online appendix of the importance of being special: repo ...jfe.rochester.edu ›...
TRANSCRIPT
-
Online Appendix of”The Importance of Being Special:Repo Markets During the Crisis” ∗
Stefano CorradinEuropean Central Bank
Angela MaddaloniEuropean Central Bank
October 6, 2019
∗Email contacts: [email protected] and [email protected]. This pa-per should be not reported as representing the views of the European Central Bank (ECB), The viewsexpressed herein are those of authors only and do not represent the position of the European Central Bankor the Eurosystem.
1
[email protected]@ecb.europa.eu
-
Contents
1 Data and summary evidence 3
2 Model 4
2.1 Population measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 Panel regressions 7
4 Panel VAR 10
4.1 Alternative ordering: SMP purchase ordered last . . . . . . . . . . . . . . . . 10
4.2 Without 10−year on-the-run bonds . . . . . . . . . . . . . . . . . . . . . . . 174.3 CCP margin (instead of CDS bond) . . . . . . . . . . . . . . . . . . . . . . . 24
2
-
1. Data and summary evidence
Table 1 Statistics on repo transactions for repo maturity - This table reports theaverage and the standard deviation of the daily distribution of the special repo transactionsfor repo maturity. The statistics are reported for the full sample and for three distinct sub-periods. First period from 1 October 2009 to 7 August 2011. Second period: from 8 August2011 to 21 December 2011. Third period: from 22 December 2011 to 12 July 2012.
Spot next Tomorrow next Over nightMean St.dev Mean St.dev Mean St.dev
Full Sample 86.315 3.154 12.047 2.908 1.638 0.8201st period 86.431 3.238 11.868 2.978 1.699 0.8442nd period 84.869 2.425 13.620 2.285 1.509 0.6223rd period 86.962 3.020 11.519 2.694 1.517 0.838
Table 2 Statistics of GC and Special repo rates - This table reports the average,the standard deviation, the 5th and 95th percentile of the daily distribution of the GC andspecial repo rates. Special and GC repo rates are based on spot next (SN), tomorrow next(TN) and overnight (ON) transactions.The statistics are reported for the full sample andfor three distinct sub-periods. First period from 1 October 2009 to 7 August 2011. Secondperiod: from 8 August 2011 to 21 December 2011. Third period: from 22 December 2011 to12 July 2012 (see Section 3 for details about the variables).
GC repo rates Special repo ratesMean St.dev 5 Perc. 95 Perc. Mean St.dev 5 Perc. 95 Perc.
Full Sample 0.618 0.377 0.246 1.590 0.491 0.390 -0.326 1.4091st period 0.593 0.347 0.254 1.622 0.496 0.382 -0.231 1.4622nd period 1.141 0.197 0.765 1.602 0.873 0.331 -0.372 1.3103rd period 0.333 0.096 0.229 0.733 0.212 0.150 -0.396 0.397
3
-
2. Model
2.1. Population measures
To solve the system, we reduce it to a simpler one in six unknowns. Adding
Sellers νiµboµli = κµsi + κµsi + λµbiµsi (1)
Non-searchers λµbiµsi = κµni + κµni (2)
we find
νiµboµli + κ̄ (−µni − µsi) + κ (−µni − µsi) = 0. (3)
Using µsni = µsi + µni we find
µsni =νiµboµliκ̄+ κ
. (4)
Then, equation
µli + µcbhi
+ µsi = S (5)
implies
µli = −µcbhi + S − µsi. (6)
Plugging equation (6) into
Borrowers F +2∑
i=1
κ(µsi + µni) = κµbo +2∑
i=1
νiµboµli (7)
we find
F =κµbo
(−ν1µcbh1 − ν2µ
cbh2
+ κ̄+ κ+ ν2S − ν2µs2 + ν1S − ν1µs1)
κ̄+ κ. (8)
Equation
Sellers νiµboµli = κµsi + κµsi + λµbiµsi (9)
implies
µsi =νiµbo
(−µcb
hi+ S − µsi
)κ̄+ κ+ λµbi
. (10)
Equation
Non-searchers λµbiµsi = κµni + κµni (11)
implies
µni =λνiµbiµbo
(−µcb
h1+ S − µsi
)(κ̄+ κ) (κ̄+ κ+ λµbi)
. (12)
Equation
Buyers κµni = κµbi + λµbiµsi (13)
4
-
implies
µbi =λνiκµbiµbo
(−µcb
hi+ S − µsi
)(κ̄+ κ) (κ̄+ κ+ λµbi) (κ̄+ λµsi)
. (14)
Combining these equations to compute µsi we find
µsi = −νiµbo
(−µcb
hi+ S − µsi
)κ̄+ κ+ λµbi
− µcbsi + µsi. (15)
Using equation
Sellers κµli + κµsi = λµbiµsi (16)
we find
κ̄(−µcb
hi+ S − µsi
)+νiκ̄µbo
(−µcb
hi+ S − µsi
)κ̄+ κ+ λµbi
− λµbi
(−νiµbo
(−µcb
h1+ S − µs1
)κ̄+ κ+ λµbi
− µcbsi + µsi
)= 0
and we solve for µbo
µbo = −(κ̄+ κ+ λµbi)
(λµbiµ
cbsi − κ̄µhi + Sκ̄− κ̄µsi − λµbiµsi
)νi (κ̄+ λµbi)
(−µcb
hi+ S − µsi
) . (17)Simplifying equation (15) we find
µsi =κ̄(−µcb
hi− µcbsi + S
)κ̄+ λµbi
. (18)
Equation
Central bank seller κcbµhi = λµbiµcbsi (19)
implies
µcbsi =κcbµhiλµbi
. (20)
Using equation (9) to compute µbo we find
µbo =F (κ̄+ κ)
κ (−νiµhi − ν2µh2 + κ̄+ κ+ ν2S − ν2µs2 + ν1S − ν1µs1). (21)
We use
Buyers F = κµb +2∑
i=1
λµsiµb (22)
to compute µb using µsi = µsi + µcbsi + µsi and equations (18), (20) and (10).
5
-
We use
Central bank buyer Fcb
= κcbµcbb
+2∑
i=1
λµsiµcbb
(23)
to compute µcbb
using µsi = µsi + µcbsi + µsi and equations (18), (20) and (10).
The reduced system consists of
µsi = µsi + µcbsi + µsi
µbi = µbi + µcbbi
+ µbi
λµsiµcbb
= κcbµcbhi,
where the latter equation corresponds to the central bank holder transition equation. These
are six equations (because one is for each asset), and the six unknowns are{µsi, µbi, µ
cbhi
}i∈{1,2}.
The system is numerically solved using Mathematica.
6
-
3. Panel regressions
Table 3 Panel Stationary Test - This table reports the results of the Fisher stationarytest whose null hypothesis is that all the panels contain a unit root for the SMP purchase,repo imbalance, specialness, cash imbalance, bid-ask spread, available for lending variablesand CDS bond (log) over the sample 8 August 2011 - 21 December 2011.
SMP purchase Repo imbalance SpecialnessStatistic P-value Statistic P-value Statistic P-value
Inverse chi-squared 393.26 0.00 305.43 0.00 187.92 0.00Inverse normal -15.36 0.00 -13.01 0.00 -8.30 0.00Inverse logit t -19.43 0.00 -15.02 0.00 -8.74 0.00Modified inv. chi-squared 29.74 0.00 21.86 0.00 11.30 0.00
Cash imbalance Bid-ask spread LendingStatistic P-value Statistic P-value Statistic P-value
Inverse chi-squared 70.26 0.00 300.23 0.00 305.89 0.00Inverse normal -4.42 0.00 -13.07 0.00 -12.77 0.00Inverse logit t -4.83 0.00 -15.07 0.00 -15.05 0.00Modified inv. chi-squared 6.67 0.00 21.93 0.00 21.90 0.00
CDS bond (log)Statistic P-value
Inverse chi-squared 136.140 0.00Inverse normal -6.23 0.00Inverse logit t -6.15 0.00Modified inv. chi-squared 6.65 0.00
7
-
Tab
le4Correlationmatrix
-T
his
table
rep
orts
the
corr
elat
ion
mat
rix
ofal
lth
eva
riab
les
use
din
the
anal
ysi
s.T
he
stat
isti
csar
ere
por
ted
for
the
full
sam
ple
and
the
seco
nd
per
iod.
Sp
ecia
lnes
san
dre
po
imbal
ance
are
bas
edon
spot
nex
t(S
N)
and
tom
orro
wnex
t(T
N)
tran
sact
ions.
The
seco
nd
per
iod:
from
8A
ugu
st20
11to
21D
ecem
ber
2011
.
Sp
ecia
l.R
epo
Imbal
ance
Bid
-ask
Len
d.
SM
PF
ail-
to-
CD
Sim
bal
ance
cash
spre
adpurc
has
edel
iver
bon
d(L
og)
Full
sam
ple
Sp
ecia
lnes
s1.
000
Rep
oim
b.
0.09
11.
000
Cas
him
b.
-0.0
22-0
.011
1.00
0B
id-a
sk-0
.012
-0.0
080.
027
1.00
0L
endin
g-0
.138
0.04
40.
013
-0.0
401.
000
SM
Ppurc
h.
0.06
90.
004
0.01
4-0
.006
-0.0
301.
000
Fai
l-to
-del
iver
0.43
40.
122
-0.0
34-0
.017
-0.0
730.
005
1.00
0C
DS
bon
d(l
og)
0.19
8-0
.068
0.00
8-0
.017
-0.1
270.
108
0.04
91.
000
2nd
per
iod
Sp
ecia
lnes
s1.
000
Rep
oim
b.
0.15
51.
000
Cas
him
b.
-0.0
15-0
.054
1.00
0B
id-a
sk-0
.023
-0.0
060.
010
1.00
0L
endin
g-0
.181
-0.0
070.
024
-0.0
331.
000
SM
Ppurc
h.
0.05
00.
048
0.04
0-0
.023
-0.0
321.
000
Fai
l-to
-del
iver
0.34
60.
169
-0.0
61-0
.010
-0.0
510.
016
1.00
0C
DS
bon
d(l
og)
0.22
0-0
.074
0.00
20.
005
-0.1
84-0
.013
0.00
61.
000
8
-
Table 5 Specialness and SMP purchases with CCP margin instead of CDS bond- The table shows the results of OLS (Columns (1)− (2)) and quantile (Columns (3)− (4))panel regressions with bond fixed effects. Specialness is expressed in basis points and is basedon spot next (SN) and tomorrow next (TN) transactions. The repo imbalance is based onspot next (SN) and tomorrow next (TN) transactions. The repo imbalance, cash imbalance,available for lending and SMP purchase variables are rescaled by the nominal outstandingamount of the bond, expressed in percentage terms and are lagged by one day (see Section3.1 for details). Bond time-to-maturity denotes the time-to-maturity of a specific security.Bid-ask spread denotes the bid-ask spread based on BGN Bloomberg prices at or before5pm. D. Auction is a dummy variable to control for the date of a primary issuance of thesecurity. CCP margins variable is bond-specific and lagged by one day. In Column (2) theobservations on the 10− year on-the-run bonds are excluded from the sample. Standarderrors are in parenthesis and are clustered by bond identifier (Columns (1) − (2))). Theresults are reported for the sub-period 8 August 2011 - 21 December 2011. Stars denotestatistical significance at 10% (∗), 5% (∗∗) and 1% (∗ ∗ ∗).
(1) (2) (3) (4)OLS OLS Q-70 Q-90
Specialness (lag) 0.802*** 0.800***(0.090) (0.099)
Repo imbalance 1.133** 0.569 0.954*** 4.007***(0.525) (0.347) (0.257) (0.688)
Cash imbalance -1.986*** -1.336*** -2.671*** -5.034*(0.738) (0.472) (1.004) (2.685)
Avail. lending -0.466** -0.477** -1.790*** -3.088***(0.202) (0.207) (0.185) (0.495)
Bond time-to-mat. -0.082 -0.077 -0.351*** 0.068(0.065) (0.067) (0.092) (0.246)
D. Auction -8.785*** -7.815*** -8.536 -18.271(2.677) (2.865) (6.724) (17.986)
Bid-ask spread -0.005 -0.009 -0.017 -0.136(0.024) (0.023) (0.107) (0.286)
CCP margin 45.536** 43.956** 203.309*** 173.414***(18.173) (18.953) (17.850) (47.743)
SMP purchase 4.614*** 5.169*** 5.060*** 15.610***(0.803) (1.576) (1.428) (3.819)
Constant 5.890** 5.869** 29.795*** 58.995***(2.645) (2.807) (1.091) (2.918)
R2 0.745 0.739Num. Obs. 4830 4742 4830 4830
9
-
4. Panel VAR
4.1. Alternative ordering: SMP purchase ordered last
Panel A - CDS bond
−.0
3−
.02
−.0
10
.01
Per
cent
age
poin
ts
0 5 10 15Day
CDS bond −> Cash imbalance
0.0
2.0
4.0
6.0
8P
erce
ntag
e po
ints
0 5 10 15Day
CDS bond −> Bid−ask spread
.02
.04
.06
.08
.1.1
2P
erce
ntag
e po
ints
0 5 10 15Day
CDS bond −> SMP purchase
−.1
0.1
.2P
erce
ntag
e po
ints
0 5 10 15Day
CDS bond −> Repo imbalance
−2
−1
01
2P
erce
ntag
e po
ints
0 5 10 15Day
CDS bond −> Specialness
−.0
2−
.01
0.0
1P
erce
ntag
e po
ints
0 5 10 15Day
CDS bond −> Sec. lending
Fig. 1. Panel VAR Cumulative IRFs - CDS bond The figure plots the cumulativeimpulse response functions to a CDS bond shock of 1% and 90% bootstrapped confidenceintervals (grey dashed lines) based on 1, 000 replications. The cumulative impulse responsefunctions are estimated from a panel VAR model in first-difference including the followingvariables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbalance, spe-cialness and available for lending of the bonds purchased under the SMP from 8 August 2011to 21 December 2011.
10
-
Panel B - Cash imbalance−
10
12
Bas
is p
oint
s
0 5 10 15Day
Cash imbalance −> CDS bond
−.0
2−
.01
0.0
1.0
2P
erce
ntag
e po
ints
0 5 10 15Day
Cash imbalance −> Bid ask spread−
.08
−.0
6−
.04
−.0
20
.02
Per
cent
age
poin
ts
0 5 10 15Day
Cash imbalance −> SMP purchase
−.6
−.4
−.2
Per
cent
age
poin
ts
0 5 10 15Day
Cash imbalance −> Repo imbalance
−4
−2
02
Per
cent
age
poin
ts
0 5 10 15Day
Cash imbalance −> Specialness
0.0
1.0
2.0
3.0
4P
erce
ntag
e po
ints
0 5 10 15Day
Cash imbalance −> Sec. lending
Fig. 2. Panel VAR Cumulative IRFs - Cash imbalance The figure plots the cumu-lative impulse response functions to a cash imbalance shock of 1% and 90% bootstrappedconfidence intervals (grey dashed lines) based on 1, 000 replications. The cumulative impulseresponse functions are estimated from a panel VAR model in first-difference including thefollowing variables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbal-ance, specialness and available for lending of the bonds purchased under the SMP from 8August 2011 to 21 December 2011.
11
-
Panel C - Bid-ask spread−
10
12
3B
asis
poi
nts
0 5 10 15Day
Bid−ask spread −> CDS bond
−.0
050
.005
.01
.015
Per
cent
age
poin
ts
0 5 10 15Day
Bid−ask spread −> Cash imbalance0
.02
.04
.06
.08
Per
cent
age
poin
ts
0 5 10 15Day
Bid−ask spread −> SMP purchase
−.1
−.0
50
.05
.1P
erce
ntag
e po
ints
0 5 10 15Day
Bid−ask spread −> Repo imbalance
−2
−1
01
Per
cent
age
poin
ts
0 5 10 15Day
Bid−ask spread −> Specialness
−.0
15−
.01
−.0
050
.005
.01
Per
cent
age
poin
ts
0 5 10 15Day
Bid−ask spread −> Sec. lending
Fig. 3. Panel VAR Cumulative IRFs - Bid-ask spread The figure plots the cumu-lative impulse response functions to a bid-ask spread shock of 1% and 90% bootstrappedconfidence intervals (grey dashed lines) based on 1, 000 replications. The cumulative impulseresponse functions are estimated from a panel VAR model in first-difference including thefollowing variables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbal-ance, specialness and available for lending of the bonds purchased under the SMP from 8August 2011 to 21 December 2011.
12
-
Panel D - SMP purchase−
10
12
3B
asis
poi
nts
0 5 10 15Day
SMP purchase −> CDS bond
−.0
2−
.015
−.0
1−
.005
0.0
05P
erge
ntag
e po
ints
0 5 10 15day
SMP purchase −> Cash imbalance−
.02
−.0
15−
.01
−.0
050
.005
Per
cent
age
poin
ts
0 5 10 15Day
SMP purchase −> Bid−ask spread
−.1
0.1
.2.3
Per
cent
age
poin
ts
0 5 10 15Day
SMP purchase −> Repo imbalance
01
23
45
Per
cent
age
poin
ts
0 5 10 15Day
SMP purchase −> Specialness
−.0
4−
.03
−.0
2−
.01
0P
erce
ntag
e po
ints
0 5 10 15Day
SMP purchase −> Sec. lending
Fig. 4. Panel VAR Cumulative IRFs - SMP purchase The figure plots the cumulativeimpulse response functions to a SMP purchase shock of 1% and 90% bootstrapped confidenceintervals (grey dashed lines) based on 1, 000 replications. The cumulative impulse responsefunctions are estimated from a panel VAR model in first-difference including the followingvariables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbalance, spe-cialness and available for lending of the bonds purchased under the SMP from 8 August 2011to 21 December 2011.
13
-
Panel E - Repo imbalance−
3−
2−
10
1B
asis
poi
nts
0 5 10 15Day
Repo imbalance −> CDS bond
−.0
4−
.02
0.0
2.0
4.0
6P
erce
ntag
e po
ints
0 5 10 15Day
Repo imbalance −> Cash imbalance−
.005
0.0
05.0
1.0
15.0
2P
erce
ntag
e po
ints
0 5 10 15Day
Repo imbalance −> Bid−ask spread
−.0
8−
.06
−.0
4−
.02
0.0
2P
erce
ntag
e po
ints
0 5 10 15Day
Repo imbalance −> SMP purchase
24
68
Per
cent
age
poin
ts
0 5 10 15Day
Repo imbalance −> Specialness
−.0
10
.01
.02
.03
Per
cent
age
poin
ts
0 5 10 15Day
Repo imbalance −> Sec. lending
Fig. 5. Panel VAR Cumulative IRFs - Repo imbalance The figure plots the cumu-lative impulse response functions to a repo imbalance shock of 1% and 90% bootstrappedconfidence intervals (grey dashed lines) based on 1, 000 replications. The cumulative impulseresponse functions are estimated from a panel VAR model in first-difference including thefollowing variables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbal-ance, specialness and available for lending of the bonds purchased under the SMP from 8August 2011 to 21 December 2011.
14
-
Panel F - Specialness0
12
34
Bas
is p
oint
s
0 5 10 15Day
Specialness −> CDS bond
0.0
2.0
4.0
6.0
8P
erce
ntag
e po
ints
0 5 10 15Day
Specialness −> Cash imbalance−
.02
−.0
10
.01
.02
Per
cent
age
poin
ts
0 5 10 15Day
Specialness −> Bid−ask spread
−.0
4−
.02
0.0
2.0
4.0
6P
erce
ntag
e po
ints
0 5 10 15Day
Specialness −> SMP purchase
−.3
−.2
−.1
0.1
Per
cent
age
poin
ts
0 5 10 15Day
Specialness −> Repo imbalance
−.0
3−
.02
−.0
10
.01
Per
cent
age
poin
ts
0 5 10 15Day
Specialness −> Sec. lending
Fig. 6. Panel VAR Cumulative IRFs - Specialness The figure plots the cumulativeimpulse response functions to a specialness shock of 1% and 90% bootstrapped confidenceintervals (grey dashed lines) based on 1, 000 replications. The cumulative impulse responsefunctions are estimated from a panel VAR model in first-difference including the followingvariables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbalance, spe-cialness and available for lending of the bonds purchased under the SMP from 8 August 2011to 21 December 2011.
15
-
Panel G - Available lending−
2−
10
1B
asis
poi
nts
0 5 10 15Day
Sec. lending −> CDS bond
−.0
4−
.03
−.0
2−
.01
0.0
1P
erce
ntag
e po
ints
0 5 10 15Day
Sec. lending −> Cash imbalance−
.01
0.0
1.0
2.0
3P
erce
ntag
e po
ints
0 5 10 15Day
Sec. lending −> Bid−ask spread
−.0
8−
.06
−.0
4−
.02
0.0
2P
erce
ntag
e po
ints
0 5 10 15Day
Sec. lending −> SMP purchase
−.1
0.1
.2.3
Per
cent
age
poin
ts
0 5 10 15Day
Sec. lending −> Repo imbalance
01
23
Per
gent
age
poin
ts
0 5 10 15day
Sec. lending −> Specialness
Fig. 7. Panel VAR Cumulative IRFs - Available lending The figure plots the cumula-tive impulse response functions to a available for lending shock of 1% and 90% bootstrappedconfidence intervals (grey dashed lines) based on 1, 000 replications. The cumulative impulseresponse functions are estimated from a panel VAR model in first-difference including thefollowing variables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbal-ance, specialness and available for lending of the bonds purchased under the SMP from 8August 2011 to 21 December 2011.
16
-
4.2. Without 10-year on-the-run bonds
Panel A - CDS bond
−.0
2−
.01
0.0
1P
erce
ntag
e po
ints
0 5 10 15Day
CDS bond −> Cash imbalance
0.0
2.0
4.0
6.0
8P
erce
ntag
e po
ints
0 5 10 15Day
CDS bond −> Bid−ask spread
.02
.04
.06
.08
.1P
erce
ntag
e po
ints
0 5 10 15Day
CDS bond −> SMP purchase
−.0
50
.05
.1.1
5P
erce
ntag
e po
ints
0 5 10 15Day
CDS bond −> Repo imbalance
−1
01
2P
erce
ntag
e po
ints
0 5 10 15Day
CDS bond −> Specialness
−.0
2−
.01
0.0
1P
erce
ntag
e po
ints
0 5 10 15Day
CDS bond −> Sec. lending
Fig. 8. Panel VAR Cumulative IRFs - CDS bond The figure plots the cumulativeimpulse response functions to a CDS bond shock of 1% and 90% bootstrapped confidenceintervals (grey dashed lines) based on 1, 000 replications. The cumulative impulse responsefunctions are estimated from a panel VAR model in first-difference including the followingvariables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbalance, spe-cialness and available for lending of the bonds purchased under the SMP from 8 August 2011to 21 December 2011.
17
-
Panel B - Cash imbalance−
.50
.51
1.5
2B
asis
poi
nts
0 5 10 15Day
Cash imbalance −> CDS bond
−.0
10
.01
.02
.03
Per
cent
age
poin
ts
0 5 10 15Day
Cash imbalance −> Bid ask spread−
.02
0.0
2.0
4P
erce
ntag
e po
ints
0 5 10 15Day
Cash imbalance −> SMP purchase
−.1
−.0
50
.05
.1P
erce
ntag
e po
ints
0 5 10 15Day
Cash imbalance −> Repo imbalance
−1.
5−
1−
.50
.5P
erce
ntag
e po
ints
0 5 10 15Day
Cash imbalance −> Specialness
−.0
2−
.01
0.0
1.0
2P
erce
ntag
e po
ints
0 5 10 15Day
Cash imbalance −> Sec. lending
Fig. 9. Panel VAR Cumulative IRFs - Cash imbalance The figure plots the cumu-lative impulse response functions to a cash imbalance shock of 1% and 90% bootstrappedconfidence intervals (grey dashed lines) based on 1, 000 replications. The cumulative impulseresponse functions are estimated from a panel VAR model in first-difference including thefollowing variables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbal-ance, specialness and available for lending of the bonds purchased under the SMP from 8August 2011 to 21 December 2011.
18
-
Panel C - Bid-ask spread0
12
3B
asis
poi
nts
0 5 10 15Day
Bid−ask spread −> CDS bond
−.0
050
.005
.01
Per
cent
age
poin
ts
0 5 10 15Day
Bid−ask spread −> Cash imbalance0
.02
.04
.06
Per
cent
age
poin
ts
0 5 10 15Day
Bid−ask spread −> SMP purchase
−.0
50
.05
.1P
erce
ntag
e po
ints
0 5 10 15Day
Bid−ask spread −> Repo imbalance
−2
−1
01
Per
cent
age
poin
ts
0 5 10 15Day
Bid−ask spread −> Specialness
−.0
2−
.01
0.0
1P
erce
ntag
e po
ints
0 5 10 15Day
Bid−ask spread −> Sec. lending
Fig. 10. Panel VAR Cumulative IRFs - Bid-ask spread The figure plots the cumu-lative impulse response functions to a bid-ask spread shock of 1% and 90% bootstrappedconfidence intervals (grey dashed lines) based on 1, 000 replications. The cumulative impulseresponse functions are estimated from a panel VAR model in first-difference including thefollowing variables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbal-ance, specialness and available for lending of the bonds purchased under the SMP from 8August 2011 to 21 December 2011.
19
-
Panel D - SMP purchase−
10
12
34
Bas
is p
oint
s
0 5 10 15Day
SMP purchase −> CDS bond
−.0
1−
.005
0.0
05.0
1P
erge
ntag
e po
ints
0 5 10 15day
SMP purchase −> Cash imbalance−
.03
−.0
2−
.01
0.0
1P
erce
ntag
e po
ints
0 5 10 15Day
SMP purchase −> Bid−ask spread
−.0
50
.05
.1.1
5P
erce
ntag
e po
ints
0 5 10 15Day
SMP purchase −> Repo imbalance
02
46
Per
cent
age
poin
ts
0 5 10 15Day
SMP purchase −> Specialness
−.0
4−
.03
−.0
2−
.01
0.0
1P
erce
ntag
e po
ints
0 5 10 15Day
SMP purchase −> Sec. lending
Fig. 11. Panel VAR Cumulative IRFs - SMP purchase The figure plots the cumu-lative impulse response functions to a SMP purchase shock of 1% and 90% bootstrappedconfidence intervals (grey dashed lines) based on 1, 000 replications. The cumulative impulseresponse functions are estimated from a panel VAR model in first-difference including thefollowing variables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbal-ance, specialness and available for lending of the bonds purchased under the SMP from 8August 2011 to 21 December 2011.
20
-
Panel E - Repo imbalance−
3−
2−
10
1B
asis
poi
nts
0 5 10 15Day
Repo imbalance −> CDS bond
−.0
1−
.005
0.0
05P
erce
ntag
e po
ints
0 5 10 15Day
Repo imbalance −> Cash imbalance−
.01
0.0
1.0
2.0
3P
erce
ntag
e po
ints
0 5 10 15Day
Repo imbalance −> Bid−ask spread
−.0
3−
.02
−.0
10
.01
Per
cent
age
poin
ts
0 5 10 15Day
Repo imbalance −> SMP purchase
12
34
5P
erce
ntag
e po
ints
0 5 10 15Day
Repo imbalance −> Specialness
0.0
05.0
1.0
15.0
2.0
25P
erce
ntag
e po
ints
0 5 10 15Day
Repo imbalance −> Sec. lending
Fig. 12. Panel VAR Cumulative IRFs - Repo imbalance The figure plots the cumu-lative impulse response functions to a repo imbalance shock of 1% and 90% bootstrappedconfidence intervals (grey dashed lines) based on 1, 000 replications. The cumulative impulseresponse functions are estimated from a panel VAR model in first-difference including thefollowing variables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbal-ance, specialness and available for lending of the bonds purchased under the SMP from 8August 2011 to 21 December 2011.
21
-
Panel F - Specialness0
12
34
5P
erce
ntag
e po
ints
0 5 10 15Day
Specialness −> CDS bond
−.0
050
.005
.01
Per
cent
age
poin
ts
0 5 10 15Day
Specialness −> Cash imbalance−
.02
−.0
10
.01
.02
.03
Per
cent
age
poin
ts
0 5 10 15Day
Specialness −> Bid−ask spread
−.0
20
.02
.04
.06
Per
cent
age
poin
ts
0 5 10 15Day
Specialness −> SMP purchase
−.1
5−
.1−
.05
0.0
5P
erce
ntag
e po
ints
0 5 10 15Day
Specialness −> Repo imbalance
−.0
3−
.02
−.0
10
.01
Per
cent
age
poin
ts
0 5 10 15Day
Specialness −> Sec. lending
Fig. 13. Panel VAR Cumulative IRFs - Specialness The figure plots the cumulativeimpulse response functions to a specialness shock of 1% and 90% bootstrapped confidenceintervals (grey dashed lines) based on 1, 000 replications. The cumulative impulse responsefunctions are estimated from a panel VAR model in first-difference including the followingvariables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbalance, spe-cialness and available for lending of the bonds purchased under the SMP from 8 August 2011to 21 December 2011.
22
-
Panel G - Available lending−
3−
2−
10
1P
erce
ntag
e po
ints
0 5 10 15Day
Sec. lending −> CDS bond
0.0
02.0
04.0
06.0
08.0
1P
erce
ntag
e po
ints
0 5 10 15Day
Sec. lending −> Cash imbalance−
.01
0.0
1.0
2.0
3P
erce
ntag
e po
ints
0 5 10 15Day
Sec. lending −> Bid−ask spread
−.0
8−
.06
−.0
4−
.02
0P
erce
ntag
e po
ints
0 5 10 15Day
Sec. lending −> SMP purchase
−.1
−.0
50
.05
Per
cent
age
poin
ts
0 5 10 15Day
Sec. lending −> Repo imbalance
0.5
11.
52
Per
gent
age
poin
ts
0 5 10 15day
Sec. lending −> Specialness
Fig. 14. Panel VAR Cumulative IRFs - Available lending The figure plots the cumula-tive impulse response functions to a available for lending shock of 1% and 90% bootstrappedconfidence intervals (grey dashed lines) based on 1, 000 replications. The cumulative impulseresponse functions are estimated from a panel VAR model in first-difference including thefollowing variables: CDS bond, cash imbalance, bid-ask spread, SMP purchase, repo imbal-ance, specialness and available for lending of the bonds purchased under the SMP from 8August 2011 to 21 December 2011.
23
-
4.3. CCP margin (instead of CDS bond)
Panel A - CCP margin
−.0
050
.005
.01
.015
Per
gent
age
poin
ts
0 5 10 15day
CCP margins −> Cash Imbalance
−.0
6−
.04
−.0
20
.02
Per
gent
age
poin
ts
0 5 10 15day
CCP margins −> Bid−ask spread
−.0
4−
.02
0.0
2.0
4P
erge
ntag
e po
ints
0 5 10 15day
CCP margins −> SMP purchase
−.1
5−
.1−
.05
0.0
5P
erge
ntag
e po
ints
0 5 10 15day
CCP margins −> Repo imbalance
−1
01
2P
erge
ntag
e po
ints
0 5 10 15day
CCP margins −> Specialness
−.0
2−
.01
0.0
1P
erge
ntag
e po
ints
0 5 10 15day
CCP margins −> Sec. lending
Fig. 15. Panel VAR Cumulative IRFs - CCP margin The figure plots the cumulativeimpulse response functions to a CCP margin shock of 1% and 90% bootstrapped confidenceintervals (grey dashed lines) based on 1, 000 replications. The cumulative impulse responsefunctions are estimated from a panel VAR model in first-difference including the followingvariables: CCP bond, cash imbalance, bid-ask spread, SMP purchase, repo imbalance, spe-cialness and available for lending of the bonds purchased under the SMP from 8 August 2011to 21 December 2011.
24
-
Panel B - Cash imbalance−
.000
4−
.000
3−
.000
2−
.000
10
.000
1P
erge
ntag
e po
ints
0 5 10 15day
Cash Imbalance −> CCP margins
−.0
2−
.01
0.0
1.0
2P
erge
ntag
e po
ints
0 5 10 15day
Cash Imbalance −> Bid ask spread−
.08
−.0
6−
.04
−.0
20
.02
Per
gent
age
poin
ts
0 5 10 15day
Cash imbalance −> SMP purchase
−.6
−.4
−.2
0P
erge
ntag
e po
ints
0 5 10 15day
Cash imbalance −> Repo imbalance
−4
−2
02
Per
gent
age
poin
ts
0 5 10 15day
Cash imbalance −> Specialness
0.0
1.0
2.0
3.0
4P
erge
ntag
e po
ints
0 5 10 15day
Cash imbalance −> Sec. lending
Fig. 16. Panel VAR Cumulative IRFs - Cash imbalance The figure plots the cumu-lative impulse response functions to a cash imbalance shock of 1% and 90% bootstrappedconfidence intervals (grey dashed lines) based on 1, 000 replications. The cumulative impulseresponse functions are estimated from a panel VAR model in first-difference including thefollowing variables: CCP margin, cash imbalance, bid-ask spread, SMP purchase, repo im-balance, specialness and available for lending of the bonds purchased under the SMP from8 August 2011 to 21 December 2011.
25
-
Panel C - Bid-ask spread−
.000
20
.000
2.0
004
.000
6P
erge
ntag
e po
ints
0 5 10 15day
Bid−ask spread −> CCP margins
−.0
1−
.005
0.0
05.0
1P
erge
ntag
e po
ints
0 5 10 15day
Bid−ask spread −> Cash imbalance0
.02
.04
.06
.08
Per
gent
age
poin
ts
0 5 10 15day
Bid−ask spread −> SMP purchase
−.0
50
.05
.1P
erge
ntag
e po
ints
0 5 10 15day
Bid−ask spread −> Repo imbalance
−2
−1
01
Per
gent
age
poin
ts
0 5 10 15day
Bid−ask spread −> Specialness
−.0
15−
.01
−.0
050
.005
.01
Per
gent
age
poin
ts
0 5 10 15day
Bid−ask spread −> Sec. lending
Fig. 17. Panel VAR Cumulative IRFs - Bid-ask spread The figure plots the cumulativeimpulse response functions to a bid-ask spread shock of 1% and 90% bootstrapped confidenceintervals (grey dashed lines) based on 1, 000 replications. The cumulative impulse responsefunctions are estimated from a panel VAR model in first-difference including the followingvariables: CCP margin, cash imbalance, bid-ask spread, SMP purchase, repo imbalance,specialness and available for lending of the bonds purchased under the SMP from 8 August2011 to 21 December 2011.
26
-
Panel D - SMP purchase−
.000
10
.000
1.0
002
.000
3.0
004
Per
gent
age
poin
ts
0 5 10 15day
SMP purchase −> CCP margins
−.0
25−
.02
−.0
15−
.01
−.0
050
Per
gent
age
poin
ts
0 5 10 15day
SMP purchase −> Cash imbalance−
.02
−.0
10
.01
Per
gent
age
poin
ts
0 5 10 15day
SMP purchase −> Bid−ask spread
−.1
0.1
.2.3
Per
gent
age
poin
ts
0 5 10 15day
SMP purchase −> Repo imbalance
02
46
Per
gent
age
poin
ts
0 5 10 15day
SMP purchase −> Specialness
−.0
4−
.03
−.0
2−
.01
0.0
1P
erge
ntag
e po
ints
0 5 10 15day
SMP purchase −> Sec. lending
Fig. 18. Panel VAR Cumulative IRFs - SMP purchase The figure plots the cumulativeimpulse response functions to a SMP purchase shock of 1% and 90% bootstrapped confidenceintervals (grey dashed lines) based on 1, 000 replications. The cumulative impulse responsefunctions are estimated from a panel VAR model in first-difference including the followingvariables: CCP margin, cash imbalance, bid-ask spread, SMP purchase, repo imbalance,specialness and available for lending of the bonds purchased under the SMP from 8 August2011 to 21 December 2011.
27
-
Panel E - Repo imbalance−
.000
4−
.000
3−
.000
2−
.000
10
.000
1P
erge
ntag
e po
ints
0 5 10 15day
Repo imbalance −> CCP margins
−.0
4−
.02
0.0
2.0
4.0
6P
erge
ntag
e po
ints
0 5 10 15day
Repo imbalance −> Cash imbalance−
.005
0.0
05.0
1.0
15.0
2P
erge
ntag
e po
ints
0 5 10 15day
Repo imbalance −> Bid−ask spread
−.0
4−
.02
0.0
2.0
4P
erge
ntag
e po
ints
0 5 10 15day
Repo imbalance −> SMP purchase
24
68
Per
gent
age
poin
ts
0 5 10 15day
Repo imbalance −> Specialness
−.0
10
.01
.02
.03
Per
gent
age
poin
ts
0 5 10 15day
Repo imbalance −> Sec. lending
Fig. 19. Panel VAR Cumulative IRFs - Repo imbalance The figure plots the cumu-lative impulse response functions to a repo imbalance shock of 1% and 90% bootstrappedconfidence intervals (grey dashed lines) based on 1, 000 replications. The cumulative impulseresponse functions are estimated from a panel VAR model in first-difference including thefollowing variables: CCP margin, cash imbalance, bid-ask spread, SMP purchase, repo im-balance, specialness and available for lending of the bonds purchased under the SMP from8 August 2011 to 21 December 2011.
28
-
Panel F - Specialness−
.000
8−
.000
6−
.000
4−
.000
20
Per
gent
age
poin
ts
0 5 10 15day
Specialness −> CCP margins
0.0
2.0
4.0
6.0
8P
erge
ntag
e po
ints
0 5 10 15day
Specialness −> Cash imbalance−
.01
0.0
1.0
2.0
3P
erge
ntag
e po
ints
0 5 10 15day
Specialness −> Bid−ask spread
−.0
4−
.02
0.0
2.0
4P
erge
ntag
e po
ints
0 5 10 15day
Specialness −> SMP purchase
−.4
−.2
0.2
Per
gent
age
poin
ts
0 5 10 15day
Specialness −> Repo imbalance
−.0
3−
.02
−.0
10
.01
Per
gent
age
poin
ts
0 5 10 15day
Specialness −> Sec. lending
Fig. 20. Panel VAR Cumulative IRFs - Specialness The figure plots the cumulativeimpulse response functions to a specialness shock of 1% and 90% bootstrapped confidenceintervals (grey dashed lines) based on 1, 000 replications. The cumulative impulse responsefunctions are estimated from a panel VAR model in first-difference including the followingvariables: CCP margin, cash imbalance, bid-ask spread, SMP purchase, repo imbalance,specialness and available for lending of the bonds purchased under the SMP from 8 August2011 to 21 December 2011.
29
-
Panel G - Available lending−
.000
20
.000
2.0
004
.000
6P
erge
ntag
e po
ints
0 5 10 15day
Sec. lending −> CCP margins
−.0
4−
.03
−.0
2−
.01
0.0
1P
erge
ntag
e po
ints
0 5 10 15day
Sec. lending −> Cash imbalance−
.01
0.0
1.0
2P
erge
ntag
e po
ints
0 5 10 15day
Sec. lending −> Bid−ask spread
−.0
8−
.06
−.0
4−
.02
0P
erge
ntag
e po
ints
0 5 10 15day
Sec. lending −> SMP purchase
−.1
0.1
.2.3
Per
gent
age
poin
ts
0 5 10 15day
Sec. lending −> Repo imbalance
01
23
Per
gent
age
poin
ts
0 5 10 15day
Sec. lending −> Specialness
Fig. 21. Panel VAR Cumulative IRFs - Available lending The figure plots the cumula-tive impulse response functions to a available for lending shock of 1% and 90% bootstrappedconfidence intervals (grey dashed lines) based on 1, 000 replications. The cumulative impulseresponse functions are estimated from a panel VAR model in first-difference including thefollowing variables: CCP margin, cash imbalance, bid-ask spread, SMP purchase, repo im-balance, specialness and available for lending of the bonds purchased under the SMP from8 August 2011 to 21 December 2011.
30
Data and summary evidenceModelPopulation measures
Panel regressionsPanel VARAlternative ordering: SMP purchase ordered lastWithout 10-year on-the-run bondsCCP margin (instead of CDS bond)