Bidding Behavior in the Longer Term Refinancing
Operations of the European Central Bank:
Evidence from a Panel Sample Selection Model
Tobias Linzert∗
European Central Bank
Dieter Nautz
Goethe University Frankfurt
Ulrich Bindseil
European Central Bank
September 2006
Abstract
This paper analyzes individual bidding data of the longer term refinancing
operations (LTROs) of the European Central Bank (ECB). We investigate how
banks’ bidding behavior is related to a series of exogenous variables including
collateral costs, interest rate expectations, market volatility and to individual
bank characteristics like country of origin, size, and experience. A specific feature
of these auctions is that the number and composition of bidders varies over time.
Therefore, we estimate panel sample selection models to control for a bank’s
endogenous participation decision. We find that bidding strategies depend on the
banks’ attributes. Yet, different bidding behavior generally does not translate
into differences concerning bidder success. There is evidence for the winner’s
curse effect in LTROs indicating a common value component in banks’ demand
for longer term refinancing.
Keywords: Central Bank Auctions, Winner’s Curse, Panel Sample Selection Model,
Bidding Behavior, Monetary Policy Instruments.
JEL classification: E52, D44, C33, C34
Financial support by the German Research Foundation (DFG) through NA-31020102 is gratefully
acknowledged. We thank Jörg Breitung, William Greene, Marco Lagana, Tuomas Välimäki, Benedict
Weller and two anonymous referees for valuable comments and suggestions. The views expressed in
this paper are those of the authors and do not necessarily reflect those of the ECB. An earlier version
of the paper was circulated as ECB Working paper, No. 359, May 2004. Corresponding address:
Tobias Linzert, European Central Bank, Kaiserstr. 29, 60311 Frankfurt am Main, Germany. Email:
[email protected]
∗
1
Introduction
Repo auctions are the predominant instrument for the implementation of monetary
policy of the European Central Bank (ECB).1 Repo rates govern short-term interest rates and the availability of repo credit determines the liquidity of the European
banking sector. The ECB conducts repo auctions as weekly main refinancing operations (MRO) and as longer term refinancing operations (LTRO) maturing after three
months. Although MROs are the ECB’s primary policy instrument, LTROs are far
from negligible. Currently, refinancing via LTROs amounts to around 120 billion Euro
which is more than 25% of overall liquidity provided by the ECB. This paper provides
a first analysis of the empirical performance of LTROs. We investigate how money
market conditions and bidder characteristics affect banks’ bidding behavior and the
auction outcome.
Our analysis is based on a unique data set of 50 LTRO auctions conducted between
March 1999 and May 2003. Bidder codes allow us to follow bidding behavior of individual banks over time and to apply panel econometric techniques. In contrast to
Treasury bill auctions where bids are placed by a small group of primary dealers (see
e.g. Nyborg, Rydqvist, and Sundaresan, 2002), the number of bidders in LTROs is
large and varies considerably in terms of number and composition of bidders. Apparently, banks’ participation in LTROs is determined endogenously. Since a bank’s
bid amount or its average bid rate can only be observed if the bank actually participated in the LTRO, estimation is subject to a selection bias, see Heckman (1979).
Therefore, accounting for banks’ participation decision is of crucial importance for
the empirical analysis of the LTRO bidding data. This paper introduces the random
effects sample selection model into the empirical auction literature which extents the
cross sectional Heckman (1979) approach to the panel case, following e.g. Verbeek and
Nijman (1992a) or Nijman and Verbeek (1992).2
1
2
Note that the operational framework refers to the Eurosystem i.e. the twelve national central banks
of the euro area and the ECB. Since the ECB is conducting the repo auctions, the term ECB is
used throughout the paper instead of Eurosystem.
The empirical auction literature has typically ignored the selection bias problem inherent to bidding
data. Exceptions are Ayuso and Repullo (2001) and Linzert, Nautz, and Breitung (2006) where the
bid volume is analyzed in a rather inflexible Tobit model framework. Jofre-Bonet and Pesendorfer
(2003) apply the two step Heckman method but do not take into account the specific selectivity
bias correction in the panel case (time and cross sectional dimension). Our approach is more closely
related to Bruno, Ordine, and Scalia (2005) who use a fixed effects panel sample selection model
to investigate the bidding behavior in the ECB’s main refinancing operations.
1
According to the ECB the LTRO auction format should be simple enough so that
longer term refinancing is equally accessible to all banks. In the same vein, LTROs
should give a good opportunity to smaller banks which have limited access to the
interbank market to receive liquidity for a longer period. We will, therefore, estimate
the impact of size, bidding experience and country of origin on banks bidding strategies. In particular, advancing on recent studies on the bidding behavior in central
bank repo auctions (see e.g. Bindseil, Nyborg, and Strebulaev, 2004, Bruno, Ordine
and Scalia, 2005 or Linzert, Nautz, and Breitung, 2006), we also shed some light on
the determinants of bidding success.
In line with the empirical literature on financial auctions, we characterize money market conditions by prevailing interest rate expectations and interest rate uncertainty.
Since longer term repos are collateralized central bank credits, we also consider the
cost of collateral on the interbank market as opportunity costs for LTRO refinancing.
With respect to the role of small banks in LTROs assigned by the ECB, we are particularly interested in how size affects a bidder’s response to changes in collateral costs,
interest rate expectations, and interest rate uncertainty.
Many financial auctions are natural examples of common value auctions where the
value of the auctioned good is uncertain. Therefore, bidding in common value auctions must rely on estimates of the unknown value. This explains the winners curse
effect: Naive bidders who win the auction typically realize that they had been overly
optimistic. Of course, rational bidders account for this effect. In particular, optimal
bidding in common value auctions entails to bid more cautiously when the uncertainty about the goods value rises. This implication has been widely used to confirm
the empirical relevance of the winners curse effect in Treasury bill auctions, see e.g.
Gordy (1999), Bjonnes (2001), and Nyborg, Rydqvist, and Sundaresan (2002). However, Bindseil, Nyborg, and Strebulaev (2004) and Bruno, Ordine, and Scalia (2005)
found only weak evidence in favor of the winner’s curse effect in the ECB’s MROs. It
appears that the demand for reserves in MROs is more closely related to the known
liquidity needs of individual banks and thus influenced less by uncertain market conditions. Therefore, empirical evidence on the winner’s curse effect in LTROs could
reveal information about the role of LTROs in the monetary framework of the ECB
and the use of longer term refinancing.
Our main results can be summarized as follows. First, we confirm that selection bias
2
is an issue in our auction data and employ a panel sample selection model accordingly.
Second, we find that bidding behavior is influenced by a bank’s size and its country
of origin. However, differences in bidders’ response to various exogenous variables do
not necessarily imply that banks bid with different success. Typically, banks with
lower refinancing costs must realize lower cover to bid ratios and vice versa. Thus,
different bidding strategies in LTROs seem to reflect different attitudes towards the
risk of going out empty handed. Even bidder experience is not an important issue
with respect to a bank’s success. Third, in contrast to the findings from MROs, there
is evidence for the winner’s curse effect in LTROs. In line with the theory of common
value auctions, banks’ participation, the bid volume and the bid rate decrease when
interest rate uncertainty rises. And finally, banks’ bidding in MROs and LTROs must
not be seen as independent. In fact, there are significant spill over effects from MROs
to LTROs, i.e. banks use LTROs to adjust the liquidity position from MROs.
The rest of the paper will proceed as follows. The next section describes the role of
LTROs in the operational framework of the ECB and the institutional background.
Some descriptive statistics on bidder behavior and performance are given in Section 3.
Section 4 introduces the variables that enter our panel regressions and discusses how
those might affect bidding behavior. The empirical results on banks’ participation
decision and their bidding behavior in terms of bid volume, bid rate and bid rate
dispersion are given in Section 5. Section 6 presents our results concerning banks’
bidding success and Section 7 offers some policy conclusions.
2
The Role of LTROs within the Operational Framework
of the ECB
The role of MROs and LTROs in the ECB’s operational framework for monetary policy
differs in two main respects. First, due to their higher frequency and their shorter
maturity, MROs are designed to actively steer the liquidity of the banking sector. In
particular, the repo volume allotted via MROs is not predetermined by the ECB but
responds to liquidity shocks and possibly to the bids of the banks. By contrast, the
allotment volumes of LTROs are changed only infrequently and are always fixed at
the auction’s outset, see Figure 1. Therefore, LTROs are not used to steer money
market conditions but provide banks with a basis stock of reserves that is unrelated
to any short-term liquidity fluctuations.
3
Figure 1: Liquidity Needs of the Banking System and Liquidity Supply (in EUR billion, daily
averages)
The second crucial difference between MROs and LTROs concerns the influence of
the ECB on the interest rate. MROs play the dominant role in steering short term
interest rates. The MRO rate serves as the ECB’s key interest rate which is either
explicitly set by the ECB (if it is conducted as fixed rate tender) or (in case of a
variable rate tender) at least strongly signaled by a pre-announced minimum bid rate,
see also Figure 4. By contrast, LTROs are always conducted as pure variable rate
tenders, i.e. without a minimum bid rate. The ECB simply accepts the allotment
rates resulting from its pre-announced supply of liquidity and the demand for LTROs
submitted by the banks. Acting as a price-taker, the ECB cannot use LTROs for
signaling intended interest rate levels.
4
Apart from the omitted minimum bid rate and the pre-announced allotment volume,
the rules of LTRO and MRO auctions are identical. Each bank can submit bids at up
to ten different bid rates at the precision of one basis point (0.01%). Both auctions are
price-discriminating, i.e. every successful bidder has to pay its bid. In both refinancing
operations, banks need to deposit eligible collateral with the Eurosystem to cover the
amounts allotted.
In order to leave enough room for manoeuvre in controlling the liquidity supply
through MROs, the share of LTROs in the total liquidity provision of the ECB has
been between 20% and 25%. Yet it is not obvious whether this share is appropriate.
On the one hand, it is argued that LTROs are superfluous for an efficient liquidity
management and should be abandoned in order to increase the transparency and simplicity of the ECB’s operational framework. On the other hand, if LTROs allow the
banks to improve the efficiency of their liquidity management without introducing
market distortions, then the importance of LTROs should even increase, not decrease.
3
A First Look at the Data
3.1
The Sample
Our data set consists of individual bidding data of all regular refinancing operations
(MROs and LTROs) conducted by the ECB from March 1999 to May 2003.3 The
focus of our analysis is on the performance of the 50 LTROs executed in this period.
6776 credit institutions in the euro area fulfilled the general conditions to participate
in the ECB’s regular refinancing operations but a lot of banks refrained from bidding
irrespectively of the prevailing situation in the money market. We, therefore, restrict
our attention to those 1809 banks that participated at least once in either a MRO or
a LTRO.
3.2
Number of Bidders
The average number of bidders in LTROs was 232. However, similar to the trend
observed in MROs the number of bidders in LTROs declines over time, see Figure 2
3
In January and February 1999 LTROs were performed as uniform price auctions in which every
successful bidder pays the marginal rate.
5
100
150
Number of Bidders
200
250
300
350
Figure 2: The Number of Bidders in LTROs
01/99
07/99
01/00
07/00
01/01
07/01
Auction Date
01/02
07/02
01/03
Notes: The number of bidders shown in the Figure refers to LTROs conducted from
March 1999 until May 2003.
and Bindseil, Nyborg, and Strebulaev (2004).4 Possible explanations for the declining
trend in both open market operations are the ongoing process of concentration and
rationalization in the banking sector and the increased efficiency of the interbank
market. The declining trend and the fact that the composition of bidders changes
across auctions suggests banks’ participation to be endogenous. Therefore, a sample
selection bias is likely to be an issue in the preceding econometric analysis.
3.3
Bid Rate Dispersion
In contrast to MROs, where many banks place their whole bid volume at a single
interest rate, the average number of bids per bank in a LTRO is close to 4. One
immediate explanation for the higher number of bids in LTRO auctions is the absence
of the minimum bid rate that constrains bidders in the MROs. More information about
the bidding strategies of banks is revealed in the variable bid rate dispersion defined as
the quantity weighted standard deviation of a bank’s bid rates. Figure 3 shows for each
LTRO the (unweighted) average of all bank specific bid rate dispersions. Apparently,
banks place their bids in a range of 2-3 basis points. Remarkable exceptions are
4
Notice, however, that this trend is not size specific. We control for the falling number of bidders by
inclusion of a logarithmic trend in the following regressions. The results are, however, qualitatively
robust in terms of the specific type of trend (linear, logarithmic, logistic) that is included into the
regressions.
6
.005
.01
Bid Rate Dispersion
.015
.02
.025
.03
Figure 3: The Aggregate Bid Rate Dispersion
01/99
07/99
01/00
07/00
01/01
07/01
Auction Date
01/02
07/02
01/03
Notes: The aggregate bid rate dispersion is an unweighted average of all individual bid
rate dispersions and is displayed over all LTROs in the sample period from March 1999
until May 2003. Note that the peak in bid rate dispersion is due to the Y2K effect.
the three tenders prior to the turn of the century at the end of 1999 where the bid
rate dispersion sharply increases.5 Therefore, the relatively large number of bids
submitted in LTROs exaggerates the extent to which banks distribute their bid amount
at different interest rates.
Typically, the marginal rate of the LTRO auctions is very close to the corresponding
money market rate, i.e. the three-month repo rate valid at the allotment day of the
auction, see Figure 4. There are only a few cases where some banks submit bids
at interest rates very much below the market consensus. Moreover, those bids were
negligible in terms of the bid amount indicating that they stem from small uninformed
bidders and not from bidder rings.
3.4
Size Effects
In the present study, we define a bank’s size with respect to its average reserve requirement. Specifically, 737 banks are called small because their average reserve requirement is below 10 million Euro. For 650 medium banks the reserve requirements
ranges from 10 million to 100 million Euro and for 141 large banks the reserve re5
Our results are robust with respect to the Y2K effect. In particular, leaving out the period until
December 1999 in our sample does not change the results in a significant way.
7
2.5
3
Interest Rates
3.5
4
4.5
5
Figure 4: LTRO Rates, 3-Month Repo Rate and MRO Minimum Bid Rate
01/99
07/99
01/00
07/00
01/01
07/01
Auction Date
LTRO Rate
MRO Minimum Bid Rate
01/02
07/02
01/03
3−M Repo
Notes: The LTRO rate displayed refers to the marginal rate of the corresponding tender.
The 3-month Repo is taken from the day of the deadline for counterparties’ submission
of bids. The rates are displayed from March 1999 to May 2003.
quirements exceeds 100 million Euro.6 The share of small, medium, and large banks
in total reserve requirements is 5%, 25% and 70%, respectively.
In line with the ECB’s original intention, LTROs are a relatively more important
refinancing tool for small and medium banks. On average, small banks satisfy 35%
of their refinancing demand with LTROs while the share is only 20% for large banks.
Compared to their bid volume, small banks receive a relatively high share (9.3% vs.
6.9%) of the total allotment whereas large banks get a small proportion (44.2% vs.
57.5%), see Table 1, columns (5) and (6). This already indicates that the bidding
strategies of large banks may differ from those of small banks.
Table 1 provides more information on the different bidding strategies of small, medium
and large banks. Column (7) of Table 1 presents the average spread between the
quantity weighted bid rate and the marginal rate for the various size groups. While
small banks bid on average very close to the marginal rate (−0.006), large banks bid on
average about 2 basis points lower. While large banks tend to receive their allotment
6
The reserve requirements refer to the period from February 1999 to August 2001. This data was
available for 1528 of the 1809 banks under consideration. All statistics and regressions accounting
for size effects are thus based on the data of these 1528 banks. It is important to note that this
sample reduction does not imply an obvious selection bias. In particular, the banks that dropped
out of the sample included small and large as well as active and inactive bidders coming from all
over the euro area.
8
at lower cost, the average cover to bid ratio is only 28% compared to 50% for small
banks. Therefore, bidding at lower interest rates decreases the volume allotted and
increases the risk of going out empty handed. Notice that the differences in the cover
to bid ratios across size groups are far less distinct in MROs. Apparently, bidding
strategies in MROs differ from those applied in LTROs, see columns (9) and (10) in
Table 1.
3.5
Country Effects
The number of banks taking part in the ECB’s open market operations differs considerably across countries. The 1235 German banks form by far the largest group of
bidders while only 10 banks come from Finland, see Table 2 in the Appendix. According to Table 2, the preference of national banking systems for LTRO versus MRO
refinancing differs across countries. LTROs are particulary relevant for banks from
Austria, Germany, Ireland, and Portugal while for banks located in Greece and Italy
LTROs play only a negligible role.
4
Variables and Theoretical Predictions
4.1
What to Explain? Variables Measuring Bidder Behavior
Each bank has to decide whether to refinance through a LTRO or through an alternative source of refinancing, like a MRO or the interbank money market. In Section 5.3,
we investigate how a bank’s participation decision with regard to a LTRO depends
on various auction as well as bidder-specific factors. In Section 5.4, we estimate the
impact of these factors on the quantity of refinancing demanded by each bank, i.e.
the log of the individual bid amount. Furthermore, we examine the determinants of
the price at which banks demand reserves in a LTRO auction, measured as the quantity weighted average bid rate. In order to account for changes of the overall interest
rate level, the actual variable explained in the regressions is the spread between the
weighted average bid rate and the 3-month repo rate observed in the money market.7
Since banks are allowed to bid at up to ten different interest rates, the average bid
rate of a bank neither determines the volume allotted nor the average interest rate to
7
The 3-month repo rate quotes are taken at 9:15 just prior to the end of the bidding period for the
LTRO at 9:30. Indeed, most bids are submitted in the last 15 minutes of the auction.
9
be paid. Understanding banks’ bidding behavior in LTROs also requires an analysis
of the distribution of bids. To that aim, we examine the factors influencing the individual bid rate dispersion, defined for each bank as the quantity weighted standard
deviation of its bid rates.
4.2
And How? Variables Explaining Bidder Behavior
The costs of collateral should be of particular importance for banks’ bidding since
LTRO refinancing blocks collateral and makes it thus unavailable for alternative uses
over a three-month horizon. Unfortunately, there is no exact measure of LTRO collateral cost available. We, therefore, define the variable collateral as the spread between
the three-month deposit and the three-month repo rate valid at the bidding day of the
auction. This spread measures the opportunity cost of general collateral which can be
used not only in LTROs but also in interbank operations.8 Therefore, an increase in
cost of general collateral might induce banks to increase participation in LTROs.
In MROs, expectations about future interest rates are crucial for banks’ bidding behavior. In particular, when banks expect decreasing interest rates, the pre-announced
minimum bid rate leads banks to underbid, i.e. they refrain from bidding, see ECB
(2003). In order to investigate whether rate change expectations also affect bidder
behavior in LTROs, we define the variable term spread as the difference between the
three-month repo rate and the biweekly MRO minimum bid rate. In accordance with
the expectations theory of the term structure of interest rates, e.g. a negative term
spread indicates that interest rates are expected to decline, see Figure 4.
The following estimations will also shed light on the winner’s curse effect.9 In a
common value auction, bidders may face the risk of winner’s curse when winning the
auction on the basis of ”too” positive signals about the true value of the auction good.
According to Gordy (1999) an empirically testable implication of the winner’s curse
theory is that bidder’s will disperse their bids more when uncertainty in the market
increases, i.e. when the true value of the good becomes more uncertain. Moreover,
8
9
Collateral useable for central bank operations additionally contains e.g. lower volume issues (Pfandbriefe) or non-marketable claims, which are not suitable for interbank repos which require standardization. It should also be noted that we could not account for the fact that availability of
different types of collateral varies across countries.
The winner’s curse effect was originally studied in single-unit auctions, see e.g. Milgrom and Weber (1982). Ausubel (2004) has shown that the winner’s curse (or champion’s plague) effect is
potentially more pronounced in the context of multi-unit auctions.
10
Nyborg, Rydqvist, and Sundaresan (2002) argue that if bidders submit downward
sloping demand schedules, the winner’s curse effect should be visible when bidders
obtain private information. This implies that in the case of rising uncertainty, one
should observe bidders adjusting for winner’s curse by bidding at lower rates, reducing
the quantity demanded and increasing the bid rate dispersion.
Estimating the effect of uncertainty in ECB’s open market operations can reveal the
common value character of this type of monetary policy operation. On the one hand,
common value character is supported by the fact that there is a competitive interbank
market for respective repos. On the other hand, fulfilling individual required reserves,
the availability of specific type of collateral and transaction costs in the secondary
market introduces a private value component to the bidding. In this case, bidders
would not hedge against winner’s curse. In the ECB’s MROs banks’ bidding seems
less subject to winner’s curse and to be dominated by other considerations such as
the fear of not obtaining any funds, see Bindseil, Nyborg, and Strebulaev (2004).
Bruno, Ordine, and Scalia (2005), in particular, found that an increase in interest
rate volatility lowers the probability of bidding, but induces bidders to bid at higher,
not lower, rates.10 In the following, interest rate uncertainty is proxied by the variable
volatility measured as the implied volatility derived from options on 3-month EURIBOR futures, see Figure 5. According to ECB (2002), the implied volatility is a useful
measure of the overall uncertainty associated with future movements in short-term
interest rates.
In order to investigate whether the preannounced LTRO volume influences the behavior of banks and the outcome of the auction we incorporate the variable auction size
that equals the LTRO allotment volume. If the central bank increases the intended
allotment volume, banks’ might expect the price for liquidity to decrease, thus, increasing their participation and bid volume accordingly.
We also consider several bidder-specific regressors. We define ∆CBRM RO as the
change of a bank’s cover to bid ratio of the previous two MRO tenders in the regressions. If ∆CBRM RO > 0, then a bank might have received more MRO repo credit
than expected which might affect its bidding behavior in the upcoming LTRO.11 A
10
11
This behavior is also predicted by multi-period reserve management models where higher interest
rate risk increases banks’ demand for reserves, see Nautz (1998).
∆CBRM RO is set to zero when a bank did not participate in one of the two previous MROs. Note
that the level CBRM RO is severely distorted by banks’ massive overbidding during the fixed rate
tender period, see Nautz and Oechssler (2003). For that reason, we defined ∆CBRM RO to be zero
11
3
Interest Rate Volatility
3.5
4
4.5
Figure 5: Interest Rate Volatility
01/99
07/99
01/00
07/00
01/01
07/01
Auction Date
01/02
07/02
01/03
Notes: Volatility is measured as the implied volatility derived from options on 3-month
EURIBOR futures. Source: ECB (2002).
further variable that relates LTRO bidding to banks’ behavior in MROs is MRO frequency that measures the average participation frequency of a bank in the MROs. It
should reveal whether e.g. bidders which are especially active in MROs are less or
more active in LTROs. Moreover, it can be seen as a proxy for the degree of bidding
experience of a bank.12 The variable maturing allotment is defined as the log of a
bank’s repo volume received three months before.13 If banks use LTROs on a revolving basis, the maturing amount of the repo credit should increase banks’ participation
probability and its bid volume.
We will also investigate how a bank’s size or its country of origin influences bidder
behavior. To that aim, we include a dummy variable for each size group (small,
medium, large) and country in the regressions. A bank’s size may not only influence
the level of an auction variable. Therefore, we also interact the size dummies with
the variables volatility, term spread, and collateral to investigate how size influences
12
13
in the first LTRO after the ECB’s switch to the variable rate tender format in June 2000. Using
the change in a bank’s cover-to-bid ratio should alleviate the distorting effect from the overbidding
period.
Notice, that strictly speaking the variable is not predetermined. However, it is constant over time
and can, therefore, be seen as equivalent to having extracted the information from a sub-sample of
the data. We perceive the variable to proxy the characteristics of a bank that are not captured by
the level of the reserve requirements determining a bank’s size.
The value of the variable is set to zero when the bank did not receive any allotment three months
ago.
12
a bank’s reaction to changing money market conditions.
5
Empirical Results
In this section, we investigate the determinants of a bank’s bid volume, its weighted
average bid rate, and its bid rate dispersion. These variables are left-censored since
they can only be observed if the bank decided to participate in the auction. In contrast
to Treasury bill auctions, in LTROs the participation of banks varies over time in terms
of the number and composition of banks. In this respect, banks refrain from bidding
in the auction in a non-random manner, i.e. banks are incompletely observed because
of an endogenous reason. Therefore, we are faced with a classical sample selection
problem that has to be taken into account in order to avoid biased estimates.
5.1
The Panel Sample Selection Model
Following the results of standard Hausman tests, banks’ bidding behavior in the ECB’s
LTROs is investigated by panel random effects sample selection models.14 Following
e.g. Verbeek and Nijman (1992b), we consider the following model:
yit = ci + β ′ xit + ǫit
(1)
with yit denoting banks’ bidding variable in auction t, ci the random individual and ǫit
the individual specific time effect which are both assumed to be normally distributed.
xit is the vector of observed determinants of banks’ bidding behavior, yit . The selection
process is characterized by the random effects probit equation:
∗
zit
= di + α′ wit + uit
(2)
∗
zit = 1(zit
> 0)
(3)
with zit being an indicator variable denoting whether a positive bid of a bank was
observed. di and uit represent the normally distributed random individual and individual specific time effect, respectively. The random effects, ci and di , are assumed
to be bivariate normally distributed with zero means, standard deviations σc and σd ,
14
The Hausman test indicated that the hypothesis of the individual effects being uncorrelated with
the other regressors cannot be rejected. In this case, the random effects model is asymptotically
efficient. An additional feature of random effects model is that they allow the inclusion of timeinvariant bidder-specific regressors.
13
and correlation σcd . wit is a vector of variables determining banks’ participation.15
If the response variable zit is conditionally independent of ci and ǫit , given wit , i.e.
cov(ǫit , uit ) = 0 and cov(ci , di ) = 0, estimating (1) with only observed bids is feasible. However, if zit depends on the unobservables (di , uit ), i.e. cov(ǫit , uit ) 6= 0 or
cov(ci , di ) 6= 0, then the usual random effects estimator is inconsistent. The source of
this inconsistency is a selection bias due to the impact of exogenous variables and the
selection decision on the expectation of ci and ǫit :
E [ci |wi1 , . . . , wiT , zi1 , . . . , ziT ] 6= E [ci ]
(4)
E [ǫit |wi1 , . . . , wiT , zi1 , . . . , ziT ] 6= E [ǫit ]
(5)
As shown by Heckman (1979) the selection problem can be seen as a simple omitted
variable problem. Therefore, one way to account for the selection bias, is to apply
the traditional two-step Heckman method typically used in the cross sectional case,
see Heckman (1979).16 However, in panel data there is the additional time dimension
which introduces an additional source for selection bias. In general, there is no reason
to assume that the sample selection process is time invariant. The intuition is that in
the spirit of the cross sectional Heckman correction, the sample selection correction
in the random effects panel case is based not only on accounting for correlation in
the random individual effects, ci and di but also on accounting for the correlation
between the idiosyncratic time effects, ǫit and uit , see Nijman and Verbeek (1992).
Our estimations build on the random effects sample selection estimators suggested in
Verbeek and Nijman (1992a) and Zabel (1992). We particularly employ the maximum
likelihood estimator to obtain efficient estimates of the parameters in our model.
15
16
Note that wit does not necessarily need to be identical with xit . In this respect, the sample selection
model used here is a more general approach than the traditional Tobit model. The Tobit model can
be too restrictive for many applications. In particular, in a Tobit model all variables determining
participation affect the censored variables with the same sign by construction. Obviously, this
assumption is not plausible in all cases: for example, one should expect that the participation
probability decreases when bid rates increase. This implies that variables that positively affect
the participation decision may have a negative impact on bid rates, and hence the signs of their
respective coefficients cannot be the same.
This traditional approach is based on the assumption that E[ǫit , zit = 1] can be expressed as a
linear function of the errors of the selection equation, uit . In the first stage, the inverse mills ratio,
φ(α′ wit )
, is constructed from the probit regression. In the second stage, the inverse Mills ratio is
Φ(α′ wit )
included as an additional variable into the primary equation to correct for sample selection bias.
Notice that the usual standard errors obtained from OLS routines are not valid if cov(ǫit , uit ) 6= 0.
Corrected standard errors are, however, available in today’s econometric packages like LIMDEP.
14
5.2
Testing for Selection Bias
To test whether selection bias is in fact an issue in the LTRO bidding behavior regressions, we employed the variable addition tests as suggested in Verbeek and Nijman
(1992b). The basic idea is that if some simple additional variables are added to the
regressions they could approximate the true correction for the selection bias. In particular, as the selection bias stems from a systematic pattern of missing observations,
a simple way to check whether such influence is present is to include a variable in the
model that captures the effect of missing data, e.g. the number of auctions an individual bank participates in the LTROs. The participation frequencies in our sample
vary strongly across individual banks. While about 800 banks never participated in
an LTRO, only 4 banks participated in all auctions.
Clearly, the information that an individual bank was observed in any of the auctions
1 to T should not be relevant for the bank’s unobservables in the model. Therefore,
following Verbeek and Nijman (1992b) we add three variables to the model in order
to check for selection bias in our data. First, the number of auctions an individual
P
bank i participates ( Ts=1 zis ) , second a dummy variable indicating whether a bank
Q
i is observed in all periods ( Ts=1 zis ) and third a dummy variable indicating whether
an individual bank i is observed in the previous period (zi,t−1 ). The results of the
tests shown in Table 3 indicate clearly the presence of selection bias in the regression
for bid volume, bid rate and bid rate dispersion as at least one of the three variables
is statistically significant.
5.3
The Participation Decision
Having shown that selection bias is an issue, in a first step, we present the results
from banks’ participation decision in the repo auctions using the first stage panel
random effects probit model estimated on 1528 banks, see Table 4, column (2), in
the Appendix.17 In line with the winner’s curse hypothesis, banks decrease their
participation significantly as interest rate volatility increases. If interest rate volatility
is high, banks bid more reluctantly in LTROs because they fear to bid at interest rates
above the uncertain market consensus.
17
Recall that we estimate the model for those 1528 banks for which reserve requirement data is
available to construct the respective size variable, compare Footnote 6 for details.
15
The significant coefficient of the variable term spread implies that e.g. banks participation decreases if interest rates are expected to fall. In contrast to MROs where the
minimum bid rate makes bidding less attractive when repo rates are expected to fall,
there is no obvious explanation for the impact of rate expectations on the bidder behavior in LTROs. However, the economic significance of rate expectations, measured
by its marginal effect, is only small. Assuming a fairly large term spread of about 50
basis points would lead to a drop in participation by only 3%.
An increase in collateral which indicates higher opportunity cost of general collateral
does not exert any significant effect on banks’ participation decision. Maturing allotment has the expected positive effect on banks’ participation indicating that banks
use LTROs on a revolving basis. Moreover, the variable trend indicates a significant
negative trend in banks participation as the number of bidders has been dropping
since 1999.
The probit model reveals interesting relations between banks’ bidding behavior in
MROs and LTROs. The significantly negative coefficient of ∆CBRM RO shows that
banks tend to participate less when they realized an unexpectedly high allotment
amount in the previous MRO and vice versa. This suggests that banks’ demand
for refinancing in LTROs depends on the refinancing they received in current MROs
which contradicts the often perceived segmentation of the two refinancing markets.
Moreover, the variable MRO frequency shows that LTROs are used more frequently
by bidders who are also active in MROs.
The probit analysis confirms that medium and large banks use LTROs significantly
more often than small banks. Thus, LTROs should not be seen as a monetary policy
instrument especially designed for small banks. In Table 4, differences in the average level of participation across the 12 EMU countries are measured by 11 countrydummies and a constant. The model uses Germany as a reference country, such that
the coefficients of the dummy variables show whether the average participation level in
a certain country is higher or lower than in Germany. The results indicate some significant country differences. In line with the descriptive statistics, Finish and Portuguese
banks participate significantly more frequently in LTROs than banks from Germany.
In contrast, the average participation frequency of Spanish banks is significantly lower.
16
5.4
Bid Volume, Bid Rate, and Bid Rate Dispersion in LTROs
The results from the panel sample selection model analyzing banks’ bid amount, its bid
rate and the bid rate dispersion is shown in Table 4. As expected, the larger the bank
the higher is its average bid volume. In line with the descriptive statistics (compare
Table 1) large bidders bid at significantly lower rates than small banks. Moreover,
large banks bid’s are also significantly less dispersed than smaller banks. Bidding
experience does not affect bidding behavior in the expected direction. Particularly,
very active bidders in MROs (MRO frequency) bid on a wider range and at higher
rates.
The significant influence of volatility on the banks’ bid volume and their average
bid rate confirms the evidence for the winner’s curse effect obtained from the probit
analysis. In line with auction theoretical predictions, banks bid at lower rates and
reduce their bid volume as interest rate uncertainty increases. However, as in MRO
auctions (see Bindseil, Nyborg, and Strebulaev, 2004) the negative effect of volatility
on the bid rate dispersion somewhat blurs the evidence for the winners curse effect.
There are significant differences in banks’ bidding behavior depending on a bank’s
country of origin.18 For example, Italian banks bid significantly larger volumes and
at lower rates than banks from the reference country Germany. In line with the
descriptive statistics (Table 2), Dutch banks bid at significantly higher interest rates
than German banks.
Overall, with respect to the sign and significance of the estimated parameters, the
results obtained for a bank’s bid volume, its bid rate and its bid rate dispersion are
very much in line with those obtained for the participation decision. In particular,
each factor that e.g. increases a bank’s probability of participation in LTROs also
increases its bid volume.
5.5
Size-specific Determinants of LTRO Bidding Behavior
Table 5, column (2) shows the results of an extended probit model where the influence
of collateral costs, the term spread, and volatility on a bank’s participation are allowed
to dependent on the bank’s size, as measured through its reserve requirement. These
18
Notice, however, that caution is warranted in interpreting the country effects since these dummy
variables merely account for what is not explained by the remaining explanatory variables.
17
size-effects are implemented by interacting the size-dummies with the variables of
interest. For example, the single variable volatility is replaced by the three size-specific
variables volatilitysmall , volatilitymedium , and volatilitylarge .
The extended probit model demonstrates that a bank’s size does not only affect the
average level of participation. In fact, size-dependent coefficients are significant for
the term spread as well as for interest rate volatility. First, small banks react more
pronounced to the term spread, i.e. on changes in interest rate expectations. However,
even for small banks there is no evidence that rate cut expectations can lead to bidder
strikes in LTROs as observed in MROs, see Linzert, Nautz, and Breitung (2006).
Second, very much in line with the predictions implied by the winner’s curse effect,
the participation decision of large banks depends stronger on prevailing interest rate
uncertainty. Since large banks are more active in the money market, they are more
interested in the common value component of reserves.
Table 5, columns (3)-(5), presents the extended models allowing for size specific coefficients for the bid volume, the bid rate and bid rate dispersion. Generally, a bank’s
size seems to be less important for the bidding variables than for the participation
decision. In particular, the coefficients in the bid volume equation do not depend on
a bank’s size at all. The impact of rate expectations is size-dependent for the bid rate
and the bid rate dispersion. In both cases, there is a weaker response to the term
spread of large banks.
The influence of collateral cost on banks’ bid rate and its dispersion is also sizespecific. Specifically large and medium banks spread their bids more when collateral
becomes more expensive in the interbank repo market. In line with results obtained
for banks’ participation decision, this points to a less active collateral management of
small banks.
6
Bidding Success
In the preceding sections, the analysis of bidder behavior in the ECB’s LTRO auctions
revealed that a bank’s bidding strategy can depend on its size, its country of origin,
and its bidding experience. In this section, we investigate whether the observed bidder
heterogeneity implies that certain types of bidders are systematically more successful
than others. In this case, LTRO auctions may be seen as unfair and the principle
18
of equal treatment emphasized by the ECB might be violated. Moreover, bidding in
LTROs should be sufficiently simple to ensure an appropriate access to longer term
refinancing even for less informed or less sophisticated bidders.
Since a bank’s true demand for repo credit is only partly revealed in its bid, measuring
the success of a bank in a LTRO auction is not straightforward. For example, suppose
a bank left a LTRO empty handed because it bid only at interest rates below the
marginal rate. From the bank’s perspective, this auction was no success only if the
low bidding rates resulted from a misperception of the situation in the money market.
If, however, the bank bid seriously and the marginal rate of the LTRO simply exceeds
the bank’s willingness to pay then a zero allotment is a ”successful” auction outcome.
In view of these problems we employ two different measures of banks’ success. First,
we assume that a banks’ success increases with its individual cover to bid ratio. This
measure captures the plausible idea that banks are the more successful the more
refinancing they receive (relative to their bid). However, according to this measure
bidding at unrealistically high interest rates would be an expensive but successful
strategy. Therefore, as a complementary measure of a bank’s success we define the
variable refinancing cost as the spread between the average allotment rate paid by an
individual bank and the marginal repo rate of the auction. Note that this measure is
not without problems either as, for example, it can indicate successful bidding even if
a bank received only a disappointingly small part of its bid volume. As a consequence,
a bank is identified to be more successful than others only if it achieves higher average
cover to bid ratios without higher refinancing cost and vice versa. Notice that since
both variables are defined with respect to aggregate bidding data which is unknown
at the auctions outset, the following empirical analysis of banks bidding success can
only serve as an ex-post evaluation of bidder performance.
The results from the two random effects panel regressions explaining banks’ individual cover to bid ratios and their refinancing cost are presented in Table 6.19 As
expected, for most regressors the signs of the estimated coefficients are the same in
both equations. In particular, medium and large banks have both, significantly lower
refinancing cost as well as lower cover to bid ratios than small banks. As a result,
the different bidding strategies of small and large banks are hard to evaluate in terms
19
Notice that we employ simple random effects regressions since the two variables are variables
describing the auction outcome and are, therefore, not subject to a sample selection bias.
19
of success. Small banks prefer a secure allotment by realizing higher cover to bid
ratios, while large banks are more flexible with respect to the allotment volume caring
more about their refinancing cost. This bidding behavior can also be observed for
banks participating very frequently in MROs (MRO frequency), i.e. more experienced
bidders.
Finally, we included country dummies into the regressions in order to investigate how
the auction outcome depends on a bank’s country of origin. Taking Germany as the
reference country, we found that refinancing cost for Italian banks are on average about
three basis points lower. Yet, Italian banks cannot be seen to be more successful than
their German counterparts because they also realize significantly smaller cover to bid
ratios. Apparently, Italian banks have different preferences regarding their means of
refinancing and hence pursue different bidding strategies. This is also reflected in the
different country shares in refinancing in LTROs, see Table 2.
There are three countries where banks appear to bid less successful than those from
Germany. Banks from Belgium and Greece realize significantly lower cover to bid
ratios but their refinancing cost are not significantly lower. Banks from Netherland
pay a higher interest rate without receiving a higher allotment. By contrast, bidders
from Ireland seem to bid particularly successful in LTRO auctions. While their refinancing cost are not significantly higher than those from German bidders, Irish banks
nevertheless achieve a significantly higher cover to bid ratio. Note that especially in
Ireland many credit institutions have been established by foreign mother companies
for the sake of perceived advantages of that location.
7
Conclusions
During the first five years of the Eurosystem up to 120 billion Euro, i.e. more than 25%
of banks’ total refinancing was provided by the ECB through longer term refinancing
operations (LTROs). This paper analyzed individual bidding data of LTRO auctions
shedding light on banks’ bidding behavior as well as their bidding success. Banks’
bidding can only be observed if a bank actually participates in an auction. In contrast
to Treasury bill auctions, in LTROs participation frequency differs considerably across
individual banks. A distinguishing feature of our contribution is that the empirical
analysis of banks bidding behavior is based on a random effects panel sample selection
20
models that accounts for the endogeneity of banks participation decision and avoids
any selectivity bias.
One of the original intentions of the ECB when establishing LTROs was to give smaller
banks with only limited access to the interbank market a comfortable source of longer
term refinancing. However, our results do not substantiate the notion that LTROs are
especially designed for and used by smaller banks. In particular, the results from a
panel probit model reveal that a bank’s participation probability in a LTRO increases
with its size.
According to the ECB, an important requirement for its refinancing operations is
that the auction procedure does not violate the principle of equal treatment. In
particular, certain types of banks should not bid a priori more successfully in LTROs
than others. Although we find significant differences in the bidding behavior across
banks of different size and from different countries, the resulting differences in terms
of bidding success were astonishingly small. The analysis of banks’ cover to bid ratio
and their refinancing cost showed that different bidding strategies in most cases do
not imply an obvious ranking of banks in terms of bidding success. For example,
small banks realize higher cover to bid ratios in LTROs than medium and large banks
but also tend to have higher refinancing cost. Moreover, we found that experienced
bidders are not significantly more successful in LTROs than less experienced ones.
Apparently, the LTRO auctions are sufficiently simple and transparent.
The analysis of banks’ bidding behavior provided strong evidence for the winner’s curse
effect in LTROs. In line with theoretical predictions, banks reduce their participation,
bid at lower interest rates and reduce their bid volume as interest rate uncertainty
increases. Interestingly, large banks react stronger to rate uncertainty than small
banks. This indicates that large banks are particulary interested in the common value
component of the longer term refinancing since they have on average a more active
interbank money market desk. The finding of a winner’s curse effect in LTROs is
in marked contrast to the evidence from the ECB’s main refinancing operations, see
e.g. Bindseil, Nyborg, and Strebulaev (2004) and Bruno, Ordine, and Scalia (2005).
This suggests that the private value component of central bank refinancing is more
pronounced in MROs than in LTROs where common money market conditions seem
to be more important for banks’ bidding behavior.
21
References
Ausubel, L. (2004): “An Efficient Ascending-Bid Auction for Multiple Objects,”
American Economic Review, 94(5), 1452–1475.
Ayuso, J., and R. Repullo (2001): “Why Did the Banks Overbid? An Empirical
Model of Fixed Rate Tenders of the European Central Bank,” Journal of International Money and Finance, 20, 857–870.
Bindseil, U., K. G. Nyborg, and I. A. Strebulaev (2004): “Bidder Behavior
and Performance in Repo Auctions: The Case of the Eurosystem,” Discussion Paper
4367, CEPR.
Bjonnes, G. H. (2001): “Winner’s Curse in Discriminatory Price Auctions: Evidence
From the Norwegian Treasury Bill Auctions,” Working Paper, Stockholm Institute
for Financial Research.
Bruno, G., M. Ordine, and A. Scalia (2005): “Banks’ Participation in the Eurosystem Auctions and Money Market Integration,” Discussion paper no. 562, Bank
of Italy.
ECB (2002): “Measures of Implied Volatility Derived from Options on Short-Term
Interest Rate Futures,” Monthly Bulletin, (May), 13–16.
(2003): “Changes to the Eurosystem’s Operational Framework for Monetary
Policy,” Monthly Bulletin, (August), 41–54.
Gordy, M. (1999): “Hedging Winner’s Curse with Multiple Bids: Evidence from the
Portuguese Treasury Bill Auction,” The Review of Economics & Statistics, 81(3),
448–465.
Heckman, J. (1979): “Sample Selection Bias as a Specification Error,” Econometrica,
47, 153–161.
Jofre-Bonet, M., and M. Pesendorfer (2003): “Estimation of a Dynamic Auction Game,” Econometrica, 71(5), 1443–1489.
Linzert, T., D. Nautz, and J. Breitung (2006): “Bidder Behavior in Central
Bank Repo Auctions: Evidence from the Bundesbank,” Journal of International
Financial Markets, Institutions and Money, 16, 215–230.
22
Milgrom, P., and R. J. Weber (1982): “A theory of auctions and competitive
bidding,” Econometrica, 50, 1089–1122.
Nautz, D. (1998): “Banks’ Demand for Reserves When Future Monetary Policy is
Uncertain,” Journal of Monetary Economics, 42(1), 161–183.
Nautz, D., and J. Oechssler (2003): “The Repo Auctions of the European Central
Bank and the Vanishing Quota Puzzle,” Scandinavian Economic Journal, 15(2),
207–220.
Nijman, T., and M. Verbeek (1992): “Nonresponse in Panel Data: The Impact on
Estimates of a Life Cycle Consumption Function,” Journal of Applied Econometrics,
7, 243–257.
Nyborg, K. G., K. Rydqvist, and S. Sundaresan (2002): “Bidder Behavior
in Multiunit Auctions: Evidence from Swedish Treasury Auctions,” Journal of
Political Economy, 110, 394–425.
Verbeek, M., and T. Nijman (1992a): “Incomplete Panels and Selection Bias,” in
The Econometrics of Panel Data, ed. by L. Matyas, and P. Sevestre, pp. 262–302.
Kluwer.
(1992b): “Testing for Selectivity Bias in Panel Data Models,” International
Economic Review, 33(3), 681–703.
Zabel, J. E. (1992): “Estimating Fixed and Random Effects Models with Selectivity,” Economics Letters, 40, 269–272.
23
Table 1: Bidder Behavior in LTROs: The Role of Banks’ Size
24
Number of
Banks
Participation
Frequency
LTRO
Participation
Frequency
MRO
Share in Total
Bid Volume
Share in Total
Allotment
Bid
Rate minus
Marg. Rate
Allotment
Rate minus
Marg. Rate
Cover/Bid
Ratio
LTRO
Cover/Bid
Ratio
MRO
Small
737
8.8%
20.8%
6.9%
9.3%
-0.006
0.042
0.50
0.69
Medium
650
16.9%
38.6%
35.6%
46.5%
-0.008
0.028
0.48
0.59
Large
141
29.4%
51.5%
57.5%
44.2%
-0.023
0.020
0.28
0.62
Notes: Small banks are those with requirement below 10 million Euro. Banks with reserve requirements ranging from 10 million to 100 million were grouped as medium and
banks with reserve requirements above 100 million were classified as large banks. The cover to bid ratio for the MROs is calculated on the basis of the variable rate tender period
starting in June 2000 only. All other number refer to the LTRO auctions conducted between March 1999 and May 2003.
Table 2: Bidder Behavior in LTROs: The Role of Banks’ Country of Origin
Country
Number
of Banks
Participation
Frequency
LTRO
Participation
Frequency
MRO
Total
Refinancing
(million Euro)
Share in
MRO
Share in
LTRO
Bid Rate
minus
Marg. Rate
Allotment
Rate minus
Marg. Rate
Cover/Bid
Ratio
LTRO
Cover/Bid
Ratio
MRO
Austria (AT)
37
29.7%
42.4%
4013
67%
33%
0.003
0.037
0.32
0.48
Belgium (BE)
28
10.5%
30.7%
9670
84%
16%
-0.023
0.030
0.13
0.67
Finland (FI)
10
12.8%
24.6%
1124
75%
25%
-0.015
0.033
0.22
0.58
France (FR)
74
14.9%
40.8%
17349
83%
17%
-0.029
0.027
0.21
0.52
1235
12.9%
28.5%
106822
68%
32%
-0.007
0.032
0.43
0.62
Greece (GR)
12
5.0%
15.2%
918
99%
1%
0.010
0.033
0.13
0.74
Ireland (IE)
34
16.8%
28.0%
8375
39%
61%
0.017
0.031
0.68
0.82
Italy (IT)
94
6.3%
30.9%
16375
96%
4%
-0.052
0.019
0.10
0.66
Luxembourg (LU)
80
14.1%
29.2%
16300
81%
19%
-0.011
0.018
0.34
0.66
Netherlands (NE)
54
8.4%
17.0%
6940
86%
14%
0.008
0.058
0.17
0.46
Portugal (PT)
38
12.9%
14.8%
1765
38%
62%
0.012
0.048
0.60
0.68
Spain (ES)
113
7.8%
21.9%
14247
81%
19%
-0.016
0.029
0.38
0.75
Germany (GE)
25
Notes: The total refinancing volume refers to the average recourse to open market operations over the sample period from March 1999 to May 2003. The numbers are based on
balance sheet data of the national central banks showing banks’ total recourse to MRO and LTRO in the particular country. The bid rate and the allotment rate are quantity
weighted average rates.
Table 3: Test of Sample Selection in Bidding Data in ECB’s LTROs: Results from Panel
Regressions
PT
s=1 zis
QT
s=1 zis
zi,t−1
Bid Volume
Bid Rate
Bid Rate
Dispersion
0.14
0.001
0.0001
(76.51)
(5.44)
(2.42)
-3.59
0.002
0.0001
(-58.44)
(1.60)
(0.16)
15.11
0.01
0.0003
(203.79)
(0.37)
(0.07)
Notes: The t-values of the parameter estimates are reported in parenthesis. The estimations
include
all
P
Q
variables presented in Section 4 but are omitted for brevity reasons. The variables Ts=1 zis , Ts=1 zis
and zi,t−1 indicate the number of auctions an individual bank i participated, a dummy variable taking
the value 1 if bank i was observed in all periods and a dummy variable taking the value 1 if bank i was
observed in the previous period, respectively, Verbeek and Nijman (1992b) for details.
26
Table 4: Bidding Behavior in LTROs: Results from Panel Sample Selection Model Regressions
Probit Model
Collateral Costs
Term Spread
Volatility
Maturing Allotment
MRO Frequency
Auction Size
∆CBRM RO
Sizemedium
Sizelarge
Trend
BE
ES
FI
FR
GR
IE
IT
LU
NE
PT
Constant
Bid Rate
Bid Rate
Dispersion
-0.17
0.07
0.63
0.02
(-1.87)
(1.25)
(105.97)
(12.57)
0.28
0.20
-0.04
0.006
(7.16)
(10.93)
(-18.23)
(10.64)
-0.17
-0.19
-0.04
-0.004
(-3.24)
(-6.52)
(-8.35)
(-3.79)
0.07
0.01
0.001
0.0001
(54.86)
(19.97)
(10.73)
(2.41)
2.44
0.72
0.008
0.004
(36.96)
(30.04)
(2.87)
(5.51)
0.77
0.17
-0.07
0.02
(15.22)
(5.20)
(-16.33)
(21.58)
-0.28
-0.12
-0.006
-0.001
(-7.14)
(-5.50)
(-1.42)
(-1.74)
0.20
1.79
0.002
-0.0004
(4.33)
(140.03)
(1.42)
(-1.17)
0.36
2.49
-0.01
-0.001
(5.61)
(126.53)
(-3.45)
(-2.34)
-0.25
0.52
0.45
0.02
0.005
(5.46)
(16.09)
(5.41)
(8.13)
(-16.15)
AT
Bid Volume
0.01
0.86
-0.007
-0.007
(0.05)
(14.45)
(-0.89)
(-4.12)
-0.36
0.48
-0.04
-0.004
(-5.87)
(15.63)
(-12.11)
(-4.58)
0.16
2.44
0.004
0.001
(4.05)
(29.15)
(0.36)
(0.45)
0.0001
-0.22
-0.07
-0.02
(-1.39)
(-2.25)
(-4.43)
(0.09)
-0.18
0.73
-0.03
-0.001
(-0.54)
(3.09)
(-0.76)
(-0.12)
-0.15
1.79
0.03
0.001
(-)
(36.57)
(4.72)
(0.67)
0.13
1.22
-0.05
-0.002
(1.92)
(41.88)
(-12.48)
(-1.91)
0.36
0.28
0.009
-0.004
(1.90)
(9.92)
(2.48)
(-4.81)
0.01
-0.60
0.04
-0.003
(0.22)
(-15.73)
(6.87)
(-2.42)
0.49
0.90
0.04
-0.002
(4.75)
(10.57)
(3.42)
(-0.62)
-8.82
14.27
0.83
-0.19
(-17.99)
(46.68)
(18.42)
(-18.26)
Notes: The t-values of the parameter estimates are reported in parenthesis. Germany is taken as the
base country so the respective dummy is omitted.
27
Table 5: Panel Sample Selection Model of Bidding Behavior: Size Dependent Coefficients
Probit Model
Maturing Allotment
MRO Frequency
Auction Size
∆CBRM RO
Sizemedium
Sizelarge
Trend
Collateralmedium
Collaterallarge
Term Spreadsmall
Term Spreadmedium
Term Spreadlarge
Volatilitysmall
Volatilitymedium
Volatilitylarge
Bid Rate
Bid Rate
Dispersion
0.07
0.01
0.001
0.0001
(55.64)
(18.91)
(17.86)
(2.72)
2.37
1.38
0.005
0.003
(24.02)
(57.33)
(2.33)
(4.85)
0.77
0.16
-0.07
0.02
(15.27)
(4.77)
(-27.93)
(20.95)
-0.27
-0.12
-0.006
-0.001
(-6.97)
(-5.14)
(-2.92)
(-1.71)
0.33
0.97
0.01
0.001
(0.91)
(3.33)
(0.47)
(0.15)
2.28
3.58
0.03
-0.009
(4.31)
(10.35)
(1.03)
(-0.73)
-0.25
-0.33
0.008
0.60
0.005
(-2.57)
(0.07)
(40.02)
(1.93)
(-16.24)
Collateralsmall
Bid Volume
-0.14
0.07
0.66
0.026
(-1.17)
(0.76)
(53.25)
(12.47)
0.23
0.17
0.54
0.021
(1.06)
(1.70)
(26.20)
(6.87)
0.45
0.26
-0.07
0.008
(7.55)
(8.35)
(-10.07)
(7.19)
0.15
0.18
-0.04
0.007
(2.93)
(6.44)
(-6.79)
(8.46)
0.24
0.18
-0.03
-0.0003
(2.58)
(4.09)
(-2.93)
(-0.34)
-0.07
-0.28
-0.03
-0.003
(-0.97)
(-4.58)
(-3.85)
(-1.95)
-0.12
-0.15
-0.04
-0.004
(-1.91)
(-3.36)
(-6.09)
(-3.09)
-0.55
-0.20
-0.06
-0.002
(-4.71)
(-3.21)
(-6.43)
(-0.79)
Notes: The t-values of the parameter estimates are reported in parenthesis. Bold numbers indicate
that the null hypothesis of no size-dependent coefficients was rejected. In this case, the response
to collateral costs, interest rate expectation and uncertainty, respectively, depends on a bank’s size.
Country dummies were included in all regressions presented in the table.
28
Table 6: Bidding Success in LTROs: Individual Cover to Bid Ratio and Refinancing Cost
Collateral Costs
Term Spread
Volatility
Maturing Allotment
MRO Frequency
Auction Size
∆CBRM RO
Cover to Bid
Ratio
Refinancing
Cost
0.05
0.18
(1.31)
(37.05)
-0.20
0.04
(-12.19)
(16.88)
0.22
0.01
(11.20)
(4.83)
0.01
0.0003
(18.18)
(5.50)
-0.13
-0.02
(-4.56)
(-4.61)
0.15
0.09
(6.56)
(28.09)
0.006
0.002
(0.42)
(0.99)
Sizemedium
-0.05
-0.01
(-3.11)
(-4.32)
Sizelarge
-0.15
-0.02
(-5.77)
(-3.57)
AT
0.03
0.007
BE
ES
FI
FR
GR
(0.80)
(1.32)
-0.20
0.006
(-3.04)
(0.49)
-0.05
-0.002
(-1.32)
(-0.37)
-0.17
0.005
(-1.91)
(0.34)
-0.08
-0.004
(-2.02)
(-0.63)
-0.35
-0.03
(-2.22)
(-0.63)
IE
0.19
0.008
(3.05)
(0.91)
IT
-0.24
-0.03
(-7.68)
(-5.54)
LU
-0.03
-0.009
NE
PT
Constant
(-0.83)
(-1.90)
0.03
0.02
(0.56)
(2.80)
0.12
0.003
(1.43)
(0.29)
-1.84
-0.87
(-8.36)
(-29.23)
Notes: The t-values of the parameter estimates are reported in parenthesis. Germany
is taken as the base country so the respective dummy is omitted. Refinancing cost are
defined as the spread between the average allotment rate paid by an individual bank and
the marginal repo rate of the auction.
29