scieee Science in your language
[en] (orig)

Do security prices rise or fall when margins are raised?

Abstract

When repo margins are raised by a central clearing counterparty (CCP), the impact on security prices depends on whether the market is `long' or `short'. As both the long and the short must post margins, the price impact depends on which side is more leveraged: traders long in the security or those short-selling it. If a raised margin forces more position unwind from the long than from the short, the price will go down to clear the market. However, if short positions are more hit then the long ones, a raised haircut leads to a higher security price!

Read accessible full text

Do security prices rise or fall when margins are raised?

Author: Bottazzi, Jean-Marc,Páscoa, Mário Rui,Ramírez, Guillermo
Publisher: Nova SBE
Year: 2017
Source: https://run.unl.pt/bitstream/10362/25470/1/WP616.pdf
Do secu i y p ices ise
o all when ma gins
a e aised?
Jean-Ma c Bo azzi
Má io R. Páscoa
Guille mo Ramí ez
Wo king Pape
# 616
2017
Do Secu i y P ices Rise o Fall When
Ma gins A e Raised?
By Jean-Ma c Bo azzi, M´
a io R. P´
ascoa and Guille mo Ram´
ı ez1
Pa is S. o Economics and Capula; Uni e si y o Su ey; No a S. o Business
and Economics
Abs ac
When epo ma gins a e aised by a cen al clea ing coun e pa y (CCP),
he impac on secu i y p ices depends on whe he he ma ke is ‘long’ o
‘sho ’. As bo h he long and he sho mus pos ma gins, he p ice impac
depends on which side is mo e le e aged: ade s long in he secu i y o
hose sho -selling i . I a aised ma gin o ces mo e posi ion unwind om
he long han om he sho , he p ice will go down o clea he ma ke .
Howe e , i sho posi ions a e mo e hi hen he long ones, a aised hai cu
leads o a highe secu i y p ice!
1We a e hank ul o commen s om he audience a he SAET 2017 con e ence (June
2017). G. Ram´ı ez was suppo ed by a doc o al g an gi en by FCT, Po ugal (g an
BD 74704/2010). P´ascoa and Ram´ı ez acknowledge suppo om p ojec PTDC/IIM-
ECO/5360/2012 (FCT, Po ugal).
1
1 INTRODUCTION 2
1 In oduc ion
Wi hou ehypo heca ion, aised ma gin ypically dec eases he p ice o he colla -
e al (houses p ices o mo gage, o he p ice o o he colla e al used o asse back
secu i ies). In ac , he only le e age p o ided is o he long side o he colla e al
ma ke , who is unding colla e al pu chases. Reduced le e age, e e y hing else
being equal, means less buying and hence a lowe colla e al p ice.
The si ua ion is less clea , howe e , in he p esence o ehypo heca ion, o
example in he secu i ies ma ke s. Secu i ies se e as colla e al when inancing
hei own pu chase and hen cash lende s manage o epledge o sho sell2 he
colla e al hey ha e accep ed. Tha is, in hese ma ke s, sel -colla e aliza ion com-
bined wi h colla e al euse allow o sho posi ions. Since le e age is p o ided o
bo h he long and he sho side, he p ice impac o a le e age educ ion can also
go ei he way. Wi h cen al coun e pa y clea ing houses (CCPs), bo h he long
side and he sho side need o pos ma gin and a e hus a ec ed by a ia ion o
hai cu 3: i is well documen ed how highe ola ili y (o jus highe sp eads on
yields o so e eign bonds ela i e o a AAA benchma k) make CCPs (and u u es
exchanges) inc ease he ini ial ma gins cha ged in epo ades o bo h sides o he
ma ke 4, agen s ha pledge secu i ies and agen s ha accep hese as colla e al
o cash loans. posi ions
I has been a gued, in mos o he li e a u e we e iew below, ha he impac
o highe hai cu s is always o lowe he p ice o he colla e al. I his we e ue his
would mean ha aising ma gin is always p ocyclical : i he p ice o he colla e al
2The coun e pa ies o hese sho s can also epledge he secu i ies (an indi ec ehypo he-
ca ion) and a new i e a ion s a s.
3S icly speaking, he hai cu indica es, in pe cen age e ms, how he loan alls below he
colla e al alue, whe eas he ini ial ma gin exp esses he colla e al as a pe cen age o he
loan (i is he e o e he in e se o one minus he hai cu ).
4I should be clea ha a simila analysis can be done eplacing secu i y wi h deli e able
u u es and hai cu wi h u u es ini ial ma gin.
1 INTRODUCTION 3
is al eady unde p essu e, a highe hai cu pushes i u he down in some icious
cycle ha in he wo s case can eed on i sel .
Howe e , o ake jus an example, i we obse e wha happened in he Eu opean
so e eign deb c isis, we see yields d opping a e some ma gins we e aised. This
is mo e no o ious o he 10 yea go e nmen bonds o he ollowing coun ies and
o he ollowing episodes: I eland, 12 and 26 No embe 2010 and again o he
six consecu i e ma gin inc eases om Ap il h ough June 2011; I aly, 9 No 2011
and July 2012; Po ugal, 11 May 2011 and 29 June 2011; Spain, 20 June 2012.
The e a e ne e heless also some down spi als in p ices5, bu his is ce ainly no
he mos equen ma ke eac ion a e ma gins a e aised. La e , in pa icula o
I ish and Spanish bonds, as he c isis s a ed o ade away, specula ion dec eased
and he all in hai cu s was accompanied by a all in yields.
The e may be o he ac o s behind such yield mo emen , bu i does sugges
ha he impac o aising ma gin is no always o push he p ice o he colla e al
down. Ou iew is ha in secu i ies ma ke s i depends on posi ioning. Fo his
eason, as posi ioning changes quickly o e ime, one can expec he impac o
ma gins o be mo e isible in sho e m ma ke eac ions a he han in long
e m ends. I is well-known ha a a ious imes specula i e posi ions we e
qui e signi ican . This aises he possibili y ha aising hai cu could a imes
ha e had mo e impac on he sho side. And a con ibu ing ac o o yields going
down is sho s being he le e aged ade s being o ced o dele e age when ma gin
is aised. In ma ke pa lance, a secu i y ma ke is long when he longs a e he mo e
le e aged agen s and sho when i is he sho s ha a e mo e le e aged ins ead.
We will gi e a mo e p ecise de ini ion o a e minology ha supe icially seems
o clash wi h ma ke clea ing (which equi es longs and sho s o be ma ched).
The emphasis is on whe e he le e age is when he hai cu ises. I he sho side
519 No embe 2010 and 25 Ma ch 2011 o I eland; h oughou Ap il 2011 and on 14 June
2011 o Po ugal
2 RELATION WITH THE LITERATURE 4
ge s o be educed mo e han he long side (which happens when ‘ he ma ke is
sho ’), hen he p ice o he secu i y going up, e e y hing else being equal, enables
ma ke s o clea .
2 Rela ion wi h he Li e a u e
Mos o he li e a u e emphasized he p ocyclical ole o epo hai cu s and he e-
sul ing p opaga ing down-spi al. Ad ian and Shin (2009) obse e a s ong posi i e
co ela ion be ween le e age and balance shee size o US inancial in e media ies,
in e ing om i a p ocyclical ole o le e age. A heo e ical wo k by B unne meie
and Pede sen (2009) no ices ha sho selling also equi es capi al in he o m
o ma gin and p o ides an in e es ing esul poin ing ou ha a ma gins spi al
a ises when long ma gins a e dec easing in p ices o sho ma gins a e inc easing
in p ices. I seems o be implici ha p ices migh inc ease wi h sho ma gins and
ha a coun e -cyclical pa e n could happen i he sho ma gins we e dec easing
in p ices, jus like he long ma gins.
The obse ed, no so in equen , escapes o wha jus be o e looked like an
implacable icious ci cle logic mo i a ed us. When, in esponse o a p ice d op
and ola ili y, ma gins a e inc eased and he p ice subsequen ly ises, is he aised
ma gin a con ibu ing ac o o his ise o does he ise in p ice happen in spi e
o a headwind ma gin e ec ? We sugges ha he ma gin e ec can some imes be
suppo i e: highe ma gin also can nudge p ices highe .
In he empi ical li e a u e, he analysis ha Go on and Me ick (2012) did
o he 2008 inancial c isis a gued ha he spi al in e ac ion o epo hai cu s and
p ices o s uc u ed secu i ies led o a un on epo and insol ency o he US banking
sys em. While ha p o-cyclicali y may ha e ac ually occu ed in wha was hen
a long ma ke o hose secu i ies, K ishnamu hy e al. (2014) downplayed i s

2 RELATION WITH THE LITERATURE 5
magni ude and impac on he whole inancial sys em. These di e en assessmen s
may be due, as In an e (2015) poin ed ou , o Go on and Me ick (2012) ha ing
ocused on bila e al ades, while in he la ge US i-pa y ma ke hai cu s may
ha e been changed much less. We a gue o a balanced iew: in a ma ke wi h
bo h longs and sho s, le e age by i sel does no ha e o be p ocyclical, i depends
whe e le e age is mo e p e alen .
In any case, he p esumed p ocyclical na u e o ma gins se by CCPs (mo e
sys ema ic han in o e - he-coun e (OTC) ma ke ) has e en led o he iew ha
CCPs migh be ac o o ins abili y. This iew was ne e heless me wi h some
scep icism by Eu opean policy-make s ha alue CCPs’ ole as a i ewall agains
he p opaga ion o de aul shocks and ind an inc easingly cen ally clea ed ma ke
easie o moni o (see Cons ˆancio (2012)).
In he case o he Eu opean so e eign deb c isis, a lo o posi ioning was done
by banks (hedges o illiquid posi ion) and hedge unds. We ind acking CCP
ma gin a good way o gauge e olu ion o ma gins ele an o hose agen s. No
jus because Banks end o use CCPs bu all ele an ma gins ha e be mo ing
b oadly oge he : (1) hedge unds use p ime b oke s who ac as clea e s o hem
(also cha ging ma gin ega dless o whe he he ade is sho o long) (2) Fu u es
con ac s (whe e he exchange asks o ini ial ma gin ega dless o posi ion di ec-
ion). Ou ocus on cen al clea ing was no chosen only o analy ical pu poses
bu also seems o i be e wi h wha happened in ha c isis we chose o illus a e
ou poin .
The e is an impo an li e a u e ha has s udied colla e al in he gene al equi-
lib ium (GE) heo y adi ion and his is he li e a u e ha is closes ela ed o ou
wo k. The pape s by Geanakoplos (1997) and Geanakoplos and Zame (1997) se
he s age: adically changing he way bo owing is ea ed in GE models, eplacing
abs ac nega i e asse posi ions (subjec o exogenous nai e cons ain s) by loans
2 RELATION WITH THE LITERATURE 6
secu ed by asse s. Subsequen ly, many con ibu ions explo ed implica ions o how
colla e al ma gins a e se (like absence o Ponzi schemes6, wel a e p ope ies7o
he ola ili y o colla e al p ices8) bu he mos ele an o ou heme a e he pa-
pe s on how le e age impac s asse p ices, s emming om Geanakoplos (2003).
In pa icula , he pape s by Geanakoplos (2010), Fos el and Geanakoplos (2012)
and Fos el and Geanakoplos (2015), con empla e a con inuum o ade s, wi h di -
e en belie s on binominal asse e u ns, and compa e unsecu ed bo owing wi h
bo owing agains a colla e al which is ei he a du able good o a inancial asse
ha canno be sho -sold.
One o he main insigh s o hese h ee seminal pape s is ha le e age makes
he p ice o he asse ha se es as colla e al ise. To s a , le e age equilib ium
p ices we e shown o be g ea e han no-le e age p ices. I was no iced ha p ices
inc ease wi h exogenous loan- o- alue a ios, bu he emphasis was hen placed
on he endogenous de e mina ion o le e age ( uling ou de aul on non- ecou se
loans). The endogenous dele e age always ein o ced a all in p ices caused by bad
news abou he wo se case payo .
We use Fos el and Geanakoplos’ (2012) and (2015) binomial model wi h a
con inuum o deg ees o agen s’ op imism on asse payo s. We ind i su p isingly
e sa ile and ap o illus a e ich and sub le phenomena o gene al ele ance. We
use i o show ha lowe ma gin migh no always aise asse p ices, i asse s can be
sho sold. The key di e ence - mo e impo an han ecou se - is i he colla e al
may be eused by he c edi o on whose hands i was pledged. In a nu shell: Asse
p ices inc ease wi h le e age in a long ma ke and dec ease in a sho ma ke . In
a long ma ke he op imis s a e le e aged, while i is he pessimis s who a e i i is
sho . To make he compa ison easie , ou examples a e as close as possible o he
6See A aujo, P´ascoa and To es-Ma inez (2002) and Kuble and Schmede s (2003).
7See A aujo, Kuble and Schomme (2012) and Geanakoplos and Zame (2014).
8See B umm, G ill, Kuble and Schmede s (2015).
2 RELATION WITH THE LITERATURE 7
amewo k used in hose pape s, in pa icula o he benchma k model in Fos el
and Geanakoplos (2012).
As in Fos el and Geanakoplos (2012), we con empla e a binomial economy ( he
ini ial node is ollowed by wo nodes) and wo asse s, one ha can be pledged as
colla e al and ano he one ha canno . The o me has s ochas ic e u ns and he
la e is iskless. We can in e p e he o me as a isky secu i y and he la e
as cash. Also as in Fos el and Geanakoplos (2012) and Fos el and Geanakoplos
(2015), he e is a con inuum o agen s, wi h he same ini ial holdings o asse s a
he i s da e bu di e ing in he weigh s a ached o he wo s a es o na u e in
hei linea u ili y unc ions.
We depa om Fos el and Geanakoplos (2012) and Fos el and Geanakoplos
(2015) in h ee aspec s. Fi s and o emos he colla e al can be eused (and sho
sold). Reuse o inancial colla e al has always been done bu had been igno ed in
he li e a u e o a long ime. Recen wo k includes he pape by Bo azzi, Luque
and P´ascoa (2012), add essing le e age and exis ence o euse equilib ia in OTC
ma ke s, and he wo king pape by B umm, G ill, Kuble and Schmede s (2017),
which s udies ola ili y and wel a e implica ions o euse.
Second, loans a e ecou se, as is ypically he case in epo (wi h an excep ion
allowed by he Fed du ing he 2008 c isis). De aul is a bank up cy igge 9and,
he e o e, no so common in epo ma ke s. Fo his eason, and also o simplici y
and ac abili y, we ocus on asse p ice luc ua ions and look o ull epaymen
ou comes10.
Thi d, agen s ha e endowmen s (o he iskless asse ) also a he wo nodes
9The colla e al alue can be below he p omised epu chase p ice and ye de aul migh
no occu . On he o he hand, epo colla e al is exemp by he US bank up cy code om
au oma ic s ay and he cash lende can liquida e he colla e al immedia ely, bu i he
liquida ion alue exceeds (o is below) he epu chase p ice, he di e ence is deemed p ope y
o he insol en es a e (o can be claimed agains ha es a e, espec i ely).
10 Wo k in p og ess by Bo azzi, P´ascoa and Rami ez (2017) discusses how bank up cy can be
modelled bu he e we lea e ha in ica e non-con ex p oblem aside.
2 RELATION WITH THE LITERATURE 8
o he second da e. Such endowmen s played no ole in ensu ing sol ency in he
non- ecou se case, bu we can now choose hem high enough so ha he sho s
can con empla e maximum le e age o a gi en le el o hai cu , wi hou going
bank up a he second da e.
Ou model is used o cons uc wo benchma k cases o a epo ma ke clea ed
h ough a CCP. One whe e he ma ke is sho and aising hai cu s/ini ial ma gin
leads o highe secu i y p ices (as in he so e eign deb s episodes men ioned abo e).
Ano he , whe e he ma ke is long and aising he hai cu makes ha he secu i y
p ice go down. Wha de e mines whe he he ma ke will be long o sho is how
secu i y e u ns a y ac oss nodes by compa ison wi h endowmen s o he iskless
asse . The la e migh be oo low (in he s a e whe e he secu i y has highe
e u n) o allow o he sho s o le e up o wha he hai cu allows o .
The pape shows how ehypo hecac ion allows o coun e -examples o well-
known esul s ha had been es ablished in he absense o such euse o he col-
la e al. In he nex sec ion we p o ide a e y b ie mo i a ion on he basis o
obse ed hai cu . In sec ion 4 we desc ibe he model o cen ally clea ed epo
and in he ollowing sec ions we p esen he examples, discussing in mo e de ail a
sho ma ke whe e p ices ise in esponse o highe ma gin. Tha being said we
canno claim ha hai cu s ha e a s abilizing e ec by hemsel es. Fi s he e a e
many ma ke s whe e le e age is only o e ed o he long side. Also when sho s a e
possible, le e aged posi ioning could s abilise he p ice when ma gin is aised only
i he ma ke would be well equipped, in he sense o ha ing plen y o le e aged
sho s willing o ake p o i in any excessi e down p ice mo e.
4 THE MODEL 15
asse Xand secu i y Y, espec i ely, and a he same ime can pledge ψiuni s
o he secu i y in epo (in exchange o ob aining a loan wi h he same alue
qψi) o accep θiuni s o he secu i y as colla e al (in exchange o gi ing a loan
wi h he same alue qθi). We assumed epo con ac s a e se led a he secu i y
ma u i y da e12 . In his simple case, in he second leg, he loan is epaid (a he
p ede e mined epu chase p ice) bu wha is ac ually gi en back o he bo owe
is jus he asse payo s.
A non-nega i i y cons ain applies o X, epo long and epo sho posi ions:
(2) xi, θi, ψi≥0
Secu i y posi ions a e bounded by he in e ac ion wi h epo posi ions. Wha can
be pledged as colla e al canno exceed he secu i y long posi ion. So a , his is
would be a adi ional colla e al cons ain . Bu we go beyond, as he colla e al
can be eused by he c edi o : can be pledged in ano he epo ade o can be
sold. Selling he colla e al ha is empo a ily in he c edi o ’s hands is wha a
sho sale is. Naked sho sales, un ela ed o he amoun o he secu i y ha was
bo owed (accep ed as colla e al), a e no allowed in eali y and a e uled ou in
ou model. Hence, secu i y posi ion can be nega i e yibu sho sales a e bounded
by he ne epo posi ion. The box cons ain cap u es he wo-sided in e ac ion
be ween secu i y and epo posi ions13:
(3) yi+θi−ψi≥0
In o he wo ds, he ne secu i y i le balance held by each agen (in each agen ’s
12 also known as epo o ma u i y...al e na i ely, a 3 da es a ian o he model could be wo ked
ou easily.
13 See Bo azzi, Luque and P´ascoa (2012) on his cons ain and also Du ie (1996) o p e-
cu so s o his cons ain .

4 THE MODEL 16
“box”) mus be non nega i e. I he euse o he colla e al we e no allowed (as
in mo gages), his cons ain would collapse o a plain colla e al cons ain o he
o m yi−ψi≥0.
Repo ades a e cen ally clea ed by an exchange (also known as cen al clea -
ing coun e pa y, CCP), ha cha ges a ma gin o bo h sides o he ma ke , in a
p opo ion 1 −ho he alue o he epo ade14. Then, agen ibudge cons ain
a da e 0 is as ollows:
(xi−1) + q(yi−1) + q(θi−ψi) + (1 −h)q(θi+ψi)≤0(4)
O equi alen ly,
(xi−1) + q(yi−1) + q(2 −h)θi−qhψi≤0(5)
A he second da e, agen s se le hei epo obliga ions, using hei endowmen s
and he e u ns om hei i s da e secu i y (long o sho ) posi ions. Repo loans
a e epaid a he epo a e ρ. The CCP gi es back o he agen s he ma gins i
had collec ed, acc ued o in e es , a he same a e, he epo a e
The budge cons ain o agen iin s a e sis as ollows:
Ci
s=ωs+xi+yiDs+ (1 + ρ)q(θi−ψi) + (1 + ρ)q(1 −h)(θi+ψi)≥0(6)
Tha is,
Ci
s=ωs+xi+yiDs+ (1 + ρ)q[(2 −h)θi−hψi]≥0(7)
A plan (xi, yi, θi, ψi) is said o be easible o agen ia p ices (q, ρ) i i sa is ies
14 See Bo azzi, Luque and P´ascoa (2012) on he OTC case and how hai cu s a e di e en in
ha case.
4 THE MODEL 17
he budge cons ain s (5) and (7), he box cons ain (3) and he non-nega i i y
es ic ions (2). A plan ha maximizes (1) among all plans ha a e easible o i
a (q, ρ) is said o op imal o igi en (q, ρ).
4.2 Equilib ium
The CCP collec s ma gins om bo h sides o he epo ma ke and is supposed o
pay back hese ma gins, a he epo in e es a e. To accomplish his, i should
in es he ma gins and he e is no be e way han o in es also in epo, ha is,
o be also a epo long (accep he secu i y as colla e al and p o ide a loan, on i s
own), wi h a posi ion ΘEX ≥0.
The ma ke clea ing condi ion o he epo ma ke is:
Z1
0
θidi + ΘEX =Z1
0
ψidi,(8)
Wha he CCP in es s in epo is he ma gins i collec s om bo h sides o he
ma ke , ha is,
ΘEX = (1 −h)·Z1
0
θidi +Z1
0
ψidi
Hence, epo ma ke s clea i
(2 −h)Z1
0
θidi =hZ1
0
ψidi(9)
The ma ke clea ing condi ions o he wo asse ma ke s a e:
Z1
0
xidi = 1 and Z1
0
yidi = 1(10)
An equilib ium consis s in p ices (q, ρ) and an alloca ion o agen s’ plans (xi, yi, θi, ψi)i∈(0,1)
5 SHORT MARKETS VERSUS LONG MARKETS 18
such ha (i) each agen ’s plan is op imal o him a hese p ices and (ii) asse
and epo ma ke s clea 15.
5 Sho ma ke s e sus long ma ke s
5.1 Le e age
Le e age in secu i ies ma ke is usually measu ed by he a io o he alue o
posi ion o he down paymen needed o build ha posi ion. In he case o a
long posi ion, he pu chase o he secu i y can be inanced in he epo ma ke -
pledging he whole long posi ion and ob aining a cash loan wi h he same alue,
bu incu ing he cos o paying he ma gin o he CCP, ha is, spending jus a
ac ion 1 −ho he secu i y alue. Le e age is gi en by 1/(1 −h). Analogously,
in he case o a sho posi ion, i he agen is sho selling he whole colla e al ha
he has accep ed, i p esumes ha he ga e a cash loan o he same alue and paid
a ma gin o he CCP. Again, he ne cos is jus he ac ion 1 −ho he secu i y
alue being sho ed. So, le e age is once mo e gi en by 1/(1 −h). Howe e , in
bo h cases, his is wha le e age is when he agen has decided o le e up o ull
po en ial gi en he hai cu . Lowe le e age would esul i he long posi ion we e
no en i ely pledged o i he sho posi ion we e lowe han he whole colla e al
ha was accep ed.
Colla e al euse may play also a ole in allowing o he longs o le e up o ull
hai cu po en ial. The agg ega e ( ully) le e aged long posi ion, which is in e sely
p opo ional o he hai cu 1 −h, may exceed wha he agg ega e ini ial holdings
we e o his secu i y. Such ou come can only sa is y ma ke clea ing i sho sales
a e done in equilib ium. Howe e , how much le e age can be achie ed depends
15 Second da e ma ke clea ing ollows om asse ma ke clea ing and he ul illmen o second
da e budge cons ain s as equali ies (due o mono onici y in p e e ences).
5 SHORT MARKETS VERSUS LONG MARKETS 19
bo h on how weal hy agen s a e and how low he hai cu is.
In equilib ium and o he ma ke a la ge we will say ha he ma ke is long i
ne ade o longs is mo e sensi i e o le e age han sho s’. The ma ke is sho i
highe sensi i i y is wi h sho s. In he examples ha we p esen in his pape we
exhibi equilib ia whe e he longs o he sho s a e le e ed up o ull po en ial bu
no bo h a he same ime, hence p o iding clea examples o ‘long’ and ‘sho ’
ma ke s. The side hi ing ha ull le e age cons ain - only one side does- will
ell us whe he he ma ke is long o sho , espec i ely. In bo h cases, being
ully le e aged means ha he secu i y posi ion (long o sho , espec i ely) can
be buil o he mos ha he ma gin downpaymen allows o (by combining he
i s da e budge and box binding equali ies)16. Second da e endowmen s should
no be an obs acle o such le e age. Tha is, in he case o a sho ma ke , s a e D
endowmen s should be high enough (o ma gins no oo low) o allow o he sho s
o ully le e while s ill ha ing non-nega i e income (no becoming insol en ) a
ha s a e, whe e secu i y e u ns ou weigh epo epaymen s. Simila ly, in he case
o a long ma ke , s a e Uendowmen s should be high enough o allow o he longs
o ully le e wi hou becoming insol en in he s a e whe e secu i y e u ns all
below he epo epaymen ha is due.
Finally, we no e ha in bo h examples, he colla e al ge s eused. In he
‘sho ’ case his is ob ious bu we obse e also ha in he ‘long’ case sho sales
a e always done in equilib ium. The longs’ weal h ( hei ini ial holdings o he wo
asse s, e alua ed a equilib ium p ices) is high enough (by compa ison wi h wha
he hai cu is), so ha agg ega e long posi ions exceed he gi en agg ega e ini ial
holdings.
We will show nex ha , in ou economy, whe he he ma ke is long o sho is
wha de e mines how inc eases in le e age ( educ ions in ma gin) a ec he p ice
16 Mo eo e , he e can no be an equilib ium whe e bo h he longs and he sho s a e ully
le e aged, as shown in Appendix A.4.
5 SHORT MARKETS VERSUS LONG MARKETS 20
o he secu i y.
We also iden i y condi ions ha gi e ise o sho and long ma ke s. Sho
ma ke s a e likely o occu when he endowmen in he s a e ha sho selle s
alue less (ωDin his economy) is high ela i e o he endowmen in he o he
s a e. When ωUis high enough ela i e o ωD, a long ma ke is mo e likely o
occu .
5.2 Le e age and p ices
Mo e o mally, le us ocus on equilib ia whe e all sho s ake he same sho
posi ion and all longs ake he same long posi ion, as will be he case in he
examples p esen ed below. We e e o such equilib ia as ypical equilib ia. The e
will be a ma ginal agen mwho is indi e en be ween buying o selling secu i y
Yand we deno e a ypical agen in he se (m, 1) by Hand a ypical agen in
he se (0, m) by L. The o me is a pessimis and will be a sho . The la e is
an op imis and will be a long. I will be shown om i s o de condi ions wi h
espec o y ha when minc eases qdec eases.
De ini ion 1. We say ha a ypical equilib ium (q.ρ, x, y, θ, ψ)is a sho (long)
ma ke equilib ium i he ne ade o a sho is mo e (less, espec i ely) elas ic
wi h espec o he loan- o- alue a io h han he ne ade o a long.
In o he wo ds, in a sho ma ke i is he sho s who a e mo e sensi i e o
hai cu s, o changes in le e age condi ions. They hey beha e as being he ones
ha a e mo e le e aged. The elas ici y o he ne ade o a sho wi h espec
o hmeasu es wha is he p opo ional change in he absolu e alue o he equi-
lib ium ne ade (yH−1) o a sho in esponse o some p opo ional change in
h, by allowing all equilib ium a iables o adjus o his change ( ha is, mu a is
mu andis). I is gi en by dln(1−yH)
dln h, as yH<0. This should no be con used wi h

5 SHORT MARKETS VERSUS LONG MARKETS 21
he pa ial elas ici y ∂ln(1−yH)
∂ln h, which assumes he secu i y p ice and he epo a e
o emain cons an (a ce e is pa ibus elas ici y). Simila ly, he elas ici y o he ne
ade o a long wi h espec o hmeasu es how sensi i e is he absolu e alue o
he ne ade o a long o changes in hand is gi en by dln(yL−1)
dln h, o yL>1.
Ou i s esul es ablishes ha he way how secu i y p ices espond o ma gins,
in ypical equilib ia, depend on whe he ma ke s a e long o sho .
P oposi ion 1. The secu i y p ice qinc eases wi h he loan- o- alue a io hin a
long ma ke bu dec eases wi h hin a sho ma ke .
P oo . Ma ke clea ing equi es myL+ (1 −m)yH= 1. As he loan- o- alue a io
hchanges, mhas o change so ha 1 emains a con ex combina ion o he new
posi ions yLand yH. Mo e o mally,
(11) m=1−yH
yL−yH=1
1 + yL−1
1−yH
We see ha mis inc easing wi h he a io 1−yH
yL−1o he absolu e alue o he
ne ade o a sho (1 −yH) o he ne ade o a long (yL−1). Hence, m
inc eases (and, he e o e, q alls) wi h hi and only i he loga i hm o his a io
inc eases wi h h, ha is, i and only i dln(1−yH)
dh >dln(yL−1)
dh , o equi alen ly, i
and only i dln(1−yH)
dln h>dln(yL−1)
dln h, as claimed. And qinc eases wi h hi and only i
dln(1−yH)
dln h<dln(yL−1)
dln h.
In he nex subsec ions we examine how sho o long ma ke s may a ise and
see how secu i y p ices ge a ec ed by ma gins in bo h cases.
5.3 Sho ma ke
We cons uc an equilib ium wi h a sho ma ke . This ype o equilib ium only
occu s when ωDis la ge enough so ha wha ends up cons aining he sho
5 SHORT MARKETS VERSUS LONG MARKETS 22
posi ion o agen s i∈(m, 1) is no he non-nega i i y o Ci
Dbu ins ead he
binding box cons ain . The sho s sell all hei endowmen s o asse s Xand Y o
build up he la ges possible sho posi ion in Y. We ind ha he posi ions o a
pessimis agen Has ollows:
yH=−θH=−1 + q
q·1
1−h
(12)
ψH=xH= 0(13)
CH
U, CH
D>0
Equa ion (12) shows ha when i is he box cons ain ha bounds sho -selle s’
posi ions, hei posi ions a e in e sely ela ed o he hai cu , ha is gi en by he
le e age 1/(1−h). We will see ha his is a sho ma ke as when ma gin is aised
sho s need o co e mo e posi ion han longs.
Deno e a ypical agen in he se (0, m) by L. This is a buye o bo h Xand
Y. F om ma ke clea ing in he secu i y we see ha he secu i y long posi ion o
his agen mus be he ollowing:
yL=1
m1 + (1 −m)1 + q
q·1
1−h
(14)
I he agen s ha a e long in Ya e pledging i (ψL>0) and no doing any e e se
epo (θL= 0), he epo ma ke clea ing condi ion (9) equi es
(15) ψL= (2−h
h)(1−m
m)1 + q
q(1 −h)
The box cons ain equi es yL−ψL≥0, which holds i and only i
(16) h/2≥(1 −m)1 + q
q
5 SHORT MARKETS VERSUS LONG MARKETS 23
The e will be a slack in he box o he non-le e ed agen Li (16) holds as a s ic
inequali y. The o he posi ions o agen Lsa is y
(17) CL
U= 0, CL
D>0, xL=1
m
Agen s in (0, m) a e mo ing all hei consump ion om s a e U o s a e D,
ha is, CL
U= 0.
To be su e ha hese plans a e op imal o each agen , we w i e agen i’s
op imiza ion p oblem in e ms o h ee choice a iables, yi,θiand ψi. We deno e
EiD=γiDU+ (1 −γi)DD, which is agen i’s expec ed paymen o secu i y Y. On
he igh o each es ic ion we w i e he co esponding Lag ange mul iplie .
max
Eiω+ (1 + q)+(EiD−q)yi+q(2 −h)ρθi−qhρψi
s. .
xi= (1 + q)−qyi−q(2 −h)θi+qhψi≥0 (µi
x)
yi+θi−ψi≥0 (µi)
Ci
s=ωs+ (1 + q)+(Ds−q)yi+q(2 −h)ρθi−qhρψi≥0 (λi
s)
θi≥0 (νi
θ), ψi≥0 (νi
ψ)
The i s o de condi ions (FOC) o he p oblem o agen i∈(0,1) a e:
yi:EiD−q=qµi
x−µi−X
s
λi
s(Ds−q)
θi:q(2 −h)ρ=q(2 −h)µi
x−µi−νi
θ−q(2 −h)ρX
s
λi
s
ψi:qhρ =qhµi
x−µi+νi
ψ−qhρ X
s
λi
s,
5 SHORT MARKETS VERSUS LONG MARKETS 24
No ice ha an agen may ha e bo h θiand ψiposi i e only i he mul iplie µi o
he box cons ain is ze o.
Gi en ha Xpays a null in e es a e, we look o equilib ia whe e he epo
a e is ze o as well17. Fo agen s i, wi h i > m, he i s o de condi ions a e
sa is ied o ρ= 0 wi h he ollowing mul iplie s:
µH
x=q−EHD
q(1 −h)>0
µH=q(2 −h)µH
x>0
νH
θ= 0
νH
ψ= 2q(1 −h)µH
x
λH
U=λH
D= 0
When i<m, he mul iplie s ha sa is y he i s o de condi ions o ρ= 0 a e:
µL
x=µL= 0
νL
θ=νL
ψ= 0
λL
U=ELD−q
q−DU
>0
λL
D= 0
Sho s ha e posi i e shadow alues o bo h he box cons ain (a possession alue
o he colla e al, due o he desi e o sho sell) and he non-nega i i y cons ain
on X(exp essing a wish o do a ”‘naked sho sale”’ in X). I he o me we e
posi i e and he la e ze o, he epo a e would ha e o be nega i e, below he
ze o in e es a e on X.
17 In Fos el and Geanakoplos (2012) he secu ed in e es a e was also ze o. I Xhad s ochas ic
e u ns, he epo a e could all in be ween he implied in e es a es on X, in e p e able as
in e es on ese es (IOR).
A APPENDIX 31
(a) yi(b) θiand ψi
Figu e 6: Equilib ium co esponding o hai cu o 1% (h= 0.99): m= 0.881,
q= 0.754, xi= 1/m o i≤mand xi= 0 o i>m(assuming ωD≥107.6).
(a) yi(b) θiand ψi
Figu e 7: Equilib ium co esponding o hai cu o 3% (h= 0.97): m= 0.801,
q= 0.789, xi= 1/m o i≤mand xi= 0 o i>m(assuming ωD≥34.7).
An agen abo e he ma ginal agen (i>m) is sho selling he whole colla e al
ha was pledged o him. An agen below he ma ginal agen (i < m) is long in
he secu i y bu pledging less han he long posi ion.
Ma ke clea ing equi es myL+(1−m)yH= 1. As he hai cu inc eases, mhas
o change so ha 1 emains a con ex combina ion o he new posi ions yLand yH.
Now, he g aphs on he le -hand-side o Figu es 6 and 7 show ha , as he hai cu

A APPENDIX 32
is aised, he ully le e aged (indi idual) sho posi ion |yH| alls much mo e han
he (indi idual) long posi ion yL. The sho posi ion yHchanges om −232.7 o
−75.6 (is 32% o wha i was be o e), while he long posi ion yLchanges om 32.7
o 20.0 (is s ill 61% o wha i was be o e). Bo h yHand yLgo close o 1, bu
he o me much mo e han he la e . The weigh s on he con ex combina ion
mus e lec ha : m alls and, he e o e, q=EmD ises.
We saw in he p oo o P oposi ion 1 ha mis inc easing wi h he a io 1−yH
yL−1
o he absolu e alue o he ne ade o a sho (1 −yH) o he ne ade o a
long (yL−1). In a sho ma ke , his a io alls as he hai cu is aised. Figu e
8 illus a es how mand q a y as his a io changes. In he i s g aph we see
ha his a io is he slope o a ay joining he o igin o he poin ep esen ing ne
ade posi ions o he wo ypical agen s. The second g aph akes an ampli ied
look a wha happens close o he o igin, whe e mcan be ound as he e ical
coo dina e o he poin whe e he ay c osses he simplex. Such minduces q=
DD−(DD−DU)m, ead on he ho izon al axis.
A APPENDIX 33
(a) Ne ades o longs and sho s (b) mand qde e mined om ne ades a io
Figu e 8
2) On o he epo posi ions o he non-le e ed agen s: as he non-le e ed longs
ha e null shadow alues o he box cons ain , he abo e secu i y ades could
be implemen ed wi h o he epo posi ions o he longs (and he CCP). Each long
could ake posi ions ˜
θLand ˜
ψLon bo h sides o he epo ma ke and ha e no slack
in he box (˜
θL=˜
ψL−yL). This is no he same as pledging he long posi ion in
exchange o he ne epo loan. Being on bo h sides o he ma ke en ails an ex a
ma gin, which will be gi en back by he CCP in he second leg. Tha is, he long
is a he same ime lending cash o he CCP (a he epo a e), he eby undoing
pa ially his secu ed bo owing. Wha ma ks he di e ence be ween le e ed and
non-le e ed agen s is no so much whe he he e is no slack in he box bu a he
how ied up is he secu i y posi ion (long o sho ) o he opposi e epo posi ion
(sho o long, espec i ely).
A APPENDIX 34
Such al e na i e epo ades would no change how q ela es o h. Ma ke
clea ing equi es bo h (2 −h)(1 −m)θH+ (2 −h)m˜
θL=hm ˜
ψLand myL=
1 + (1 −m)|yH|, whe e |yH|=θH. We ge ˜
ψL=2−h
1−h·1
2m. I is easy o see ha
˜
θL≥0 (and ˜
ψL≥yL≥ψL) i and only i (16) holds and, he e o e, equilib ium
p ices qa e s ill ela ed o hacco ding o he solid black cu e o Figu e 5.
3) On ne ing o clien s epo posi ions: we ha e assumed ha he epo sho
and long posi ions o each agen do no ge ne ed o he pu poses o assessing
ma ginable posi ions. Howe e , ou esul s a e obus o ull o pa ial ne ing18.
As shown in sec ion A.5, condi ions (16) and (19) s ill hold unde ull o pa ial
ne ing (and, he e o e, also Figu e 5). The only in e es ing di e ence is ha unde
ull ne ing, epo po olios o non-le e ed agen s will be uniquely de e mined, o
each hai cu : non-le e ed agen s can no longe ha e al e na i e epo po olios
wi h posi ions on bo h sides o he ma ke ( hose discussed in Commen 2)).
A.2 Long ma ke
A long ma ke is mo e likely o occu when ωUis la ge ela i e o ωD. The
equilib ium is again cha ac e ized by a ma ginal agen m∈(0,1). The se (m, 1)
o agen s has a ypical agen deno ed by Hand he se (0, m) has a ypical agen
deno ed by L. Now, he agen s wi h high alua ions ( he op imis s) o he expec ed
e u n o secu i y Yha e long posi ions on he secu i y and a e le e aged, leading
18 LCH.Clea ne does pa ial ne ing (see LCH.Clea ne SA (2011)): he ma ginable posi ion
is a epo long o sho posi ion dec emen ed by he ollowing: he minimum o he wo
mul iplied by a ac o signi ican ly less han one.
A APPENDIX 35
na u ally o a long ma ke as can be seen below.
yL=ψL=1 + q
q·1
1−h
(20)
θL=xL= 0(21)
CL
U, CL
D>0
while agen s wi h low alua ions o secu i y e u ns a e sho and no le e aged,
yH=1
1−m1−m·1 + q
q·1
1−h
(22)
θH=h
2−h
m
1−m
1 + q
q(1 −h)
(23)
ψH= 0(24)
xH=1
1−m
(25)
CH
U>0CH
D= 0
The box cons ain o he non-le e ed sho s holds i and only i yH+θH≥0, o
equi alen ly, o
(26) h/2≤1−m1 + q
q
The p oblem o he agen s is he same as in he sho ma ke example. Fo
he posi ions p esen ed abo e, he i s o de condi ions a e sa is ied when agen s’
posi i e mul iplie s a e he ollowing: µL
x=ELD−q
q(1−h),µL=qhµL
x,νL
θ= 2q(1 −h)µL
x
and λH
D=q−EHD
DD−q.
We will conside he case when he pa ame e s o he economy a e all as be o e
wi h only one excep ion: we in e change he alues o he endowmen s ωUand ωD.
Le DU= 0.70, DD= 1.15. We assume now
A APPENDIX 36
Assump ion 2:ωD= 0 and ωU≥(1 + DD)DD−(2−h)DU
DU(1−h)( o ensu e ha
CL
U≥0).
Agen s’ belie s a e again gi en by γi=i. Now, CH
D= 0 implies ha le e age
should be such ha
(27) 1
1−h=(1 −m)(1 + q)
DD−q+ 1q
(1 + q)m
whe e m=DD−q
DD−DU.
In Figu e 9, condi ion (26) holds he egion abo e he g ay cu e. Combining
condi ions (27) and (26) we ge a lowe bound on hcompa ible wi h ψH≥0. Fo
habo e ha lowe bound (h≥0.96189), equa ion (27) desc ibes how qe ol es
wi h h. The black cu e in Figu e 9 illus a es how he secu i y p ice inc eases
wi h he loan- o- alue a io h(and, he e o e, wi h le e age) in he long ma ke 19.
This ela ionship is gene alized in he ollowing p oposi ion, p o en in sec ion A.3:
P oposi ion 3. I commodi y endowmen s sa is y Assump ion 2, hen he ma -
ke is long and he secu i y p ice inc eases as le e age goes up (and he e o e as
ma gins dec ease).
Obse e ha sho sales a e always done in equilib ium. In ac , yH<0 i and
only i m1+q
q
1
1−h≥1. By (27), we ge yH<0 i and only i (1−m)(1+q)
DD−q+ 1 ≥1,
which holds i ially.
19 As in he sho ma ke examples, we could ha e allowed now each non-le e ed sho o be
jus p edominan ly epo long, a he han exclusi ely epo long, wi h no slack in he box,
bu s ill wi h a (sho ) secu i y posi ion less ied up he (long) epo posi ion, han is he
case o he long coun e pa ies.

A APPENDIX 37
Figu e 9: q s. h- Long ma ke
A.3 P oo o P oposi ions 2 and 3
We show ha unde Assump ions 1 o 2, he secu i y p ice dec eases o inc eases,
espec i ely, wi h h, which also implies ha he ma ke is sho o long, espec-
i ely, by he a gumen in he p oo o P oposi ion 1.
Fo gi en a alue o h, he equilib ium p ice o he secu i y mus sa is y he
ollowing nonlinea sys em o wo equa ions:
H≡1
1−h=m(1 + q)
q−DU
−1q
(1 + q)(1 −m)
(19)
q=EmD(18)
F om equa ion (19) we see ha i we keep qcons an and inc ease m, le e age (H)
A APPENDIX 38
inc eases:
∂H
∂m =1 + q
q−DUq
(1 + q)(1 −m)+m(1 + q)
q−DU
−1q
(1 + q)(1 −m)2
=q
(1 + q)(1 −m)2(q−DU)[(1 −m)(1 + q) + m(1 + q)−q+DU]
=q(1 + DU)
(1 + q)(1 −m)2(q−DU)>0
whe e he las inequali y ollows om q > DU, due o (18). Now le ¯q= 1/q and
no e ha :
∂H
∂¯q=1
(1 −m)(1 + ¯q)2(1 −¯qDU)2(1 −¯qDU)2+m(1 + ¯q)2DU>0(28)
This implies ha ∂H
∂q <0 and ha along any isole e age cu e (ob ained by ixing
a alue o Hin (19)) we ha e:
dq
dm =−
∂H
∂m
∂H
∂q
>0
F om ∂H
∂m >0 we ge ha , o a ixed alue o q, inc easing mimplies mo ing o
a isole e age cu e co esponding o a highe alue o H. The new secu i y p ice
ha sol es he wo-equa ion sys em is hen ound a he in e sec ion o his cu e
and he s aigh line gi en by (18) (q=DD+m(Du−DD)). This implies ha as
le e age Hinc eases, he p ice o he secu i y qmus dec ease as shown in Figu e
10.
A APPENDIX 39
Figu e 10: Impac o le e age on equilib ium when he ma ke is sho
The e is ano he condi ion ha mus hold in equilib ium. I is condi ion (16),
ensu ing ha he box cons ain s is sa is ied o he non-le e ed longs. Elimina ing
hin he sys em o med by equa ion (18) and he equali y e sion o (16), we ge
a quad a ic equa ion in q, o each m, whose wo oo s a e desc ibed by cu es c1
and c2in Figu e 10. The g ey egion in Figu e 10 is he se o poin s (m, q) whe e
q∈[DU, DD] and (16) holds. Equilib ium pai s (m, q) a e ound wi hin his egion
whe e each isole e age ( o some h) c osses he dashed line ep esen ing (18). The
isole e ages po ayed in he igu e a e o he hai cu s o 1% and 3%, as in he
nume ical example o sec ion 5.3 (and o i s pa ame e alues o Dand ω).
An analogous a gumen p o es ha , in a long ma ke , p ices inc ease wi h
inc eases in le e age, as in P oposi ion 3. To see his simply no e ha now equi-
lib ium p ices mus sa is y bo h (18) and
(27) 1
1−h=(1 −m)(1 + q)
DD−q+ 1q
(1 + q)m
A APPENDIX 40
De ining ¯m= 1 −mwe ha e:
∂H
∂¯m=q(1 + DD)
(1 + q)(1 −¯m)2(DD−q)>0
Which in u n implies ha now ∂H
∂m <0. Le ing again ¯q= 1/q we ha e:
∂H
∂¯q=−1
m(1 + ¯q)2(DD¯q−1)2(DD¯q−1)2+ (1 −m)(1 + ¯q)2DD<0(29)
This implies ha ∂H
∂q >0 and, he e o e, along each isole e age we ha e dq
dm >0.
Now highe le e age implies mo ing o a isole e age o he le . This implies a
highe p ice q o he secu i y and a lowe alue o m. This is illus a ed in Figu e
11.
Figu e 11: Impac o le e age on equilib ium when he ma ke is long
Equilib ium equi es in addi ion ha (26) should hold in a long ma ke . This
ensu es ha he box cons ain o he non-le e ed sho s is sa is ied. Fo q∈
[DU, DD], such condi ion holds in he g ey egion o Figu e 11 (cons uc ed om