those stocks than the less liquid ones.
With respect to the IRFs, we constrain our discussion to the response in group returns
and ∆OIBG from a one standard deviation shock to ∆Libor-OIS, as this is our main
focus of interest. The IRFs are computed in the same way as for the market as a whole
with the following ordering of the endogenous variables: ∆Libor-OIS, ∆VIX, ∆OIBG ,
∆VolumeG , ∆Bidask G , Return G , where G refers to liquidity group 1-10 or the market
(M). Table 13 lays out these IRFs in tabular form showing the impulse response levels at
different steps and the accompanying t-statistics. Period 20 is included to illustrate that
the IRFs converge to zero. The table also provides the IRFs for the market and upper and
lower confidence bands at the 10% level.
For all groups, Table 13, Panel A shows that shocking ∆Libor-OIS has a statistically
significant (1% level for Groups 1-9) contemporaneous negative effect on group returns.
This is consistent with liquidity pull-back. Furthermore, for the seven most liquid groups,
except Group 3, there is a statistically significant (10% or better) partial price reversal
after 3 days. The IRF value at step 3 for Group 3 is also positive, with a p-value of 0.12.
The negative contemporaneous effect and the subsequent price reversal parallels what we
saw for the market as a whole in Figure 3. Given that Groups 1-2 and 4-7 represent close
to 94% of volume on an average day, this is strong evidence in support of the liquidity
pull-back hypothesis.
Table 13, Panel A, also allows us to examine whether the magnitude of the contempora-
neous reactions to ∆Libor-OIS are similar across liquidity groups, as we would expect if all
stocks are equally subject to liquidity pull-back. The contemporaneous impulse responses
are indeed similar for the most liquid groups, the range being −0.122% to −0.080% for
the six most liquid groups. They decline (in absolute value) as liquidity declines, reaching
−0.032% for Group 10. Looking at the confidence bands reported in the table, we see
that the point estimate of the contemporaneous IRF value for the market of −0.122% falls
within the 10% confidence bands of each of the six most liquid groups. Conversely, the
pull-back hypothesis. We find that including Return 1 in the VAR systems does not alter the impact of
∆Libor-OIS in any significant way. Details are available from the authors upon request.

31
point estimates of the contemporaneous IRF values of the six most liquid groups all fall
within the 10% confidence bands of the market as well as those of each other. Thus the
magnitudes of the contemporaneous effects from shocks to ∆Libor-OIS are not statistically
different for the six most liquid groups. The results are weaker for the four least liquid
groups. These results indicate that there is an equalizing effect from liquidity pull-back
for the six most liquid groups. That liquidity pull-back appears to be especially relevant
for the more liquid groups is in line with our results in Section 3.35
With respect to price reversals, the magnitudes of the reactions at step 3 are similar
for the seven most liquid groups. As seen in Table 13, Panel A, the value of the IRF in
period 3 for the market is 0.049%, with values for Groups 1 to 7 falling between 0.042%
(Group 7) to 0.060% (Group 6). Furthermore, for these groups, the point estimates of
the individual IRFs at step 3 are within the confidence bands of the IRFs of the other six
groups as well as the market. This supports our other results that liquidity pull-back is
especially relevant for the most liquid groups.36
35
As a robustness check to our finding that the response to returns to a shock in ∆Libor-OIS is similar
across the most liquid groups, we have run VAR systems with return differences as one of the endogenous
variables. In particular, for each pair of groups, g and h (g > h), we define Return gh = Return g −Return h .
We define ∆OIBgh , ∆Volumegh , and ∆Bidask gh analogously. We then run VAR systems with ∆Libor-OIS,
∆VIX, Return gh , ∆OIBgh , ∆Volumegh , and ∆Bidask gh as endogenous variables. We are interested in the
impulse responses of Return gh to ∆Libor-OIS and in particular whether these are 0. Focusing especially
on steps 0 and 3, we find that we cannot reject the null hypothesis that the IRF has a value of zero when
g and h are in the set {1, 3, 4}. For Group 2, we have statistical significance at step 0 in the system
with Return 32 and Return 42 . For Groups g = 5 and higher, we have more spotty results and always
significance at step 0 for Return g1 . These results, the details of which are available upon request, support
the conclusion that there is an equalizing effect from liquidity pull-back for the most liquid groups.
36
We have also looked at the VAR coefficients. In the return regressions for the individual groups, only
the third lag of ∆Libor-OIS is statistically significant in any regression (as for the market as a whole).
Using Newey-West standard errors (6 lags), the coefficients on ∆Libor-OISt−3 are statistically significant
at 10% or better for Groups 1, 5, 6, 7, and 10 and with p-levels of 0.109, 0.114, and 0.101 for Groups 2, 3,
and 4, respectively. (For Groups 2, 3, and 4, the z test for the coefficient on ∆Libor-OISt−3 , which does
not take account of the Newey-West correction, have p-values of 0.065, 0.067, 0.072, respectively.) For
the first seven groups, the coefficients have similar magnitudes, ranging from 0.040 (Group 7) to 0.057
(Group 5). These results are consistent with a price reversal (after an initial negative reaction) as seen in

32
 The IRFs of ∆OIB are laid out in Table 13, Panel B. The contemporaneous effect from
a positive shock to ∆Libor-OIS is negative for all groups and statistically significant at
the 10% level or better for Groups 1, 3, 4, 6, and 10. The p-value for Group 2 is 0.12.
There is a statistically significant bounce-back after 1 day for Groups 1-4 and 6. In other
words, as for the market as a whole, increases in the price of liquidity are associated with
increased selling pressure, especially for stocks in the more liquid groups.
The results in this section can be summarized as follows: The relations between returns
and changes in order imbalance, on the one hand, and changes in the Libor-OIS spread,
on the other, are consistent with the liquidity pull-back hypothesis. First, the market as
a whole and all groups react negatively to changes in the Libor-OIS spread. For the six
most liquid groups, representing more than 99% of total volume on an average day, the
reaction is of similar magnitude and statistically not different from each other. This is
consistent with the impact of liquidity pull-back being equalized across stocks of different
liquidity levels. Second, there is a partial price reversal, as evidenced by the positive
period 3 impulse from shocking ∆Libor-OIS, especially for the market as a whole and the
seven most liquid groups. Third, the IRFs of changes in order imbalance to ∆Libor-OIS
show that for the market and four of the six most liquid groups, an increase in the price
of liquidity is associated with contemporaneous selling pressure. In short, our findings
are consistent with liquidity pull-back being a part of day-to-day activity in the equity
markets, especially for stocks in the 4-6 most liquid groups. With respect to portfolio
rebalancing in response to changes in market-wide uncertainty, we have found that an
increase in the VIX is associated with selling pressure and an instantaneous negative stock
price reaction.

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
the IRFs in Table 13, Panel A.

33
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 implications that
relate to volume, returns, and order imbalance, and the body of the paper is devoted to
testing these. The implications are verified in the data.
First, consistent with the liquidity pull-back hypothesis, we find that the market share
of volume of relatively more liquid stocks is increasing in 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, without this changing the qualitative findings. Interestingly, we find strong
evidence for liquidity pull-back 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. For the pre-crisis period, the results are especially strong for the four decile
portfolios consisting of the 40% most liquid stocks, representing approximately 97% of
daily volume on average.
By using the VIX as a control variable, we have also examined the alternative hy-
pothesis 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
pull-back, however, the evidence for portfolio rebalancing is also strong post-TAF.
Second, consistent with the liquidity pull-back hypothesis, we also find that increases
in the Libor-OIS spread are associated with selling pressure in the market as a whole and
especially for more liquid stocks. Third, increases in the Libor-OIS spread have a transitory
negative impact on returns. Fourth, the magnitudes of the return effect are similar for
the portfolios consisting of the 60% most liquid stocks, representing approximately 99%
of daily volume on average.
Increases in the VIX are also associated with selling pressure and negative returns. The
effect is differentiated from that of the Libor-OIS spread, however, by the impact of the
VIX being instantaneous. Overall, our findings on volume, order imbalance, and returns
provide strong supportive evidence for the liquidity pull-back hypothesis.

34
 Our results are especially strong for the 40% most liquid stocks. The results are weaker
for the 40% least liquid stocks. These stocks, which constitute less than 0.6% of total
volume on a given day, may simply be too small and illiquid to be a significant systematic
part of liquidity pull-back trading. Amihud and Mendelson (1986) and Longstaff (2009)
show that portfolios may be optimally tilted away from illiquid financial assets by agents
that have more immediate liquidity needs.
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 high-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 high powered money (liquidity). When the price of liquidity goes up,
the price of conversion also rises and asset prices therefore fall. Thus asset prices fluctuate
with the price of liquidity, which is consistent with our findings. On a more macro level,
the implication is that if there is not sufficient liquidity in the system to satisfy banks’
demand in aggregate, asset prices fall as banks start to engage in liquidity pull-back. In
such a state, if additional liquidity is injected by the central bank, the value of higher
powered money should fall and this would show up as an increase in the prices of financial
assets. The fall in asset prices during the crisis and their subsequent rise following the
Fed’s first quantitative easing program in March 2009 could be viewed as an example of
this phenomenon.
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 other 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

35
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
are in the interbank market, the larger are asset cross-correlations and the stronger will
flight to quality appear to be.

36
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, 65-69.

Anderson, Anne M., and Edward A. Dyl, 2007, Trading Volume: NASDAQ and the NYSE,
Financial Analysts Journal 63, 115-131.

Ang, Andrew, Robert J. Hodrick, Yuhang Xing, and Xiaoyan Zhang, 2006, The cross-
section of volatility and expected returns, Journal of Finance 61, 259-299.

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.

Brown, David B., Bruce Ian Carlin, and Miguel Sousa Lobo, 2010, Optimal portfolio
liquidation with distress risk, Management Science 56, 1997-2014.

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.

37
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
trading activity, Journal of Finance 56, 501-530.

Chordia, Tarun, Richard Roll, and Avanidhar Subrahmanyam, 2002, Order imbalance,
liquidity, and market returns, Journal of Financial Economics 65, 111-130.

Chordia, Tarun, Richard Roll, and Avanidhar Subrahmanyam, 2008, Liquidity and market
efficiency, Journal of Financial Economics 87, 249-268.

Chordia, Tarun, Asani Sarkar, and Avanidhar Subrahmanyam, 2005, An empirical analysis
of stocks and bond market liquidity, Review of Financial Studies 18, 85-130.

Chordia, Tarun, Asani Sarkar, and Avanidhar Subrahmanyam, 2011, Liquidity dynamics
and cross-autocorrelations, Journal of Financial and Quantitative Analysis 46, 709-736.

Chordia, Tarun and Avanidhar Subrahmanyam, 2004, Order imbalance and individual
stock returns: Theory and evidence, Journal of Financial Economics 72, 485-518.

Christie, William G. and Paul H. Schultz, 1999, The initiation and withdrawal of odd-
eighth quotes among Nasdaq stocks: An empirical analysis, Journal of Financial Eco-
nomics 52, 409-442.

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.

Fama, Eugene F., and James D. MacBeth, 1973, Risk, return, and equilibrium: Empirical
tests, Journal of Political Economy 81, 607-636.

Fair, Ray, C., 2002, Events that shock the market, Journal of Business 75, 713-731.

Fecht, Falko, Kjell G. Nyborg, and Jörg Rocholl, 2011, The price of liquidity: the effects of
market conditions and bank characteristics, Journal of Financial Economics 102, 344-362.

Ferson, Wayne E. and Campbell R. Harvey, 1993, The risk and predictability of interna-
tional 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

38
78, 193-238.

Gallant, A. Ronald, Rossi, Peter E., and George Tauchen, 1992, Stock prices and volume,
Review of Financial Studies 5, 199-242.

Greene, William H., 2012, Econometric analysis, Prentice Hall 7th edition.

Goetzmann, William, N., and Massimo Massa, 2002, Daily momentum and contrarian
behavior of index fund investors, Journal of Financial and Quantitative Analysis 37, 375-
389.

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, 2012, Securitized banking and the run on repo,
Journal of Financial Economics 104, 425-451.

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
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.

Lee, Charles, M., C., and Mark J. Ready, 1991, Inferring trade direction from intraday
data, Journal of Finance 46, 733-746.

Lo, Andrew, W., and Jiang Wang, 2000, Trading volume: Definitions, data analysis, and
implications of portfolio theory, Review of Financial Studies 13, 257-300.

39
Lo, A. W. and A. C. MacKinlay, 1990, When are contrarion profits due to stock market
overreaction? Review of Financial Studies 3, 175-205.

Longstaff, Francis A., 2009, Portfolio Claustrophobia: Asset pricing in markets with illiq-
uid assets, American Economic Review 99, 1119-1144.

Lütkepohl, Helmut, 2007, New Introduction to Multiple Time Series Analysis, Springer-
Verlag, Berlin, Heidelberg, New York.

Newey, Whitney K., and Kenneth D. West, 1987, A simple, positive semi-definite, het-
eroskedasticity and autocorrelation consistent covariance matrix, Econometrica 55, 703–
708.

O’Hara, Maureen, Chen Yao, and Mao Ye, 2012, What’s not there: The Odd-Lot bias in
TAQ data, Johnson School Working Paper, Cornell University.

Pastor, Lubos and Robert Stambaugh, 2003, Liquidity risk and expected stock returns,
Journal of Political Economy 111, 642-685.

Piazzesi, 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.

Scholes, Myron S., 2000, Crisis and risk management, American Economic Review 90,
17-21.

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, 676-
685.

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.

40
Basis Points Figure 1: Ted and Libor‐OIS spreads
600

500
TED

Libor‐OIS
400

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 Asian crisis

50

0
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.
 .001 .004

.002

−.001

−.002

−.002 −.004
0 5 10 15 20 0 5 10 15 20

a. Response of market return to ∆Libor-OIS b. Response of ∆OIBM to ∆Libor-OIS

.01

−.002
0

−.004

−.01

−.006

−.008 −.02
0 5 10 15 20 0 5 10 15 20

c. Response of market return to ∆VIX d. Response of ∆OIBM to ∆VIX

Fig. 3. Impulse response functions (IRFs). The VAR is run using daily data over the
pre-crisis period (December 4, 2001 to June 29, 2007). The number of lags is determined
by the Hannan-Quinn criterion and equals four. The IRFs are computed by orthogonal-
izing the covariance matrix of the VAR innovations using the Cholesky decomposition
with the following ordering of the endogenous variables: ∆Libor-OIS, ∆VIX, ∆OIBM ,
∆VolumeM , ∆Bidask M , Return M . Graphs a. and b. show the responses to Return M and
∆OIBM , respectively, from shocking the innovation in ∆Libor-OIS by one standard devi-
ation. Graphs c. and d. do the same by shocking the innovation in ∆VIX by one standard
deviation. To understand the units on the y-axes, note for example that 0.001 represents
0.1%.
 Table 1
Descriptive statistics of the Libor-OIS and TED spreads
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 whole sample periods and, for each spread, two subsamples: (i) Non-crisis and (ii) Crisis, which are
defined as in the notes to Figure 2. bps: basis points.

Spread Units Mean Std Error Std Dev Median Min Max N
All Days Libor-OIS bps 25.80 1.01 41.74 11.75 2.51 366.33 1,716
TED bps 55.80 0.58 43.85 41.77 2.63 456.87 5,648
Non-Crisis Libor-OIS bps 11.06 0.10 3.59 10.52 2.51 28.40 1,402
TED bps 47.95 0.45 31.82 38.44 2.63 258.00 5,013
Crisis Libor-OIS bps 91.62 3.64 64.58 72.83 30.17 366.32 314
TED bps 117.70 2.75 69.25 99.50 26.00 456.87 635
1
 Table 2
Liquidity group transition matrix
Each month, stocks are sorted into liquidity deciles (1 most liquid and 10 most 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. The sample period is December 2001 to November 2008.

month t + 1
Group 1 2 3 4 5 6 7 8 9 10
1 90.90 8.97 0.09 0.01 0.02 0.00 0.00 0.01 0.01 0.00
2 9.07 75.58 15.07 0.23 0.03 0.01 0.00 0.00 0.00 0.00
3 0.28 14.80 66.96 17.37 0.52 0.03 0.03 0.00 0.00 0.00
4 0.11 0.50 16.46 62.97 19.17 0.70 0.07 0.02 0.00 0.00
month t 5 0.03 0.16 0.83 17.69 60.28 20.05 0.87 0.09 0.00 0.00
6 0.01 0.07 0.18 1.05 17.81 59.24 20.59 0.95 0.10 0.01
7 0.00 0.01 0.07 0.20 1.41 17.54 58.53 20.96 1.19 0.09
8 0.00 0.00 0.03 0.07 0.20 1.62 17.60 57.99 21.34 1.16
9 0.00 0.00 0.01 0.03 0.09 0.29 1.77 18.11 59.90 19.79
10 0.01 0.00 0.00 0.01 0.02 0.08 0.28 1.77 17.87 79.96
2
 Table 3
Descriptive statistics of ILLIQ and Market share of volume
Panel A reports descriptive statistics for Amihud’s (2002) ILLIQ. For each stock, ILLIQ is calculated for each month according to equation (1), with volume
measured in millions of dollars. Each month stocks are sorted into ten decile groups based on their ILLIQ, Group 1 is the decile of the most liquid stocks, Group 10 is
the decile of the most illiquid stocks. For each group, we pool the firm-month ILLIQ values across months and report descriptive statistics for the resulting group level
samples. All Groups is the pooled sample of firm-month ILLIQ values across all groups. The sample period is December 2001 to November 2008. Panel B reports
descriptive statistics for the daily Market share of volume of each liquidity group. The Market share of volume of Group G on a given day is defined as the total dollar
volume of all stocks in Group G on that day divided by the total dollar volume of all stocks in the sample on that same day. The sample period is December 4, 2001
to November 11, 2008.

Liquidity Group
All Groups 1 2 3 4 5 6 7 8 9 10
Panel A: ILLIQ
Mean 4.807 0.000 0.001 0.003 0.007 0.019 0.056 0.174 0.578 2.357 44.911
Std. Error 0.118 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.005 0.020 1.161
Std. Dev. 66.789 0.000 0.001 0.002 0.008 0.025 0.080 0.257 0.859 3.532 206.986
Median 0.016 0.000 0.001 0.001 0.003 0.008 0.020 0.054 0.179 0.726 7.437
Min 0.000 0.000 0.000 0.000 0.001 0.002 0.004 0.010 0.032 0.100 0.468
Max 16,116.883 0.001 0.005 0.016 0.051 0.173 0.546 1.735 5.882 24.783 16,116.883
N (firm-months) 318,315 31,796 31,836 31,840 31,839 31,824 31,853 31,847 31,832 31,844 31,804
3

Panel B: Market share of volume (in %)

Mean n.a. 75.171 13.250 5.893 2.930 1.467 0.722 0.326 0.137 0.069 0.035
Std. Error n.a. 0.101 0.038 0.026 0.019 0.013 0.008 0.004 0.002 0.002 0.002
Std. Dev. n.a. 4.194 1.567 1.085 0.804 0.533 0.312 0.169 0.080 0.088 0.087
Median n.a. 74.103 13.409 6.104 3.120 1.579 0.779 0.337 0.133 0.053 0.017
Min n.a. 56.598 8.490 2.991 0.949 0.336 0.132 0.049 0.021 0.008 0.003
Max n.a. 86.692 18.001 10.794 5.466 3.944 3.264 1.572 0.815 1.478 1.545
N (daily obs.) n.a. 1,716 1,716 1,716 1,716 1,716 1,716 1,716 1,716 1,716 1,716
 Table 4
Regressions of Market share of volume on spreads
Each column represents a separate regression, one for each liquidity group, G:
Yt = α + β1 spreadt + β2 Yt−1 + εt ,
where Yt is Market share of volume Gt /mean(Market share of volume G) and the mean is taken over the sample period. 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 as an explanatory variable. 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. In Panels A (B) standard errors are adjusted using the Newey-West (1987) procedure
with 6 (9) lags. t-statistics are in brackets. a, b, and c denote statistical significance (two-tailed) at the 1%, 5%, and 10% level, respectively. A t − 1
subscript indicates a lagged variable. Both spreads are in percentage points.

Liquidity Group
1 2 3 4 5 6 7 8 9 10
Panel A
Constant 0.092a 0.178a 0.146a 0.113a 0.113a 0.176a 0.217a 0.207a 0.300a 0.342a
(7.55) (10.70) (8.20) (7.54) (6.45) (4.54) (5.62) (7.00) (6.30) (4.28)
Libor-OIS 0.003a -0.004 -0.014a -0.019a -0.023a -0.044a -0.071a -0.081 -0.125a
a
-0.201a
(3.97) (-1.64) (-4.69) (-4.45) (-4.08) (-3.77) (-4.25) (-4.55) (-4.12) (-3.32)
Yt−1 0.907a 0.823a 0.858a 0.893a 0.893a 0.836a 0.801a 0.814a 0.732a 0.710a
(73.07) (51.80) (51.92) (67.30) (57.67) (24.07) (24.26) (33.53) (17.00) (10.28)
N 1,715 1,715 1,715 1,715 1,715 1,715 1,715 1,715 1,715 1,715
4

Adj. R2 0.8310 0.6785 0.7475 0.8110 0.8105 0.7179 0.6713 0.6949 0.5466 0.5105
Panel B
Constant 0.089a 0.155a 0.123a 0.104a 0.127a 0.247a 0.211a 0.212a 0.267a 0.372a
(16.07) (16.11) (16.17) (15.63) (12.93) (4.45) (9.51) (8.34) (9.15) (5.03)
TED 0.003a -0.009a -0.017a -0.019a -0.024a -0.049a -0.045a -0.041a -0.055a -0.100a
(4.38) (-3.16) (-4.73) (-5.07) (-5.10) (-3.70) (-5.75) (-4.78) (-4.02) (-3.51)
Yt−1 0.909a 0.850a 0.887a 0.906a 0.886a 0.780a 0.814a 0.811a 0.764a 0.684a
(160.31) (96.98) (136.16) (158.81) (111.20) (16.07) (44.56) (37.81) (31.95) (10.25)
N 5,647 5,647 5,647 5,647 5,647 5,647 5,647 5,647 5,647 5,647
Adj. R2 0.8357 0.7283 0.8003 0.8320 0.7972 0.6260 0.6769 0.6670 0.5891 0.4718
 Table 5
Descriptive statistics: Libor-OIS, VIX, and control variables
Panel A is for the pre-crisis period, December 4, 2001 to June 29, 2007. Panel B is for the post-TAF period, December 17, 2007 to November
11, 2008. Daily data. Market return is the CRSP value weighted market return. TAF: term auction facility. bps: basis points.

Units Mean Std Error Std Dev Median Min Max N

Panel A: Pre-crisis period
Libor-OIS bps 11.09 0.10 3.60 10.60 2.51 28.40 1,375
VIX % 17.73 0.19 6.92 15.50 9.89 45.08 1,375
Group 1 Relative Bid Ask % 13.20 0.17 6.48 10.94 4.92 42.44 1,375
Group 2 Relative Bid Ask % 17.95 0.19 7.14 15.43 7.60 48.45 1,375
Group 3 Relative Bid Ask % 22.60 0.19 6.90 20.18 11.73 49.16 1,375
Group 4 Relative Bid Ask % 29.26 0.20 7.31 26.78 15.96 51.53 1,375
Group 5 Relative Bid Ask % 38.98 0.24 8.81 36.83 21.48 62.86 1,375
Group 6 Relative Bid Ask % 56.98 0.31 11.66 54.70 32.97 88.72 1,375
Group 7 Relative Bid Ask % 89.01 0.38 14.03 84.93 64.93 128.71 1,375
Group 8 Relative Bid Ask % 142.09 0.35 12.94 140.14 110.51 189.79 1,375
Group 9 Relative Bid Ask % 227.48 0.37 13.72 226.81 184.25 296.53 1,375
Group 10 Relative Bid Ask % 416.95 1.02 37.77 424.37 304.90 513.92 1,375
Market return % 0.04 0.03 0.96 0.08 -3.82 5.32 1,375
Normalized market dollar volume % 101.00 0.43 15.96 99.79 27.87 226.78 1,375
5

Correlation between VIX and Libor-OIS: 0.61

Panel B: Post-TAF period
Libor-OIS bps 100.48 4.91 73.68 73.93 30.17 366.32 225
VIX % 28.67 0.90 13.50 23.82 16.30 80.06 225
Group 1 Relative Bid Ask % 9.48 0.20 3.02 9.13 4.60 31.68 225
Group 2 Relative Bid Ask % 12.67 0.21 3.10 12.40 7.46 36.89 225
Group 3 Relative Bid Ask % 14.50 0.21 3.11 14.31 9.29 41.76 225
Group 4 Relative Bid Ask % 17.88 0.24 3.53 17.65 12.73 54.16 225
Group 5 Relative Bid Ask % 23.90 0.31 4.72 23.65 15.89 70.58 225
Group 6 Relative Bid Ask % 36.62 0.40 6.06 36.23 25.53 101.57 225
Group 7 Relative Bid Ask % 70.05 0.65 9.68 68.50 55.18 152.82 225
Group 8 Relative Bid Ask % 137.61 1.03 15.40 135.43 112.87 185.31 225
Group 9 Relative Bid Ask % 245.48 1.30 19.57 241.42 198.94 305.13 225
Group 10 Relative Bid Ask % 507.55 2.74 41.10 509.10 256.94 591.56 225
Market return % -0.19 0.15 2.25 -0.05 -9.00 11.52 225
Normalized market dollar volume % 100.70 1.17 17.58 98.85 34.31 179.91 225
Correlation between VIX and Libor-OIS: 0.89
 Table 6
Regressions of Market share of volume on the Libor-OIS spread, VIX, and control variables
Each column represents a separate regression, one for each liquidity group, G:
Yt = α + β1 spreadt + β2 ResidualZ|X,t + β3 Yt−1 + Γ0 Wt + ηt ,
where Yt is Market share of volume Gt /mean(Market share of volume G) and the mean is taken over the sample period. The Market share of volume is
divided by its time series average to facilitate comparisons among groups. Daily data (t denotes an individual day). ResidualZ|X,t is the residual (εbt ) from
the following first step regression:
Zt = α + γXt + εt ,
where Z and X are either the VIX or the Libor-OIS spread. W is a vector of control variables [market returnt−1 , Group Relative bid-ask, and Normalized
market dollar volume] 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. In Panels A and B (C and D) standard errors are adjusted using the Newey-West (1987) procedure with 6 (4) lags.
t-statistics are in brackets. a, b, and c denote statistical significance (two-tailed) at the 1%, 5%, and 10% level, respectively. Market return is the CRSP
value weighted market return. A t − 1 subscript indicates a lagged variable. The Libor-OIS spread is in percentage points and the VIX is in decimals.

Liquidity Group
1 2 3 4 5 6 7 8 9 10
Panel A: Pre-crisis period. X is the Libor-OIS spread.
Constant 0.288a 0.489a 0.768a 0.791a 0.785a 1.044a 1.128a 1.130a 1.689a 1.526a
(9.46) (13.57) (14.88) (13.25) (9.67) (9.36) (10.01) (7.46) (3.38) (3.02)
Libor-OIS 0.292a -0.524a -1.191a -1.260a -1.211a -1.886a -1.555a -1.380a -1.166a -2.428a
(10.26) (-9.13) (-11.33) (-8.82) (-5.78) (-4.93) (-3.98) (-4.86) (-2.64) (-3.02)
6

ResidualVIX|Libor-OIS 0.183a -0.285a -0.723a -0.913a -1.000a -1.418a -1.451a -1.354a -1.942a -1.941b
(9.60) (-7.12) (-10.59) (-10.22) (-8.12) (-6.09) (-5.52) (-6.17) (-6.17) (-2.58)
Yt−1 0.647a 0.638a 0.521a 0.602a 0.653a 0.564a 0.612a 0.696a 0.708a 0.700a
(20.61) (24.59) (15.45) (18.19) (16.31) (8.63) (13.33) (23.62) (15.39) (9.75)
Market returnt−1 -0.110c 0.203 0.420c 0.292 0.961b 0.725 0.936 1.188c 1.434 5.179b
(-1.71) (1.24) (1.67) (0.93) (2.19) (1.02) (1.18) (1.65) (1.04) (2.12)
Group Relative Bid-Ask 0.101a -0.189a -0.417a -0.555a -0.569a -0.543a -0.374a -0.222b -0.237c 0.009
(8.16) (-6.97) (-7.74) (-7.14) (-5.52) (-4.24) (-2.76) (-2.46) (-1.70) (0.08)
Norm. market dollar volume 0.019a -0.035b -0.062a -0.090a -0.081c -0.088 -0.233b -0.354a -0.721a -0.985a
(2.85) (-2.20) (-2.79) (-3.22) (-1.77) (-1.06) (-2.17) (-3.72) (-2.69) (-3.70)
N 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374
Adj. R2 0.8717 0.7352 0.8039 0.8546 0.8436 0.7694 0.7086 0.7046 0.5558 0.5090
Panel B: Pre-crisis period. X is the VIX.
Constant 0.284a 0.492a 0.783a 0.824a 0.830a 1.098a 1.205a 1.205a 1.840a 1.609a
(9.47) (13.30) (14.82) (13.47) (9.98) (9.60) (10.00) (8.14) (3.53) (3.05)
VIX 0.207a -0.345a -0.832a -0.972a -1.011a -1.488a -1.403a -1.286a -1.586a -1.988a
(10.61) (-9.89) (-11.96) (-10.31) (-7.52) (-5.81) (-5.32) (-6.20) (-6.45) (-3.08)
ResidualLibor-OIS|VIX 0.077a -0.188a -0.340a -0.186c -0.035 -0.218 0.151 0.213 1.118b -0.145
(3.96) (-2.98) (-4.28) (-1.91) (-0.25) (-1.01) (0.51) (0.93) (1.97) (-0.18)
Control variables, their coefficients and t-statistics, N , and Adj. R2 are by construction identical to those in Panel A.
 Table 6 – continued

Liquidity Group
1 2 3 4 5 6 7 8 9 10
Panel C: Post-TAF period. X is the Libor-OIS spread.
Constant 0.458a 0.563a 0.518a 0.530a 0.446a 0.306 0.687a 0.707a 1.607b 2.919a
(5.88) (7.63) (7.16) (5.12) (3.14) (1.56) (3.61) (2.73) (2.24) (4.21)
Libor-OIS 0.011a -0.025a -0.039a -0.060a -0.053a -0.108a -0.168a -0.132a -0.033 -0.254a
(4.83) (-3.92) (-5.09) (-4.91) (-2.94) (-2.78) (-3.22) (-3.67) (-0.86) (-4.95)
ResidualVIX|Libor-OIS 0.092a -0.256a -0.285a -0.380a -0.434a -0.605b -1.008a -0.543b -0.229 -1.325a
(4.65) (-3.91) (-4.16) (-4.18) (-3.23) (-2.55) (-2.87) (-2.54) (-0.71) (-2.94)
Yt−1 0.517a 0.505a 0.566a 0.536a 0.613a 0.468b 0.350c 0.572a 0.684a 0.427a
(6.85) (7.46) (9.30) (7.16) (4.76) (2.32) (1.70) (6.81) (8.98) (3.91)
Market returnt−1 0.042 0.036 -0.292 -0.360 -0.653 -0.602 -1.229 -0.828 -0.238 -1.339c
(0.87) (0.24) (-1.56) (-1.25) (-1.60) (-1.35) (-1.54) (-1.07) (-0.30) (-1.67)
Group Relative Bid-Ask 0.041 -0.126 -0.033 0.019 0.262 0.738 0.297 0.134 -0.251 -0.318a
(0.79) (-1.21) (-0.23) (0.07) (0.80) (1.44) (1.23) (0.73) (-1.19) (-2.89)
Norm. market dollar volume 0.009 -0.027 -0.040 -0.010 -0.071 0.061 -0.080 -0.332c -0.641b -0.481b
(0.81) (-1.19) (-0.89) (-0.15) (-0.63) (0.40) (-0.28) (-1.67) (-2.25) (-2.19)
N 224 224 224 224 224 224 224 224 224 224
7

Adj. R2 0.6344 0.5294 0.6035 0.5611 0.5488 0.4707 0.2569 0.4326 0.5471 0.3884
Panel D: Post-TAF period. X is the VIX.
Constant 0.448a 0.589a 0.550a 0.577a 0.494a 0.385b 0.814a 0.791a 1.634b 3.099a
(5.83) (7.62) (7.24) (5.31) (3.35) (1.97) (4.40) (3.13) (2.32) (4.25)
VIX 0.076a -0.179a -0.252a -0.373a -0.350a -0.654a -1.033a -0.758a -0.209 -1.517a
(5.51) (-4.53) (-5.70) (-5.47) (-3.54) (-2.89) (-3.30) (-3.84) (-0.96) (-4.67)
ResidualLibor-OIS|VIX -0.003 0.016 0.007 0.001 0.018 -0.010 -0.005 -0.045 0.004 -0.040
(-1.05) (1.62) (0.59) (0.09) (0.73) (-0.33) (-0.13) (-1.15) (0.08) (-0.77)
Control variables, their coefficients and t-statistics, N , and Adj. R2 are by construction identical to those in Panel C.
 Table 7
Regressions of Relative volume on the Libor-OIS spread, VIX, and control variables over
the pre-crisis period
This repeats the regressions from Table 6, Panels A and B, 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:
Yt = α + β1 Xt + β2 ResidualZ|X,t + β3 Yt−1 + Γ0 Wt + ηt ,
where Yt is Relative volume of Group G to Group Ht /mean(Relative volume of Group G to H) and the
mean is taken over the sample period. Daily data (t denotes an individual day). ResidualZ|X,t is the
residual (εbt ) from the following first step regression:
Zt = α + γXt + εt ,
where Z and X are either the VIX or the Libor-OIS spread. W is a vector of control variables [market
returnt−1 , Group Relative bid-ask, and Normalized market dollar volume] and Γ is a vector of coefficients.
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). Standard errors are adjusted using the Newey-
West (1987) procedure with 6 lags. a, b and c denote statistical significance (two-tailed) at the 1%, 5%
and 10% level, respectively. The Libor-OIS spread is in percentage points and the VIX is in decimals.

Liquidity Group G (less liquid)

Group H 2 3 4 5 6 7 8 9 10
Panel A: Pre-crisis period. X is the Libor-OIS spread.
1 -0.802a -1.477a -1.495a -1.485a -2.277a -1.838a -1.641a -1.308a -2.543a
(-10.09) (-10.89) (-8.70) (-5.90) (-5.05) (-4.14) (-4.94) (-2.82) (-3.08)
2 -0.700a -1.101a -0.971a -1.583a -1.370a -1.078a -0.893b -2.274a
(-7.92) (-9.00) (-5.18) (-5.05) (-3.59) (-4.07) (-2.06) (-2.86)
3 -0.597a -0.713a -1.178a -0.966b -0.742a -0.635c -2.231a
(-5.64) (-4.79) (-4.64) (-2.40) (-3.15) (-1.72) (-2.79)
4 -0.690a -1.105a -0.806b -0.527b -0.415 -1.718b
(-4.90) (-5.27) (-2.21) (-2.42) (-1.20) (-2.49)
5 -0.401b -0.103 0.023 -0.084 -1.603b
(-2.44) (-0.21) (0.12) (-0.31) (-2.24)
6 0.159 0.503b 0.281 -1.103c
(0.35) (2.32) (1.16) (-1.72)
7 0.371 0.273 -1.130
(1.61) (1.05) (-1.64)
8 -0.057 -1.121c
(-0.25) (-1.67)
9 -0.937
(-1.47)

8
Table 7 – continued

Liquidity Group G (less liquid)

Group H 2 3 4 5 6 7 8 9 10
Panel B: Pre-crisis period. X is the VIX.
1 -0.524a -1.005a -1.109a -1.189a -1.734a -1.573a -1.451a -1.671a -2.038a
(-10.78) (-11.53) (-9.94) (-7.34) (-5.84) (-5.25) (-5.95) (-6.44) (-3.07)
2 -0.536a -0.936a -0.884a -1.327a -1.362a -1.139a -1.475a -1.972a
(-10.91) (-11.79) (-7.54) (-6.34) (-5.02) (-5.91) (-6.31) (-3.07)
3 -0.574a -0.683a -1.030a -1.093a -0.907a -1.218a -1.930a
(-9.75) (-7.50) (-6.17) (-4.29) (-5.22) (-6.12) (-3.12)
4 -0.574a -0.896a -0.976a -0.738a -1.025a -1.547a
(-7.30) (-7.32) (-4.77) (-4.76) (-5.53) (-2.83)
5 -0.375a -0.504b -0.349a -0.722a -1.411a
(-4.48) (-2.25) (-2.63) (-4.73) (-2.74)
6 -0.337 -0.042 -0.389a -1.155b
(-1.50) (-0.29) (-2.99) (-2.49)
7 0.160 -0.205 -1.031b
(1.00) (-1.54) (-2.40)
8 -0.299b -0.881b
(-2.09) (-2.11)
9 -0.578
(-1.44)

9
 Table 8
NASDAQ stocks excluded: Regressions of Market share of volume on the Libor-OIS spread, VIX, and control variables over the pre-crisis period
This table runs the same regressions as in Table 6, Panels A and B, but with NASDAQ stocks excluded from the whole analysis. Each column
represents a separate regression, one for each liquidity group, G:
Yt = α + β1 spreadt + β2 ResidualZ|X,t + β3 Yt−1 + Γ0 Wt + ηt ,
where Yt is Market share of volume Gt /mean(Market share of volume G) and the mean is taken over the sample period. The Market share of volume
is divided by its time series average to facilitate comparisons among groups. Daily data (t denotes an individual day). ResidualZ|X,t is the residual (εbt )
from the following first step regression:
Zt = α + γXt + εt ,
where Z and X are either the VIX or the Libor-OIS spread. W is a vector of control variables [market returnt−1 , Group Relative bid-ask, and
Normalized market dollar volume] 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. Standard errors
are adjusted using the Newey-West (1987) procedure with 6 lags. t-statistics are in brackets. a, b, and c denote statistical significance (two-tailed) at
the 1%, 5%, and 10% level, respectively. Market return is the CRSP value weighted market return. A t − 1 subscript indicates a lagged variable. The
Libor-OIS spread is in percentage points and the VIX is in decimals.

Liquidity Group
1 2 3 4 5 6 7 8 9 10
Panel A: Pre-crisis period. X is the Libor-OIS spread.
Constant 0.250a 0.468a 0.578a 0.660a 0.712a 0.675a 0.802a 0.846a 1.083a 1.256a
(12.13) (13.27) (14.96) (12.83) (11.48) (8.45) (9.84) (8.34) (6.07) (5.02)
Libor-OIS 0.202a -0.532a -0.902a -1.372a -1.204a -1.467a -1.412a -2.012a -2.071a -1.962a
10

(10.68) (-8.70) (-8.59) (-8.03) (-6.94) (-6.12) (-4.72) (-6.65) (-4.90) (-3.77)
ResidualVIX|Libor-OIS 0.109a -0.224a -0.552a -0.694a -0.844a -1.158a -1.238a -1.798a -1.932a -2.309a
(9.29) (-5.92) (-8.98) (-8.04) (-7.98) (-8.12) (-6.62) (-9.08) (-7.98) (-5.41)
Yt−1 0.696a 0.684a 0.670a 0.680a 0.668a 0.713a 0.712a 0.694a 0.677a 0.650a
(30.67) (29.94) (29.39) (25.29) (26.63) (27.42) (20.72) (25.54) (16.29) (12.68)
Market returnt−1 -0.034 0.201 -0.021 0.107 1.109b -0.164 0.541 -0.279 0.358 0.570
(-0.69) (1.09) (-0.08) (0.30) (2.06) (-0.25) (0.65) (-0.36) (0.41) (0.41)
Group Relative Bid-Ask 0.065a -0.150a -0.176a -0.190a -0.261a -0.099b -0.109a 0.022 -0.021 0.019
(9.77) (-9.22) (-7.31) (-6.62) (-6.47) (-2.45) (-3.00) (0.59) (-0.77) (0.93)
Norm. market dollar volume 0.020a -0.053a -0.088a -0.105a -0.104b -0.149b -0.224a -0.355a -0.460a -0.789a
(4.62) (-3.25) (-4.23) (-4.17) (-2.53) (-2.50) (-3.08) (-4.47) (-4.62) (-4.23)
N 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374
Adj. R2 0.9021 0.8223 0.8468 0.8687 0.8058 0.7833 0.7527 0.7246 0.6806 0.5051
Panel B: Pre-crisis period. X is the VIX.
Constant 0.249a 0.464a 0.590a 0.663a 0.740a 0.724a 0.862a 0.936a 1.184a 1.406a
(12.20) (13.18) (15.02) (13.19) (11.76) (9.21) (10.40) (9.11) (6.73) (5.23)
VIX 0.132a -0.310a -0.632a -0.871a -0.911a -1.191a -1.224a -1.765a -1.868a -2.069a
(11.10) (-8.91) (-9.99) (-8.68) (-8.45) (-8.20) (-6.41) (-9.18) (-7.61) (-5.83)
ResidualLibor-OIS|VIX 0.074a -0.268a -0.253a -0.555a -0.212 -0.105 0.044 0.102 0.201 0.754
(4.79) (-4.16) (-2.85) (-4.56) (-1.47) (-0.51) (0.18) (0.38) (0.52) (1.25)
Control variables, their coefficients and t-statistics, N , and Adj. R2 are by construction identical to those in Panel A.
 Table 9
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
and report the average of all the cross-sectional coefficient estimates (α̂m , β̂m ) with corresponding t-statistics. 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. 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 of normalized share volume on high and low spread days. ILLIQG,m−1 is the equally weighted
average month m − 1 ILLIQ across stocks in liquidity group G. Specification (1) uses the Libor-OIS spread to identify high and
low spread days. Specification (2) uses the TED spread. 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). Libor-OIS and TED spread sample periods are, respectively, December 4, 2001
to November 11, 2008 and January 2, 1986 to December 31, 2008. Standard errors in specifications (2) and (5) are adjusted using
the Newey-West (1987) procedure with four lags while the remaining specifications use three lags. The a, b, and c denote statistical
11

significance (two-tailed) at the 1%, 5%, and 10% level respectively.

Spread (and Period) Constant t-stat ILLIQG t-stat Adj. R2 N

(1) Libor-OIS (complete sample period) 0.034 (1.21) -0.015c (-1.80) 0.1807 83
(2) TED (complete sample period) 0.023 (1.65) -0.004b (-2.28) 0.1428 275
(3) TED (Libor-OIS period) 0.019 (0.66) -0.013b (-2.21) 0.2030 83
(4) Libor-OIS (pre 07/2007) 0.026 (0.83) -0.017a (-2.79) 0.1649 66
(5) TED (pre 07/2007) 0.019 (1.50) -0.004b (-2.05) 0.1370 257
 Table 10
Descriptive statistics: Daily returns and change in order imbalance (∆OIB)
Panel A reports descriptive statistics of daily value weighted market and group returns. The market is the union of all groups. Panel B
reports descriptive statistics for daily ∆OIB × 100 [see equation (8)], meaning the variable is expressed as a percentage. For both panels, the
sample period is the Libor-OIS pre-crisis period (December 4, 2001 to June 29, 2007).

Liquidity Group
Market 1 2 3 4 5 6 7 8 9 10
Panel A: Returns (daily, %)
Mean 0.037 0.031 0.057 0.060 0.059 0.067 0.064 0.073 0.068 0.064 0.079
Std. Error 0.027 0.027 0.029 0.031 0.033 0.033 0.031 0.026 0.019 0.016 0.017
Std. Dev. 0.994 0.983 1.092 1.154 1.227 1.219 1.166 0.974 0.720 0.597 0.629
Median 0.075 0.057 0.116 0.108 0.093 0.127 0.118 0.124 0.112 0.097 0.062
Min -3.866 -4.072 -3.575 -4.132 -4.510 -5.004 -4.503 -4.187 -3.361 -3.009 -2.436
Max 5.483 5.844 5.913 5.464 4.927 4.306 3.802 3.388 2.576 2.196 3.627
N 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374
Panel B: Change in order imbalance (∆OIB, daily, %)
Mean -0.005 -0.006 -0.006 -0.003 -0.002 0.000 0.001 0.003 0.017 0.029 0.005
Std. Error 0.071 0.074 0.070 0.090 0.110 0.135 0.158 0.173 0.199 0.269 0.314
Std. Dev. 2.630 2.738 2.610 3.323 4.084 4.991 5.873 6.410 7.364 9.966 11.657
12

Median -0.077 -0.002 -0.091 -0.096 -0.148 -0.167 -0.008 -0.047 0.004 -0.153 0.022
Min -11.746 -13.333 -10.383 -12.155 -17.450 -20.912 -25.503 -40.662 -33.737 -51.985 -55.837
Max 10.506 15.357 10.409 11.766 22.587 22.531 26.291 35.603 33.393 45.808 65.723
N 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374 1,374
Table 11
Vector autoregression, market: Innovation correlations and Granger causality tests
Using daily data over the pre-crisis period (December 4, 2001 to June 29, 2007), we run vector
autoregression with the following endogenous variables Return M , ∆OIBM , ∆Libor-OIS, ∆VIX,
∆VolumeM , and ∆Bidask M , where M refers to the market. The number of lags is determined by
the Hannan-Quinn criterion and equals four. Panels A and B report on the correlations between
VAR innovations and χ2 statistics from Granger causality tests (with p-values in brackets). a, b,
and c denote statistical significance (two-tailed) at the 1%, 5%, and 10% level, respectively.

Return M ∆OIBM ∆Libor-OIS ∆VIX ∆VolumeM ∆Bidask M

Panel A: Correlations between VAR innovations
Return M 1.00
∆OIBM 0.78a 1.00
a
∆Libor-OIS -0.12 -0.09a 1.00
∆VIX -0.79a -0.59a 0.12a 1.00
a a
∆VolumeM 0.07 0.10 0.04 0.02 1.00
∆Bidask M -0.11a -0.13a 0.02 0.08a -0.07a 1.00
2
Panel B: χ statistics from Granger causality tests. Null hypothesis: Row
variable does not Granger-cause column variable
Return M 59.095a 4.311 1.479 2.710 17.088a
(0.00) (0.37) (0.83) (0.61) (0.00)
c
∆OIBM 1.419 6.319 2.711 8.404 4.318
(0.84) (0.18) (0.61) (0.08) (0.37)
∆Libor-OIS 5.291 8.016c 6.763 4.264 0.497
(0.26) (0.09) (0.15) (0.37) (0.97)
∆VIX 3.405 6.553 12.265b 8.414c 9.783b
(0.49) (0.16) (0.02) (0.08) (0.04)
∆VolumeM 1.233 0.818 11.504b 8.906b 0.730
(0.87) (0.94) (0.02) (0.06) (0.95)
∆Bidask M 1.856 3.593 7.344 3.453 1.734
(0.76) (0.46) (0.12) (0.49) (0.79)

13
 Table 12
Granger causality tests of returns and changes in order imbalance: Liquidity groups
Using daily data over the pre-crisis period (December 4, 2001 to June 29, 2007), we run ten vector autoregression systems with the
following endogenous variables Return G , ∆OIBG , ∆Libor-OIS, ∆VIX, ∆VolumeG , and ∆Bidask G , where G refers to liquidity group
1-10. The number of lags for each VAR system is determined by the Hannan-Quinn criterion and is reported in Panel C. Panel A
reports on χ2 statistics (with p-values in brackets) from Granger causality Wald tests of Return G . For each cell, the null hypothesis
is that the row variable does not Granger cause returns for the portfolio indicated in the column heading. Panel B does the same for
∆OIBG ([see equation (8)]).

Liquidity Group
1 2 3 4 5 6 7 8 9 10
Panel A: Returns
∆Libor-OIS 6.573c 4.263 4.093 3.676 6.598 4.576 5.223 2.958 1.805 6.259c
(0.09) (0.37) (0.39) (0.45) (0.25) (0.33) (0.27) (0.57) (0.61) (0.10)
∆VIX 3.970 1.973 2.488 2.340 3.743 0.911 0.677 5.508 13.006a 22.461a
(0.27) (0.74) (0.65) (0.67) (0.59) (0.92) (0.95) (0.24) (0.01) (0.00)
∆OIBG 1.867 2.450 2.535 2.780 16.480a 4.560 0.743 3.329 0.513 2.073
(0.60) (0.65) (0.64) (0.60) (0.01) (0.34) (0.95) (0.50) (0.92) (0.56)
∆VolumeG 0.849 1.424 0.756 0.808 2.461 3.788 3.673 5.775 1.030 6.464c
(0.84) (0.84) (0.94) (0.94) (0.78) (0.44) (0.45) (0.22) (0.79) (0.09)
14

∆Bidask G 2.410 0.538 2.339 6.480 9.597c 7.558 8.815c 3.362 0.977 12.388a
(0.49) (0.97) (0.67) (0.17) (0.09) (0.11) (0.07) (0.50) (0.81) (0.01)
Panel B: Change in order imbalance (∆OIB)
∆Libor-OIS 6.099 7.451 4.399 5.900 8.873 2.228 3.162 1.401 5.112 3.313
(0.11) (0.11) (0.36) (0.21) (0.11) (0.69) (0.53) (0.84) (0.16) (0.35)
∆VIX 4.246 7.340 5.246 2.318 8.954 6.209 2.705 16.161a 9.303b 13.339a
(0.24) (0.12) (0.26) (0.68) (0.11) (0.18) (0.61) (0.00) (0.03) (0.00)
Return G 54.539 85.591 53.861 50.874a
a a a
53.748a 30.848a 30.553a 17.803a 6.471c 12.532a
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.09) (0.01)
∆VolumeG 0.364 1.763 1.237 0.214 1.128 2.017 2.280 1.703 2.021 2.167
(0.95) (0.78) (0.87) (1.00) (0.95) (0.73) (0.68) (0.79) (0.57) (0.54)
∆Bidask G 6.257c 3.446 1.233 2.587 9.800c 22.339a 45.458a 3.888 10.516b 5.029
(0.10) (0.49) (0.87) (0.63) (0.08) (0.00) (0.00) (0.42) (0.02) (0.17)
Panel C: Lags in VAR system
3 4 4 4 5 4 4 4 3 3
 Table 13
Impulse response functions: Returns and ∆OIB on ∆Libor-OIS
Using daily data over the pre-crisis period (December 4, 2001 to June 29, 2007), we run eleven vector autoregression systems with the following endogenous variables Return G ,
∆OIBG , ∆Libor-OIS, ∆VIX, ∆Volume G , and ∆Bidask G , where G refers to the market (M ) or a liquidity group (1-10). The number of lags are selected by the Hannan-Quinn
criterion and are reported in Table 12 for the groups (it is 4 for the market). Panel A reports the impulse response of Return G to a one standard deviation shock in ∆Libor-OIS,
with t-statistics in brackets and 10% lower and upper confidence bands. The impulse response functions are computed by orthogonalizing the covariance matrix of the VAR
innovations using the Cholesky decomposition with the following ordering of the endogenous variables: ∆Libor-OIS, ∆VIX, ∆OIBG , ∆Volume G , ∆Bidask G , Return G . Values
for the first five periods as well as the 20th are reported. Panel B does the same for ∆OIBG .

Liquidity Group
Market 1 2 3 4 5 6 7 8 9 10
Panel A: Returns (%)
Period 0 -0.122a -0.122a -0.118a -0.109a -0.095a -0.083a -0.080a -0.066a -0.054a -0.061a -0.032b
(-4.60) (-4.64) (-4.04) (-3.52) (-2.88) (-2.55) (-2.57) (-2.54) (-2.91) (-4.01) (-1.95)
Lower -0.165 -0.165 -0.166 -0.160 -0.149 -0.136 -0.131 -0.108 -0.085 -0.086 -0.059
Upper -0.078 -0.078 -0.070 -0.058 -0.041 -0.029 -0.029 -0.023 -0.024 -0.036 -0.005

Period 1 -0.014 -0.020 0.009 0.007 0.007 0.000 0.006 -0.022 -0.030 -0.027c -0.031c
(-0.52) (-0.77) (0.30) (0.23) (0.21) (0.01) (0.18) (-0.85) (-1.57) (-1.75) (-1.87)
Lower -0.057 -0.063 -0.039 -0.044 -0.047 -0.053 -0.046 -0.065 -0.062 -0.052 -0.058
Upper 0.030 0.023 0.057 0.058 0.061 0.054 0.057 0.020 0.001 -0.002 -0.004

Period 2 0.016 0.019 0.016 0.017 0.021 0.031 0.026 0.036 0.000 -0.011 0.001
(0.60) (0.75) (0.56) (0.55) (0.65) (0.97) (0.86) (1.40) (0.00) (-0.73) (0.09)
15

Lower -0.027 -0.023 -0.031 -0.033 -0.032 -0.022 -0.024 -0.006 -0.031 -0.036 -0.025
Upper 0.059 0.061 0.064 0.067 0.074 0.084 0.077 0.078 0.031 0.014 0.028

Period 3 0.049c 0.054b 0.049c 0.046 0.059c 0.055c 0.060b 0.042c 0.014 0.004 0.018
(1.91) (2.20) (1.74) (1.56) (1.86) (1.72) (2.00) (1.68) (0.77) (0.25) (1.15)
Lower 0.007 0.014 0.003 -0.003 0.007 0.002 0.011 0.001 -0.016 -0.020 -0.008
Upper 0.091 0.094 0.095 0.095 0.111 0.107 0.109 0.083 0.045 0.028 0.043

Period 4 -0.032 -0.027b -0.030 -0.028 -0.023 -0.032 -0.010 -0.002 0.012 -0.006c -0.011b
(-1.28) (-2.13) (-1.10) (-0.99) (-0.75) (-1.03) (-0.36) (-0.07) (0.69) (-1.66) (-2.10)
Lower -0.072 -0.048 -0.074 -0.075 -0.073 -0.083 -0.058 -0.041 -0.017 -0.012 -0.020
Upper 0.009 -0.006 0.015 0.019 0.027 0.019 0.037 0.038 0.041 -0.000 -0.002

Period 5 -0.004 -0.006 -0.008 -0.009 -0.005 -0.04 -0.008 -0.010 -0.011b -0.003 0.000
(-0.29) (-1.28) (-0.64) (-0.70) (-0.38) (-1.32) (-0.63) (-1.10) (-2.38) (-1.33) (-0.11)
Lower -0.024 -0.014 -0.028 -0.029 -0.028 -0.089 -0.029 -0.024 -0.019 -0.007 -0.003
Upper 0.017 0.002 0.012 0.012 0.017 0.010 0.013 0.005 -0.004 0.001 0.003

Period 20 0.000 0.000 0.000 -0.000 -0.000 -0.000 0.000 0.000 -0.000 -0.000 -0.000
(0.06) (0.31) (0.07) (-0.39) (-1.00) (-0.53) (0.97) (0.90) (-0.54) (-0.03) (-0.58)
Lower -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000
Upper 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
 Table 13 – continued

Liquidity Group
Market 1 2 3 4 5 6 7 8 9 10
Panel B: Change in order imbalance (∆OIB, %)
Period 0 -0.201a -0.167a -0.090 -0.128c -0.216b -0.148 -0.415a -0.191 -0.254 -0.096 -0.800a
(-3.52) (-2.85) (-1.57) (-1.77) (-2.39) (-1.37) (-3.14) (-1.36) (-1.53) (-0.44) (-3.07)
Lower -0.294 -0.264 -0.183 -0.248 -0.364 -0.326 -0.631 -0.422 -0.526 -0.459 -1.228
Upper -0.107 -0.071 0.004 -0.009 -0.067 0.030 -0.198 0.041 0.019 0.266 -0.372

Period 1 0.218a 0.151b 0.115c 0.160c 0.204c 0.057 0.281c 0.115 0.057 0.023 -0.022
(3.14) (2.09) (1.66) (1.81) (1.88) (0.43) (1.81) (0.69) (0.29) (0.09) (-0.07)
Lower 0.104 0.033 0.001 0.014 0.026 -0.159 0.026 -0.159 -0.262 -0.407 -0.525
Upper 0.333 0.270 0.230 0.305 0.383 0.273 0.537 0.389 0.376 0.453 0.481

Period 2 0.030 0.089 -0.043 0.006 -0.104 0.059 0.046 0.032 -0.048 -0.454c 0.288
(0.43) (1.24) (-0.63) (0.07) (-0.97) (0.44) (0.3) (0.19) (-0.25) (-1.75) (0.95)
Lower -0.083 -0.029 -0.157 -0.138 -0.280 -0.159 -0.208 -0.242 -0.363 -0.881 -0.210
Upper 0.144 0.207 0.070 0.150 0.072 0.276 0.299 0.306 0.267 -0.027 0.785
16

Period 3 0.077 0.069 0.144b 0.086 0.235b 0.216c 0.044 0.294c 0.085 0.318 -0.019
(1.13) (0.96) (2.11) (1.00) (2.22) (1.66) (0.29) (1.77) (0.45) (1.24) (-0.06)
Lower -0.036 -0.049 0.031 -0.056 0.061 0.002 -0.207 0.021 -0.227 -0.105 -0.514
Upper 0.190 0.186 0.256 0.229 0.410 0.430 0.295 0.566 0.396 0.742 0.477

Period 4 -0.167a -0.171a -0.147b -0.144c -0.204b -0.321a -0.175 -0.204 0.214 0.088 0.046
(-2.45) (-2.80) (-2.16) (-1.67) (-1.93) (-2.47) (-1.15) (-1.24) (1.14) (0.41) (0.19)
Lower -0.279 -0.271 -0.258 -0.286 -0.379 -0.535 -0.423 -0.475 -0.094 -0.263 -0.364
Upper -0.055 -0.070 -0.035 -0.002 -0.030 -0.108 0.074 0.067 0.522 0.439 0.456

Period 5 0.043 0.038c 0.020 0.024 0.028 -0.102 0.086 -0.086 -0.226 -0.073 -0.093
(0.73) (1.78) (0.34) (0.32) (0.31) (-0.79) (0.68) (-0.62) (-1.46) (-0.89) (-0.98)
Lower -0.053 0.003 -0.075 -0.097 -0.120 -0.314 -0.121 -0.313 -0.479 -0.209 -0.248
Upper 0.139 0.072 0.115 0.144 0.175 0.111 0.292 0.141 0.028 0.063 0.063

Period 20 -0.000 -0.000 0.000 -0.000 0.000 -0.001 0.000 0.000 -0.001 -0.000 -0.000
(-0.25) (-0.23) (0.23) (-0.47) (0.14) (-0.9) (0.47) (0.13) (-1.33) (-0.18) (-0.07)
Lower -0.000 -0.000 -0.000 -0.000 -0.000 -0.003 -0.000 -0.001 -0.002 -0.000 -0.000
Upper 0.000 0.000 0.000 0.000 0.000 0.001 0.001 0.001 0.000 0.000 0.000
ECF-SFI 06 25.1.2006 16:05 Page 24

c/o University of Geneva

40 bd du Pont d'Arve
1211 Geneva 4
Switzerland

T +41 22 379 84 71
F +41 22 379 82 77
RPS@sfi.ch
www.SwissFinanceInstitute.ch