Stephen Morris
Princeton University
Hyun Song Shin
Princeton University
January 25, 2014
Abstract
We explore a global game model of the impact of monetary policy shocks. Risk-neutral
asset managers interact with risk-averse households in a market with a risky bond and a
oating rate money market fund. Asset managers are averse to coming last in the ranking
of short-term performance. This friction injects a coordination element in asset managers
portfolio choice that leads to large jumps in risk premiums to small future anticipated
changes in central bank policy rates. The si!e of the asset management sector is the key
parameter determining the extent of market disruption to monetary policy shocks.
2
D /e py2 /12
'here y is the risy )ond holding o% the household, e is the endo'ment and is ris tolerance!
>e assume that the endo'ment e is large enough that the 8rst5order condition determines
the optimal port%olio! *rom the 8rst5order condition 'ith respect to y and summing across
households, the aggregate demand %or the risy )ond %or the household sector is
p F v
2
h
h
y
F v cy /22
'here c is the positive constant de8ned as c F
2
/
h
h
, and
h
h
is the aggregate ris
tolerance %or the household sector as a 'hole! *igure 2 sho's the determination o% the price p
5
"igure %. Market clearing of the long asset. The price of the long asset at date $ is p. Asset managers
hold A units and households hold S A units.
o% the risy )ond %rom maret clearing! 9sset managers hold A units o% the )ond, 'here A is
exogenous %or no'! Households hold the remainder S A! "he price p and the resulting ris
premium v/p clears the maret )y re'arding households %or )earing ris!
9lthough asset managers consume once at the terminal date /date 22, they su?er %rom last
place aversion!
1
>e assume that there is a penalty su?ered )y any asset manager 'hose port%olio
value is raned last at date 1! "he penalty is in the %orm o% a decline in the asset managerGs
%unds under management, interpreted as 'ithdra'als )y their customers!
;n particular, i% any asset manager is raned last /or eHual last2 at date 1, and proportion x
o% asset managers has a strictly higher port%olio values, then the asset managerGs %unds under
management declines )y a %actor o% x, 'here is a positive constant strictly )et'een 0 and
1! ;n other 'ords, i% the asset manager initially holds 1 dollar o% %unds under management, )ut
comes last, and proportion x o% %und managers has strictly higher port%olio value, then the asset
1
The term &last place aversion' is taken from (uell) et al. *%+$,- who have used the concept in the very
di.erent context of the welfare economics of social deprivation.
<
managerGs %unds under management shrins toE
1 x /&2
9t date 1, asset managers allocate their %unds under management )et'een the risy )ond
and the 6oating rate account! "he asset managers initially start 'ith a holding o% A units o%
the risy )ond! ;% they choose to hold the )ond, each unit o% the )ond yields expected payo?
o% v!
;% the asset manager decides to sell the risy )ond, their sell order is executed simultaneously
'ith the other asset managers 'ho have decided to sell! "he aggregate sale o% the risy security
is matched 'ith the competitive demand curve o% the household investors, and each seller is
matched 'ith household )uyers!
;% proportion x o% the asset managers decide to sell their risy )ond holding, the total supply
o% the risy )ond is xA, and each seller has eHual chance o% placed in the Hueue B0, xC %or order
executiion! "here%ore, i% x asset managers sell the risy )ond, the expected revenue %rom sale
o% one unit is
p
1
2
cx /42
*igure 2 depicts the expected revenue curve p
1
2
cx, 'hose slope is hal% o% the competitive
demand curve p cx!
3iven the last place aversion o% the asset managers, the expected payo? %rom holding the
risy )ond 'hen proportion x sell the risy )ond is
u/x2 F v /1 x2 /52
9lthough the asset manager is ris5neutral and has a long hori.on, the short5term %riction %rom
last place aversion generates element o% short5termism!
;% the asset manager sells the risy )ond at date 1, the proceeds o% the sale are put into the
6oating rate account, 'here it earns interest rate r! Hence, the expected payo? o% the asset
manager %rom selling the risy )ond 'hen proportion x sell is given )y
w/x2 F /1 D r2
p
1
2
cx
/<2
I
"igure ,. /ayo. functions from holding long-dated security and switching to oating rate
*igure & plots the payo?s %rom the t'o strategies as a %unction o% x, the proportion o% asset
managers 'ho sell! "he payo? di?erence u /x2 w/x2 is indicated )y the shaded region!
"he payo? %unctions u/x2 and w/x2 are linear in x, and the payo? di?erence u/x2 w/x2 is
monotonic in x! ;% either u/x2w/x2 is positive %or all x, or negative %or all x, then the pro)lem
is trivial as asset managers have dominant actions! "here%ore, in 'hat %ollo's, 'e %ocus on the
case 'hen u/x2 w/x2 crosses the hori.ontal axis at some point!
! Global Game
"he 6oating rate r ruling )et'een date 1 and date 2 is uncertain, )ut investors have good
in%ormation a)out it! 9t date 1, asset manager i o)serves signal
i
o% the true interest rate r
given )y
i
F r D s
i
/I2
'here s
i
is a uni%ormly distri)uted noise term, 'ith reali.ation in B, C %or small positive
constant ! "he noise terms {s
i
} are independent across asset managers! >e %urther assume
that the ex ante distri)ution o% r is uni%orm! "he assumption that r and the noise term s
i
are uni%ormly distri)uted is %or expositional simplicity only! "he solution to )e o)tained )elo'
1
holds under general conditions on the ex ante distri)ution o% r and the noise structure /Morris
and Shin /200&, section 222!
,ased on their respective signals, asset managers decide 'hether to hold the risy )ond or
sell it! Since asset managers are ris5neutral, it is 'ithout loss o% generality to consider the
)inary choice o% #hold$ or #sell$! 9 strategy %or an asset manager is a mappingE
i
{Hold, Sell} /12
9 collection o% strategies /one %or each asset manager2 is an equilibrium i% the action prescri)ed
)y iGs strategy maximi.es iGs expected payo? at every reali.ation o% signal
i
given othersG
strategies!
9s the 8rst step in the solution, consider s'itching strategies o% the %orm
Sell i% >
Hold i%
/02
%or some threshold value
such that every asset manager using the same s'itching strategy around
! >e appeal to the %ollo'ing result in glo)al games! -ecall that x is our notation %or the
proportion o% investors 'ho sell!
#emma 1 Suppose that investors follow the switching strategy around
"he ans'er to Huestion /A2 gives the cumulative distri)ution %unction o% x evaluated at z, 'hich
'e denote )y G/z|
2! "he
steps to ans'ering Huestion /A2 are illustrated in *igure 4!
>hen the true interest rate is r, the signals {
i
} are distri)uted uni%ormly over the interval
Br , r D C! ;nvestors 'ith signals
i
>
2
/102
>hen do 'e have x < zK "his happens 'hen r is lo' enough, so that the area under the
density to the right o%
2
F z /112
or
r
0
F
D 2z /122
10