49
NBER WORK[NG PAPE} SERIES RATIONAL SPEaJLTIVE 3JBBLES IN AN E0(CHNGE RATE TA3ET ZONE Wiflem H. Beiter Paolo A. Pesenti Workim Paper No. 3467 NATIONAL JREPU OF EOJNCMIC RESEARUI 1050 Massachusetts Avenue Cambridge, NA 02138 October 1990 We would lBce to thank Torsten Persson, thristoçtier Sims, the participants at the Warwick Sumser Research Workshop on 'Exchare Rates ax Financial Markets", July 9-27, 1990, in particular Penzo Avesani, Giuseppe Bertola, Ardy Caplin, Bernard tmas, Marcus Miller, Rick van der Ploeg, Paul Weller ani the participants at the International Finance Workshop at Yale, in particular Giancarlo Corsetti, Vittorio Grilli, Ecidii Hamada and Nourie]. Roubini. Paolo Pesenti would like to ackelge financial support from the Mate per gli Studi Monetari, Bancari e Finanziari L. Einaudi, Rome. This paper is part of NBER' s research prcgram in International Stislies. ny opinions eressi are those of the authors and not those of the National Bereau of Economic Research.

Rational Speculative Bubbles in an Exchange Rate Target Zone

Embed Size (px)

Citation preview

NB

ER

WO

RK

[NG

PA

PE}

SER

IES

RA

TIO

NA

L

SPE

aJL

TIV

E 3

JBB

LE

S IN

AN

E0(

CH

NG

E R

AT

E T

A3E

T Z

ON

E

Wif

lem

H.

Bei

ter

Paol

o A. Pesenti

Wor

kim

Paper No. 3467

NA

TIO

NA

L

JRE

PU O

F E

OJN

CM

IC R

ESE

AR

UI

1050

Mas

sach

uset

ts A

venu

e C

ambr

idge

, NA

02

138

Oct

ober

199

0

We

wou

ld l

Bce

to

than

k T

orst

en P

erss

on,

thri

stoç

tier

Sim

s,

the

part

icip

ants

at

the

War

wic

k Su

mse

r Res

earc

h W

orks

hop

on 'E

xcha

re R

ates

ax

Fina

ncia

l M

arke

ts",

Jul

y 9-

27,

1990

, in

par

ticul

ar P

enzo

Ave

sani

, G

iuse

ppe

Ber

tola

, A

rdy

Cap

lin,

Ber

nard

tmas

, M

arcu

s M

iller

, R

ick

van

der

Ploe

g,

Paul

Wel

ler

ani t

he p

artic

ipan

ts a

t the

Inte

rnat

iona

l Fi

nanc

e W

orks

hop

at Y

ale,

in

pa

rtic

ular

Gia

ncar

lo C

orse

tti, V

ittor

io G

rilli

, E

cidi

i Ham

ada

and

Nou

rie]

. R

oubi

ni.

Paol

o Pe

sent

i w

ould

lik

e to

ack

elge

fina

ncia

l su

ppor

t fr

om t

he

Mat

e pe

r gl

i St

udi

Mon

etar

i, B

anca

ri e

Fin

anzi

ari

L.

Ein

audi

, Rom

e.

Thi

s pa

per

is p

art o

f N

BE

R' s

res

earc

h pr

cgra

m i

n In

tern

atio

nal

Stis

lies.

ny

op

inio

ns e

ress

i are

thos

e of

the

auth

ors

and

not

thos

e of

the

Nat

iona

l B

erea

u of

Eco

nom

ic R

esea

rch.

NB

ER

Wor

king

Pa

per

#346

7 O

ctob

er 1

990

RA

TIO

NA

L S

PEJI

AT

IVE

JB

BL

ES

IN A

N E

XQ

1PN

GE

R

AT

E T

AJE

I Z

ON

E

The

rec

ent

theo

ry o

f ex

chan

ge

rate

dyn

amic

s w

ithin

a t

arge

t zon

e ho

lds

that

exc

iiari

je r

ates

und

er a

cur

renc

y ba

rd a

re l

ess

resp

ansi

ve to

furd

asen

tal

shoc

ks t

han

exch

ange

rate

s un

der

a fr

ee f

loat

, pr

ovid

ed t

hat

the

inte

rven

tion

rule

s of

the

Cen

tral

Ban

k(s)

are

cat

uwn

kncM

ledg

e.

The

se r

esul

ts a

re d

eriv

ed

afte

r hav

ing

assu

med

a p

rior

i th

at e

xces

s vo

latil

ity du

e to

ratio

nal b

ubbl

es

does

not

occ

ur in

the

for

eign

exc

hang

e m

arke

t. In

this

pape

r we

cons

ider

inst

ead

a se

tup

in w

hich

the

exis

tenc

e of

spe

cula

tive

beha

vior

is a

dat

um t

he

Cen

tral

Ban

k ha

s to

dea

l w

ith.

We

show

tha

t th

e de

fens

e of

the

targ

et z

one

in t

he p

rese

nce

of b

ubbl

es i

s vi

able

if

the

Cen

tral

Ban

k ac

com

imxl

ates

spec

ulat

ive

atta

cks w

hen

the

latte

r ar

e co

nsis

tent

with

the

sur

viva

l of

the

targ

et z

one

itsel

f an

d ex

pect

atio

ns a

re s

elf-

fulf

illir

. T

hese

reB

ults

hol

d

for

a la

rge

clas

s of

exn

geno

us a

nd f

unda

men

tal-

depe

nden

t bub

ble

proc

esse

s.

We

show

tha

t th

e in

stan

tane

ous v

olat

ility

of

exch

ange

ra

tes

with

in a

bar

d is

no

t ne

cess

arily

les

s th

an th

e vo

latil

ity un

der f

ree

floa

t an

d an

alyz

e th

e

ixpl

icat

ions

for

inte

rest

rat

e di

ffer

entia

l dy

nam

ics.

Will

em F

l. B

aite

r Pa

olo

A.

Pese

nti

Ipar

toen

t of

Eco

nom

ics

Cep

arbn

ent o

f E

cono

mic

s Y

ale

Uni

vers

ity

Yal

e U

nive

rsity

19

72 Y

ale

Stat

ion

1972

Yal

e St

atio

n N

ew H

aven

, C

I 06

520—

1972

N

ew H

aven

, cr

065

20—

1972

I. I

ntro

duct

ion

The

rec

ent

liter

atur

e on

exc

hang

e ra

te t

arge

t zo

nes

insp

ired

by

Kru

gman

's

sem

inal

ar

ticle

on

the

subj

ect

(Kru

gman

[199

0])

repr

esen

ts

a su

cces

sful

mar

riag

e of

polic

y re

leva

nce a

nd t

echn

ical

inno

vatio

n. F

orm

al t

arge

t zon

es l

ike

the

exch

ange

rat

e

arra

ngem

ents

of th

e E

urop

ean M

onet

ary

Syst

em a

nd i

nfor

mal

tar

get z

ones

bet

wee

n th

e

U.S

. do

llar,

the

Japa

nese

yen

and

the

D—

Mar

k si

nce

the

mid

dle o

f th

e la

st d

ecad

e w

ere

a pr

omin

ent f

eatu

re o

f the

exc

hang

e ra

te s

yste

m o

f the

Eig

htie

s and

pro

mis

e to

be

so

for t

he N

inet

ies.

T

he t

echn

ical

mod

elin

g in

nova

tion

cons

ists

in

the

appb

catio

n of th

e

theo

ry o

f re

gula

ted

Bro

wni

an m

otio

n (s

ee e

.g.

Mal

liari

s an

d B

rock

[19

82],

Har

riso

n

[198

5], D

ixit

[198

8], K

arat

zas

and

Shre

ve [

1988

], D

umas

[1

989]

and

Flo

od a

nd G

arbe

r

[198

9]) to

the

stud

y of

the

beha

vior

of

a fl

oatin

g ex

chan

ge

rate

that

is c

onst

rain

ed

by

appr

opri

ate i

nter

vent

ions

not

to s

tray

out

side

som

e gi

ven

rang

e or t

arge

t zon

e.

The

lite

ratu

re o

n th

is s

ubje

ct i

s gr

owin

g ve

ry f

ast.

Am

ong

the

man

y re

cent

cont

ribu

tions

are

Kru

gman

[198

7], K

rugm

an

and

Rot

embe

rg [1

990]

, Mill

er a

nd W

elle

r

[198

8a,b

; 19

89;

1990

a,b,

c],

Mill

er a

nd S

uthe

rlan

d [1

990]

, Fro

ot a

nd O

bstf

eld

[198

9a,b

],

Ber

tola

an

d C

abal

lero

[1

989,

199

0],

Kle

in [

1989

], S

vens

son

[198

9; 1

990a

,b],

Pes

enti

[190

0], D

elga

do a

nd D

umas

[19

90],

Lew

is [1

990]

, A

vesa

ni [

1990

], Ic

hika

wa,

M

iller

and

Suth

erla

nd [1

990]

, and

Sm

ith a

nd S

penc

er [

1990

].

A c

omm

on

elem

ent

in a

ll of

the

se a

naly

ses

is t

hat

a "f

unda

men

tals

onl

y"

solu

tion

is c

hose

n fo

r th

e ex

chan

ge r

ate:

The

exc

hang

e ra

te, a

non—

pred

eter

min

ed s

tate

vari

able

, is e

xpre

ssed

as

a fu

nctio

n of

the

curr

ent a

nd e

xpec

ted

futu

re v

alue

s of

the

fund

amen

tals

onl

y.

The

lite

ratu

re i

n fa

ct

conc

entr

ates

on

mod

els

with

a

sing

le

fund

amen

tal.

Thi

s fu

ndam

enta

l is g

over

ned

by r

egul

ated

Bro

wni

an m

otio

n, th

at i

s by

unre

gula

ted

Bro

wni

an m

otio

n as

lon

g as

the

fund

amen

tal s

tays

with

in t

he lo

wer

and

uppe

r in

terv

entio

n th

resh

olds

and

by

(int

erm

itten

t) i

nter

vent

ions

tha

t ke

ep

the

fund

amen

tal

with

in t

hese

thr

esho

lds

whe

neve

r th

e un

regu

late

d B

row

nian

m

otio

n

thre

aten

s to

tak

e th

e fu

ndam

enta

l ou

tsid

e th

e in

terv

entio

n lim

its (

and

thus

the

—2—

exch

ange

ra

te o

utsi

de i

ts t

arge

t zo

ne).

T

his

perm

its t

he c

urre

nt

valu

e of

the

spot

exch

ange

ra

te to

be

expr

esse

d as

a n

on—

linea

r fun

ctio

n of

the

cur

rent

val

ue o

f th

e

fund

amen

tal o

nly.

The

firs

t ste

p in

this

arg

umen

t, th

e de

cisi

on to

sol

ve f

or th

e cu

rren

t val

ue o

f th

e

exch

ange

ra

te a

s a

func

tion

only

of

the

cur

rent

an

d ex

pect

ed fu

ture

val

ues

of t

he

fund

amen

tal,

mea

ns th

at s

pecu

lativ

e bu

bble

s ar

e ru

led o

ut.

In d

ynam

ic

linea

r ra

tiona

l exp

ecta

tions

mod

els

the

argu

men

ts

for

rulin

g ou

t

spec

ulat

ive

bubb

les

are

wel

l—kn

own,

if n

ot n

eces

sari

ly w

holly

con

vinc

ing.

In

man

y of

the

mos

t po

pula

r m

odel

s th

e bu

bble

pr

oces

ses

are

non—

stat

iona

ry.'

The

se

non—

stat

iona

ry

bubb

le p

roce

sses

th

en

lead

to

non

—st

atio

nary

be

havi

or o

f st

ate

vari

able

s su

ch a

s as

set

pric

es.

Unb

ound

ed g

row

th o

r de

clin

e in

the

se s

tate

vari

able

s for

cons

tant

val

ues o

f th

e fu

ndam

enta

ls is

arg

ued

to v

iola

te,

even

tual

ly, c

erta

in o

ften

onl

y

impl

icitl

y st

ated

feas

ibili

ty co

nstr

aint

s.2

Bla

ncha

rd [1

979]

ha

s de

velo

ped

an e

xam

ple

of a

non—

stat

iona

ry (

stoc

hast

ic o

r

dete

rmin

istic

) lin

ear b

ubbl

e th

at h

as a

fini

te ex

pect

ed l

ifet

ime.

T

he p

roba

bilit

y of

the

bubb

le su

rviv

ing

for

a gi

ven

peri

od o

f tim

e de

clin

es w

ith th

e le

ngth

of th

at p

erio

d at

a co

nsta

nt e

xpon

entia

l ra

te a

nd a

ppro

ache

s ze

ro a

sym

ptot

ical

ly

(see

als

o B

lanc

hard

and

Fisc

her

[198

9, c

hapt

er 5

]).

Alth

ough

th

e bu

bble

rem

ains

non

-sta

tiona

ry a

nd i

ts

expe

cted

val

ue g

row

s ex

pone

ntia

lly

with

out b

ound

, th

e fi

nite

exp

ecte

d lif

etim

e of

this

bubb

le m

ay m

ake

it le

ss v

ulne

rabl

e to

the

stan

dard

cri

tique

of n

on-s

tatio

nary

bub

bles

.

Mill

er a

nd W

elle

r [1

990a

[ an

d M

iller

, W

elle

r an

d W

illia

mso

n [1

989]

use

the

Bla

ncha

rd

bubb

le

to a

naly

ze e

xcha

nge

rate

beh

avio

r (b

oth

with

and

with

out a

tar

get

zone

) in

a

stoc

hast

ic ve

rsio

n of

the

Dor

nbus

ch o

vers

hoot

ing

mod

el (

Dor

nbus

ch

[197

6]).

Cer

tain

no

n—lin

ear m

odel

s al

so p

osse

ss n

on—

stat

iona

ry b

ubbl

e so

lutio

ns.

For

inst

ance

, the

def

latio

nary

an

d in

flat

iona

ry b

ubbl

es

stud

ied

by H

ahn

[198

2] a

nd b

y

Obs

tfel

d an

d R

ogof

f [1

983]

, al

thou

gh o

btai

ned

in n

on—

linea

r m

odel

s,

are

also

non—

stat

iona

ry.

In m

ore

gene

ral

non—

linea

r mod

els

it it

how

ever

qui

te c

omm

on t

o

—3—

find

stat

iona

ry b

ubbl

es.3

The

exc

hang

e ra

te t

arge

t zo

ne m

odel

pot

entia

lly p

rovi

des

a pa

rtic

ular

ly h

appy

non—

linea

r br

eedi

ng g

roun

d fo

r sp

ecul

ativ

e bu

bble

s.

Prov

ided

the

targ

et z

one

is

cred

ible

, th

at i

s pr

ovid

ed

ther

e is

cer

tain

ty th

at t

he i

nter

vent

ions

requ

ired

to

defe

nd

the

targ

et z

one

will

ta

ke

plac

e,

the

exch

ange

ra

te c

an n

eith

er ri

se n

or f

all

with

out

boun

d.

The

re a

re i

nfin

itely

man

y in

terv

entio

n ru

les

that

are

com

patib

le w

ith t

he

defe

nse

of a

giv

en t

arge

t zo

ne.

For

a ra

tiona

l sp

ecul

ativ

e bu

bble

to

exis

t, th

e

inte

rven

tion

rule

mus

t of c

ours

e be

con

sist

ent,

both

with

in th

e ta

rget

zon

e an

d at

the

boun

dari

es, w

ith th

e be

havi

or o

f th

e (r

egul

ated

) fun

dam

enta

ls.

Thi

s pa

per d

evel

ops

a

sim

ple

and

intu

itive

int

erve

ntio

n ru

le th

at e

ncom

pass

es a

s a

spec

ial

case

the

sol

utio

n

with

out

a bu

bble

, ye

t is

ge

nera

l en

ough

to

perm

it a

wid

e ra

nge

of e

xoge

nous

and

fund

amen

tal—

depe

nden

t de

term

inis

tic or

sto

chas

tic b

ubbl

es.

With

thi

s in

terv

entio

n

rule

the

argu

men

ts a

gain

st b

ubbl

es l

ose

thei

r bite

. A

non

—st

atio

nary

bub

ble

does

not

now

im

ply

non—

stat

iona

ry b

ehav

ior o

f th

e ex

chan

ge r

ate.

T

his

argu

men

t is

not

new

.

As

earl

y as

198

7 W

illia

m H

. B

rans

on,

in c

onve

rsat

ion

and

disc

ussi

on,

repe

ated

ly ra

ised

the

poss

ibili

ty th

at ta

rget

zon

es m

ight

lead

to "

inde

term

inac

y" of

the

exc

hang

e ra

te.

Thi

s pa

per c

an b

e vi

ewed

as

an a

naly

tical

conf

irm

atio

n of h

is c

onje

ctur

e.

In t

he n

ext

Sect

ions

we

expo

sit

the

sim

ples

t ve

rsio

n of

the

cred

ible

tar

get

zone

mod

el a

nd a

naly

ze its

beh

avio

r with

or

with

out s

pecu

lativ

e bu

bble

s.

II.

A T

arge

t Zon

e M

odel

With

or W

ithou

t Spe

cula

tive

Bub

bles

.

(a) T

he M

odel

.

Our

ana

lysi

s of

the

pla

ce a

nd r

ole

of s

pecu

lativ

e bu

bble

s in

an

exch

ange

rat

e

targ

et z

one

requ

ires

the

exch

ange

ra

te t

o be

a no

n—pr

edet

erm

ined

(fo

rwar

d—lo

okin

g)

—4—

,

stat

e va

riab

le,

but

does

not

m

ake

furt

her

dem

ands

on

de

scri

ptiv

e re

alis

m.

We

ther

efor

e fe

el

com

fort

able

in

usi

ng

Occ

am's

ra

zor

and

follo

win

g th

e bu

lk

of t

he

liter

atur

e on

tar

get

zone

s by

cas

ting

the

anal

ysis

in

ter

ms

of th

e si

mpl

est

poss

ible

linea

r in

tert

empo

ral s

ubst

itutio

n eq

uatio

n fo

r th

e ex

chan

ge

rate

and

a (

cont

inuo

us

time)

rand

om w

alk

with

dri

ft f

or th

e un

regu

late

d fun

dam

enta

l. (N

otab

le d

epar

ture

s

from

the

cult

of e

xtre

me

sim

plic

ity

are

Mill

er a

nd W

elle

r [1

988a

,b;

1989

; 19

90a,

b,c]

who

ana

lyze

stoc

hast

ic ve

rsio

ns o

f the

rich

er D

ornb

usch

ove

rsho

otin

g m

odel

.)

The

exc

hang

e ra

te p

roce

ss i

s gi

ven

in e

quat

ion

(1).

(1)

s(t)

dt =

f(t)

dt +

a'E

tds(

t)

a >

0.

Her

e s(

t)

is th

e na

tura

l lo

gari

thm

of t

he s

pot

nom

inal

exc

hang

e ra

te a

nd f(

t)

the

fund

amen

tal

dete

rmin

ant

of

the

exch

ange

ra

te.

Et

is

the

mat

hem

atic

al

expe

ctat

ion o

pera

tor c

ondi

tiona

l on

the

info

rmat

ion

avai

labl

e at t

ime

t to

the

priv

ate

sect

or a

nd th

e re

gula

tor.

Bot

h sO

) an

d f(

t)

are

assu

med

to b

e ob

serv

able

at ti

me

t,

i.e.

Ets

(t) =

sO)

and

Etf

(t) =

f(t)

, an

d th

e st

ruct

ure

of th

e m

odel

is

know

n.

For

the

purp

ose

of th

is p

aper

, the

econ

omic

int

erpr

etat

ion

of

f(t)

is

irr

elev

ant

exce

pt i

nsof

ar a

s it

affe

cts

the

cred

ibili

ty of

our

ass

umpt

ion t

hat,

with

eert

alnt

y, th

e

targ

et z

one

will

be

succ

essf

ully

def

ende

d.

A c

omm

on i

nter

pret

atio

n of

f is

in t

erm

s of

rela

tive

nom

inal

mon

ey s

tock

s m

inus

rela

tive

real

—in

com

e—re

late

d de

man

ds

for

real

mon

ey b

alan

ces,

i.e

.

* *

f=m

—m

k(y—

y )

whe

re

m i

s th

e lo

gari

thm

of th

e ho

me

coun

try n

omin

al m

oney

sto

ck,

y th

e lo

gari

thm

of h

ome

coun

try

real

out

put

and

k is

the

(co

mm

on)

inco

me

elas

ticity

of

mon

ey

dem

and.

St

arre

d va

riab

les

deno

te fo

reig

n qu

antit

ies,

In

thi

s ca

se

a de

note

s th

e

—5—

inve

rse o

f the

(com

mon

) in

tere

st s

emi—

elas

ticity

of m

oney

dem

and.

4

If th

e tw

o na

tiona

l rea

l out

puts

are

gov

erne

d by

exo

geno

us B

row

nian

mot

ion,

inte

rven

tion

mea

ns ch

ange

s in

the

rela

tive

mon

ey s

uppl

ies b

roug

ht a

bout

by

mon

etar

y

fina

ncin

g of

pub

lic

sect

or f

inan

cial

def

icits

, th

roug

h op

en m

arke

t op

erat

ions

or

by

unst

erili

zed

fore

ign

exch

ange

m

arke

t in

terv

entio

n.

As

show

n in

Bui

ter

[198

9] s

uch

inte

rven

tion

polic

ies

will

in

gene

ral

requ

ire

adju

stm

ents

to p

rim

ary

(non

—in

tere

st)

fisc

al

surp

luse

s to

be

sust

aina

ble

inde

fini

tely

fr

om a

tec

hnic

al p

oint

of

view

(i.e

. in

orde

r to

prev

ent e

ither

exh

aust

ion

of f

orei

gn e

xcha

nge

rese

rves

or u

nbou

nded

grow

th of

the

publ

ic d

ebt)

. A

pro

of th

at t

hey

are

polit

ical

ly f

easi

ble

and

cred

ible

is

beyo

nd t

he

scop

e of

this

pap

er.

In th

e ab

senc

e of

inte

rven

tion,

the

(unr

egul

ated

) fun

dam

enta

l f w

ould

fol

low

a

Bro

wni

an m

otio

n pr

oces

s w

ith d

rift

p an

d in

stan

tane

ous v

aria

nce c

.

di(t

) =

pdt

+ u1

dW1(

t)

clf>

0.

W1(

t)

is a

sta

ndar

d W

iene

r pr

oces

s (d

Wf

e 1 im

w

here

is

at

-to

inde

pend

ently

and

norm

ally

dist

ribu

ted

with

zer

o m

ean

and

unit

vari

ance

).

We

now

al

low

fo

r in

terv

entio

n at

the

upp

er a

nd l

ower

bo

unda

ries

of

the

exch

ange

rat

e ta

rget

zon

e. T

he re

gula

ted f

unda

men

tal

is g

over

ned

by

(2)

df(t

) =

pdt

+ yI

Wf(

t) —

dlu

+ d

lt

and I a

re t

he c

umul

ativ

e int

erve

ntio

ns

at th

e up

per

and

low

er b

ound

s of

the

exch

ange

rat

e ta

rget

zon

e, g

iven

by

5u an

d s1

resp

ectiv

ely

with

<

<

5u

< .

A

deta

iled s

peci

fica

tion

of th

ese

inte

rven

tions

is g

iven

bel

ow.

—6-

(b)

The

Sol

utio

n

The

gen

eral

sol

utio

n of

(1)

is

give

n in

equ

atio

n (3

).

B(t

) is

the

val

ue o

f th

e

spec

ulat

ive

bubb

le a

t tim

e t.

(3)

5(t)

=

i: E

tf(v)

e(v_

t)dv

+ B

(t).

The

bub

ble

B(t

) ca

n be

an

y st

ocha

stic

or

st

rict

ly d

eter

min

istic

pro

cess

satis

fyin

g

(4)

B(t

)dt

= a'

Etd

B(t

).

Not

e tha

t si

nce

s(t)

is

part

of t

he in

form

atio

n se

t at t

, so

is B

(t).

W

e re

stri

ct

B, w

here

it i

s st

ocha

stic

, to

follo

w a

dif

fusi

on p

roce

ss.

Exc

ept f

or t

he c

ase

of t

he

"Bla

ncha

rd b

ubbl

e",

all

exam

ples

w

ill

in f

act

rest

rict

it

furt

her

to b

e a

twic

e

cont

inuo

usly

di

ffer

entia

ble

func

tion o

f Bro

wni

an m

otio

ns.

Fina

lly,

it is

impo

rtan

t to

note

that

the

aut

hori

ties a

re a

ssum

ed t

o be

una

ble t

o

inte

rven

e di

rect

ly in

the

bub

ble

proc

ess

(and

in

deed

in t

he e

xcha

nge

rate

pro

cess

).

The

y ca

n on

ly r

egul

ate t

he fu

ndam

enta

l £

In t

he i

nter

ior

of th

e ta

rget

zon

e, t

he g

ener

al s

olut

ion

for

s as

a f

unct

ion

of

curr

ent f a

nd c

urre

nt B

is g

iven

by5

.A2(

f(t)

—C

(t))

(5

) s(

t) =

f(t)

+ B

(t) +

+

A1(

t)e

+ A

2(t)

w

ith

/7 +

2nc

2 crf

—7—

A1,

A2

and

C a

re c

onst

ant

as l

ong

as s

rem

ains

in t

he i

nter

ior

of t

he t

arge

t

zone

, hut

can

cha

nge

whe

n s

is a

t its

upp

er o

r lo

wer

bou

ndar

y.

In fa

ct w

e ha

ve

(6)

A1(

t) =

A1(

r)

A2(

t) =

A2(

r)

C(t

) =

—B

(r)

B(r

) is

the

val

ue o

f th

e bu

bble

at t

ime

r, w

hich

is th

e da

te o

f the

mos

t rec

ent

visi

t pri

or to

t of

the

exch

ange

rat

e s

to o

ne o

f th

e bo

unda

ries

, th

at i

s

(7)

r a S

upre

mum

{t' �

t, su

ch th

at s

(t')

= 5u

or S

O')

=

We

also

def

ine s to

be

a va

lue

off

both

str

ictly

gre

ater

than

s an

d ar

bitr

arily

clos

e to

. S

imila

rly,

s

is a

valu

e of

f st

rict

ly le

ss t

han

5 an

d ar

bitr

arily

clo

se to

s.

We

next

def

ine

two

criti

cal v

alue

s of

the

bubb

le,

Bu

and

Bt

Bu

is d

eter

min

ed,

toge

ther

with

A1,

A2

and

by e

quat

ions

(8a

—d)

.

A(s

-l-B

) A

(s-4

-B)

(8a)

5u

5u+

+B

uie

1 u

u+A

e 2

u u

(8b)

=

(1

+ 5u

X +

i A1

eA1(

5u+

+ A

2 eA

2(5u

)

A(s

+B

)

A(s

1-i-B

)

(Sc)

si

=ft

++

Bu+

Aie

1 u+

A2e

2 u

(Sd)

=

(1

+

+ A

1A1

eA1(

51+

+ A

2 eA

2(5)

.

—8—

Ana

logo

usly

, B

1 is

det

erm

ined

, to

geth

er w

ith A

1,

A2

and

f fr

om e

quat

ions

U

(9a-

d).

(9a)

s

= f

+

+ B

1 +

A1

eA1

1)

A2 e2

1)

U

U

a

(9b)

0

= (1

+

)(1

+ A

1 A

1 eA

1(5u

1) +

A2 e2

1))

U

A

(s1-

l-B

1)

(9c)

s1

= 8

1+ +

B

+ A

1 e1

(11)

+ A2

e 2

A (s

1+B

1) +

A2 e2

1)).

(9

d)

0 =

(1 +

f=

s1 +

A1A

1 e

1

If B

> B

(r)

> B

1, th

en

A1(

r) an

d A

2(i-

) ar

e de

fine

d, t

oget

her w

ith f

(r)

and

f1(r

) by

equ

atio

ns (l

Oa—

d).

A [f

(r)+

B(r

)]

(ba)

s

=f (

r)+

B(r

)++

A1(

r)e

l +

A2(

r)e

u u

a

(lob

) =

f1(r

) + B

() +

+

A1(

r)e

+ A

2(r)

e A

(bc)

0 =

(1 +

)(

1 +

A1

A1(

r)e

+ A

2A2(

r)e 2(

TT

) U

(lO

d) 0

= (1

+ q

T)-

)(l +

A1 A

1(r)

e'' + A

2 A

2(r)

e )

—9.

-.

Whe

n B

(r) � B

u A

1(r)

and

A2(

r) a

re d

eter

min

ed by

equ

atio

ns (l

la,b

).

(ha)

s =

s +

+ B

(r)

Ai(

r) e

iu +

A2(

r) eA

250

(ub)

0

= (1

+

+ A

1 A

1 ei

u +

A2

A2

eA25

).

Whe

n B

(r)

B1

, A

1(r)

and A

2(r)

are

dete

rmin

ed b

y eq

uatio

ns (l

ic,d

).

- .\i

A

s1

(hic

) 5r

s1 +

+

B(r

) + A

1(r)

e +

A2(

r) e

(lid

) 0

= (

1 +

+

A

1 el

l + A

2 A

2 e2

1).

Equ

atio

ns (i

oa,b

) de

fine

u

and

f1 fo

r giv

en

8u' s

1 an

d B

. E

quat

ions

(ioc

,d)

are

the

fam

iliar

tang

ency

or n

o—ar

bitr

age

cond

ition

s of

ten

refe

rred

to, i

nacc

urat

ely,

as

'sm

ooth

pa

stin

g' c

ondi

tions

, th

at e

xpec

ted

inte

rven

tions

in f

shou

ld

not

chan

ge t

he

valu

e of

the

exc

hang

e ra

te.

As

we

shal

l se

e in

Sec

tion

IV b

elow

, th

e va

lue

of t

he

bubb

le

may

dep

end

oo

the

curr

ent

valu

e of

th

e un

regu

late

d fu

ndam

enta

l. T

he

cond

ition

s (1

0c—

d), (

iib) a

nd (l

id) a

llow

for

thi

s.

Not

e th

at fo

r B0

< B

C B

t A1

and

A2

are

inde

pend

ent o

f the

val

ue o

f the

bub

ble,

and

the

refo

re c

onst

ant o

ver t

ime.

It

is

also

read

ily se

en t

hat

in th

is c

ase

df

d11

—10

Equ

atio

ns (l

la,b

) ch

arac

teri

ze a

one

—si

ded

'sm

ooth

pas

ting'

con

ditio

n at

the

uppe

r bo

unda

ry (

s =

on

ly.

Equ

atio

ns (l

lc,d

) ch

arac

teri

ze a

one

—si

ded

'sm

ooth

past

ing'

con

ditio

n at

the

low

er b

ound

ary

(s =

on

ly.

The

se n

on—

stan

dard

bou

ndar

y

cond

ition

s w

ill b

e in

terp

rete

d fu

rthe

r be

low

.

We

now

det

ail

the

inte

rven

tions

at

the

uppe

r an

d lo

wer

bo

unda

ries

of

the

exch

ange

rat

e ta

rget

zon

e.

From

equ

atio

n (5

), t

he e

xcha

nge

rate

in th

e ta

rget

zon

e ca

n

be w

ritte

n as

5(

t) =

g(f

(t),

B(r

)) +

B(t

).

The

bub

ble

at ti

me

t can

be

a fu

nctio

n of

the

curr

ent u

nreg

ulat

ed fu

ndam

enta

l f(

t) a

nd/o

r of

oth

er v

aria

bles

z(t

).

Not

e th

at

z(t)

co

uld

be a

fun

ctio

n of

nM

or

exoe

cted

fqe

valu

es o

f th

e un

regu

late

d fu

ndam

enta

l. T

here

fore

B(t

) =

b(f

(t),

z(t)

).

The

exc

hang

e ra

te i

n th

e ta

rget

zon

e

can

ther

efor

e be

wri

tten

as

5 =

h(f,

•), w

here

a

+ . Fo

r sim

plic

ity

we

Of

rest

rict

the

ana

lysi

s to

tho

se b

ubbl

es f

or w

hich

, ev

en i

f B(t

) de

pend

s on

1(

t),

the

func

tion

h(f,

.),

reta

ins

its f

amili

ar s

sha

pe w

ith a

uni

que

inte

rior

min

imum

an

d a

uniq

ue i

nter

ior

max

imum

. In

the

cas

e w

here

Bu

< B

(r) c

B,

u is

the

refo

re th

e

valu

e of

f fo

r w

hich

s =

5u an

d th

e h

func

tion

has

an i

nter

ior m

axim

um, t

hat

is u

is

defi

ned

by h

(fu;

)

= 5u

' =

0

and

•) <

0.

Not

e th

at b

ecau

se of

the

bubb

le,

s m

ay r

each

5u

at v

alue

s of

fun

dam

enta

l dif

fere

nt f

rom

u•

The

se w

ill

be

deno

ted f

. Sim

ilarl

y, s

may

reac

h s

at v

alue

s of

f d

iffe

rent

fro

m f

The

se w

ill b

e

deno

ted

f1.

We

cons

ider

th

e si

mpl

est

poss

ible

exa

mpl

e of

a t

arge

t zo

ne.

A p

laus

ible

inte

rpre

tatio

u of

the

form

al m

odel

is th

at t

here

is a

fully

cre

dibl

e co

mm

itmen

t tha

t the

valu

e of

the

exch

ange

ra

te w

ill b

e co

ntai

ned

with

in a

n ex

ogen

ousl

y gi

ven

boun

ded

rang

e, t

hat i

s —

z C

s

s 5u

< z

. T

he in

terv

entio

ns in

the

fun

dam

enta

l pro

cess

that

kee

p th

e ex

chan

ge r

ate

with

in th

is ra

nge o

ccur

onl

y w

hen

the e

xcha

nge

rate

is at

the

boun

dari

es.

Tw

o di

stin

ct k

inds

of

inte

rven

tions

are

requ

ired

to d

efen

d th

e ta

rget

zo

ne in

the

pres

ence

of a

bubb

le.

11 —

The

fir

st i

s a

pair

of

infi

nite

sim

al re

flec

ting

inte

rven

tions

, _dl

u <

0 a

t th

e

uppe

r bou

ndar

y if

f =

u an

d dI

1> 0

at t

he l

ower

bou

ndar

y of

the

targ

et z

one

if

=

The

se n

egat

ive i

nfin

itesi

mal

inte

rven

tions

at th

e up

per b

ound

ary

s =

s a

nd

posi

tive

infi

nite

sim

al in

terv

entio

ns a

t the

low

er b

ound

ary

s =

5, c

orre

spon

d to

the

orig

inal

sch

eme

anal

yzed

by

Kru

gman

[1

990]

. II

inte

rven

tions

in t

he i

nter

ior

of th

e

targ

et

zone

("

intr

amar

gina

l" i

nter

vent

ions

) w

ere

perm

itted

, a

pair

of

refl

ectin

g

inte

rven

tions

of

exog

enou

sly

give

n fi

nite

mag

nitu

des,

_A

Iu <

0

and

Al1

> 0

co

uld

be a

llow

ed (

see

Floo

d an

d G

arbe

r [19

89])

.

The

sec

ond

clas

s of

inte

rven

tions

com

pris

es i

nter

vent

ions

of f

inite

mag

nitu

de,

that

are

fun

ctio

ns o

f the

cha

nge

in th

e m

agni

tude

of th

e bu

bble

sin

ce t

he p

revi

ous

visi

t

to th

e bo

unda

ries

. T

hey

too

occu

r onl

y w

hen

the

exch

ange

rat

e is

at t

he b

ound

arie

s of

the

targ

et z

one.

T

hey

alw

ays

are

in t

he o

ppos

ite

dire

ctio

n fr

om t

he in

fini

tesi

mal

(typ

e I)

int

erve

ntio

ns at

the

sam

e bo

unda

ry.

We

inte

rpre

t th

ese

(typ

e II

) di

scre

te

inte

rven

tions

as

th

e pa

ssiv

e

acco

mm

odat

ion

of r

atio

nal

spec

ulat

ive

atta

cks

at t

he b

ound

arie

s of

the

tar

get

zone

,

that

are

con

sist

ent

with

the

sur

viva

l of

the

targ

et z

one.

W

e re

fer t

o su

ch s

tock

—sh

ift

port

folio

res

huff

ies

as "

sust

aini

ng"

or "

frie

ndly

" sp

ecul

ativ

e at

tack

s.

The

beh

avio

r of

the

regu

late

d fun

dam

enta

l is

give

n in

equ

atio

ns (1

2a) t

o (1

2c).

The

co

nditi

ons

in (

12a)

, (1

2b),

(12

c(a)

) an

d (1

2c(y

))

are

the

stan

dard

one

s fo

r

regu

late

d Bro

wni

an m

otio

n (s

ee e

.g.

Har

riso

n ]1

985,

p.

80])

. If

>

u an

d if

f =

whe

n =

5u' t

hey

char

acte

rize

a r

efle

ctin

g ba

rrie

r at

f =

u (

give

n by

(12

a),

(12b

)

and

(12c

(a))

. If

f1 c

s an

d if

f =

f1

whe

n s

= s,

th

ey c

hara

cter

ize

a re

flec

ting

barr

ier a

t f =

f1 g

iven

by

(12a

), (

12b)

and

(12

c('y

)).

The

oth

er c

ondi

tions

ar

e no

n-st

anda

rd a

nd a

re r

equi

red

for

our

anal

ysis

of

ratio

nal

bubb

les

in a

targ

et

zone

. If

>

5u'

they

cha

ract

eriz

e (d

iscr

ete)

acco

mm

odat

ed

spec

ulat

sve

atta

cks

from

f =

f to

I =

u' fo

llow

ed b

y (i

nfin

itesi

mal

)

refl

ectio

n at

f =

u (

give

n by

(12

a),

(12b

) an

d (1

2c(f

i)).

If

f1 <

s,

they

cha

ract

eriz

e

—12

acco

mm

odat

ed sp

ecul

ativ

e atta

cks

from

f

= f

to

I =

f, fo

llow

ed b

y re

flec

tion a

t

=1

(giv

en

by (

12a)

, (1

2b)

and

(12c

(6))

. If

f0

s,

they

cha

ract

eriz

e

acco

mm

odat

ed

spec

ulat

ive a

ttack

s fr

om I =

I (to

I =

s (g

iven

by

(12a

), (

12b)

and

(12c

(c))

. If

I � s

, th

ey c

hara

cter

ize

acco

mm

odat

ed

spec

ulat

ive a

ttack

s fr

om I

=

to I

= s

(giv

en b

y (1

2a),

(12

b) a

nd l

2c(s

)).

The

rea

sons

for

thes

e no

n—st

anda

rd

feat

ores

will

bec

ome

clea

r be

low

.

We

rest

rict

atte

ntio

n to

sta

rtin

g st

ates

s

and

star

ting

date

t

= 0

and

defi

ne t

as th

e in

stan

t im

med

iate

ly be

fore

t, th

at is

t_e

lim (

t—ht

) t-

4O

at>

o

The

regu

late

d fun

dam

enta

l is fo

rmal

ly d

efin

ed a

s fo

llow

s:

(12a

) 1(

t) e

pt +

dW

1—10

(t)

+ I/

t) fo

r all

t � 0

(12b

) 1n

and

I ar

e co

nsta

nt if

s1 <

s <

(12c

) jO

an

d I v

ary

whe

n =

or

s

= s

(I

) 1f

f0 >

5 an

d 1

< 5

then

:

(a)

II s

= s

and

I = 1

then

1U

is

cont

inuo

us

and

incr

easi

ng.

(jl)

If

s(t

) =

s an

d 1(

t) =

10,

then

1(

t) =

f0(t

) an

d (a

) ap

plie

s he

ncef

orth

.

('y)

II s

= s1

an

d I =

1 th

en I

is co

ntin

uous

and

incr

easi

ng.

( II

s(t_

) =

s1 an

d 1(

t) =

1,

then

1(t

) =

f/t)

and

()

appl

ies

henc

efor

th.

— 1

3—

(ha)

1f

f � s

then

(c)

If s

(t_)

= s

then

1(t

) =

s

If s

= s

then

(y)

or (

8)

appl

y.

(JIb

) If

1� s

then

(77)

If

s(t)

= s

then

1(t

) =

s1

If s

= s

then

(a)

or

(fi)

app

ly.

It w

ill h

e ap

pare

nt t

hat,

for

any

size

bub

ble

we

can

calc

ulat

e exa

ctly

bot

h th

e

mag

nitu

des

of th

e pa

ssiv

e or

acc

omm

odat

ing

disc

rete

int

erve

ntio

ns re

quir

ed to

allo

w

the

targ

et z

one

to s

urvi

ve a

nd th

e va

lues

of

the

fund

amen

tals

at w

hich

they

will

occ

ur.

In o

ur

rudi

men

tary

mod

el,

ther

e is

no

pos

sibi

lity

of t

he s

pecu

lativ

e bu

bble

infl

uenc

ing

the

beha

vior

of th

e fu

ndam

enta

l. A

bub

ble

in th

is m

odel

can

onl

y in

flue

nce

the

exch

ange

ra

te d

irec

tly.

If th

e pr

oces

s go

vern

ing

the

unre

gula

ted

fund

amen

tal

incl

uded

th

e ex

chan

ge

rate

as

an a

rgum

ent

then

, be

caus

e th

e bu

bble

af

fect

s th

e

exch

ange

ra

te,

it w

ill

indi

rect

ly i

nflu

ence

th

e be

havi

or o

f th

e fu

ndam

enta

l. A

n

inte

rest

ing

anal

ysis

of

a m

odel

in

whi

ch t

he b

ubbl

e af

fect

s th

e ex

chan

ge

rate

bot

h

dire

ctly

and

ind

irec

tly th

roug

h th

e fu

ndam

enta

l is

ana

lyze

d by

Mill

er—

Wel

ler [

1990

a]

and

by M

iller

, Wel

ler

and

Will

iam

son

[198

9].

Fina

lly,

the

beha

vior

of

the

auth

oriti

es at

the

edge

s of

the

tar

get

zone

can

be

sum

mar

ized

as

fol

low

s:

"Ref

lect

w

hen

this

is

suff

icie

nt;

acco

mm

odat

e whe

n th

is i

s

nece

ssar

y.

Ill.

Smoo

th P

astin

g a

Bub

bly E

xcha

nge R

ate

Con

side

r the

cas

e of

an e

xoge

nous

bub

ble,

tha

t is

one

tha

t do

es n

ot d

epen

d on

1.

The

bub

ble

can

be ei

ther

det

erm

inis

tic

or s

toch

astic

. T

he g

raph

ic re

pres

enta

tion o

f

this

cas

e is

sho

wn

in F

igur

es la

and

lb.

We

now

def

ine

the

bubb

le e

ater

of

the

exch

ange

rat

e fo

r a

give

n va

lue

of th

e

— 1

4—

bubb

le,

B(t

). T

he b

ubbl

e ea

ter

of s

fo

r B

(t)

satis

fies

equ

atio

n (5

) w

ith

A1,

A2

and

C

dete

rmin

ed by

equ

atio

ns (6

) an

d (l

Oa—

d) f

or th

e ac

tual

ly p

reva

iling

va

lue

of

the

bubb

le, t

hat

is f

or

C(t

) =

—B

(t).

T

hat

is t

he b

ubbl

e ea

ter

of

s fo

r B

=

B is

give

n by

A1(

f + B

) A

2(f +

B)

(13a

) s=

f++

B+

A1e

+

A2e

A1(

f +

B)

A2(

f +

B)

(13b

) 5u

=fu

+ +

B + A

1 e

u +

A2e

u

A (f

+

B)

A2(

f +

B)

(13c

) O

=l+

A1A

1e 1

u +

A2A

2e

u

A1(

f1 +

B)

A2(

f1 +

B)

(13d

) s1

=f,

+-j

-B +

A1e

+

A2e

A1(

f1+

B)

A2(

f1+

B)

(13e

) O

=1+

A1A

1e

+A

2A2e

Thi

s bu

bble

ea

ter

shif

ts

cont

inuo

usly

w

ith

B;

an in

crea

se (

decr

ease

) in

B

corr

espo

nds

to a

n eq

ual h

oriz

onta

l di

spla

cem

ent

to t

he le

ft (

righ

t) o

f th

e gr

aph.

N

ote

that

the

bubb

le e

ater

giv

es a

ref

eren

ce t

raje

ctor

y on

ly

and

does

not

sho

w t

he a

ctua

l

mov

emen

t of

f an

d s

whe

n B

ch

ange

s un

less

Bu

> B

(r)

> B

1 an

d ei

ther

s =

an

d f=

f or

s=

s1 a

nd f

=f1

. W

hen

s is

in t

he in

teri

or o

f th

e ta

rget

zon

e an

d B

u >

B(r

) > B

1, its

pos

ition

ca

n be

def

ined

with

ref

eren

ce t

o th

e pe

riod

r bu

bble

eat

er,

that

is

the

bubb

le

eate

r

corr

espo

ndin

g to

the

valu

e of

B t

hat p

reva

iled a

t the

pre

viou

s (m

ost

rece

nt) v

isit

of

ui oq 03 sAnqe lnq) 'j jo q2u aqj o; aq osjt w

oo j 'ojqqnq gems A

puapgjns t °a

au!! .ç aq jo 24211 a; o3 si 434JM 'J =

; axoq'T v A

npunoq zaddn 042 03

Aoua2utl jo lulod e oe 'T

a + (1a ';)2 =

g £q uaA

I2 '1 ioj ;aoa aqquq 0112 '(ui2uo alp q2rIoflp 01511 ,ç alp uo si qotqA

t) "iJ Jo 3J0 01J3 02 51

J O

jII{M 3043 03°N

;o; 10300 oqqnq 0112 0AO

O 0a —

1a 100!310A

0 ' + (a ')2 =

qdei2aipuoarT

put u put

20813043'11J=J put

°J=j

;o; uoddoq w

oo siq; °s

puxq joddu 531 o 0201 a2uoqoxo 012 s2uuq

1H

02 ojqqnq

042 JO otis alp us aseanui w

e 1104M suodctsq T

eqht .ioprsuoo M

ON

A

repunoq IO

MO

j 0112

30 9jj

3ulod Aoua2w

e3 t put Arepunoq ;oddn 0l

0u 3uiod A

ouo2ut t swq 4314M

+ (0a 'J)2 =

s A

q UO

A!2 '0

IOJ

10200 oqqnq 042 31e;s eJ ain2sj uj

Aouonuo 0111 JO

uoireiooidop

popadxa ss 3043 '5 JO

oPuoqo jo 0201 popodxo OA

i3iSOd 0 0A

04 J4M O

uiJ Ji 043 JO 3J31

012 02 tusod Any

AO

UO

IIUO

042 Jo uOiltl001dcFe

po330dxa we 43!M

Si 3eq 's Jo a2wsqo

JO 0201 pa300dxa O

A!202011 0 q2!M

p0301301st oq 1PM 0114

.91' 043 JO 342!

042 03 3uiOd L

ull

'3p — s)o =

spa aousg

J = s

UIPIIO

013 4200143 0114 .91' 013 JO 212!; 012 03 504 (0

a;tuep.rooo put 0J ossiosqo) [0 0] usod oqi '(js <

°j q3!M) ojqqnq jrem

s o ;oj

us aqqnq aAs3ssO

cI v Jo osoo 043 10tS11O3 L

juo OM

'L3SA

OIq JO

ots ioj V

a 5 a) s 02

renba 10 11043 1030012 J stq ojqqnq oA

i3o8ou o2itj o put (1q < a <

o) s 11042 °1

504 ajqqnq O

A!3020u

'Asno2otuy °H

<

H ! 3043

115 03 rnba 10 1101{3

8101

oeq ajqqnq aAs3s od o2rei y

°a > a >

o ! 3043 11043 1030012 J

JO O

fijEA

0 0A2

(p—oi)

suolsonbo 4O0M

loJ 0110 51 ajqqnq oAslisod item

s y oqquq O

A!3!SO

d 0a2rej o Jo 3013 q o;n2sj 'ajqqnq aA

r3ssod j[tlS1 0 JO 0503 042

S3uOSO

ldOl

02 0102!J

•1a + (0a 'flP =

Lq U

OA

! s s

JO uo131sod 100300 012 '1H

5! a JO

Ofl[O

A [n30t 01J2

2 011112

30 J! U041

=

0 L3it1auo2 Jo 5501 315042!M

) 0a +

(0a 'j)2 = s

pa3ouap aq

UR

O

iOJ

10200 aqqnq 041 0H

aq

H Jo O

O10A

301J3 301 50fl0U

flO4 01j3 JO

ouo 02

— PT

—16

of

hi).

*2

w

ill

alw

ays

be to

the

lef

t of

ll•

From

11

*1 (o

r 11

2)

ther

e is

a

disc

rete

inte

rven

tion

(a s

tock

—sh

ift i

ncre

ase

in f

) w

hich

, at

a c

onst

ant

exch

ange

rat

e

= s,

mov

es t

he s

yste

m t

o Q

, the

poin

t of

tang

eocy

of

the

bobb

le e

ater

of

B1

and

the

uppe

r lim

it of

the

targ

et z

one.

T

he h

oriz

onta

l dis

tanc

e be

twee

n an

d hi

is

jus

t eq

ual

to t

he d

iffe

renc

e be

twee

n B

1 an

d B

°, i

.e.

— fj

= B

1 —

B°.

Mor

eove

r,

the

hori

zont

al d

ista

nce

betw

een

(1 an

d f

is a

lway

s gr

eate

r th

an t

he

dist

ance

bet

wee

n 11

0 an

d T

hese

tw

o di

stan

ces

wou

ld b

e eq

ual

only

if th

e sl

ope

of

the

bubb

le e

ater

wer

e 45

de

gree

s, w

hich

is n

ot p

ossi

ble.

*1

N

ote

that

at

Il

the

smoo

th p

astin

g (o

r no

—ar

bitr

age)

con

ditio

ns

(lO

a—d)

are

not s

atis

fied

. A

t Il

l in

Fig

ure

la th

e sm

ooth

pas

ting

cond

ition

s ar

e sa

tisfi

ed a

nd t

he

expe

cted

ra

te o

f cha

nge

of th

e ex

chan

ge r

ate

is n

egat

ive

(Ili

is to

the

righ

t of

the

45

line

thro

ugh

the

orig

in).

Il

l is

the

refo

re

cons

iste

nt w

ith c

redi

ble

inte

rven

tion

to

prev

ent

s fr

om ri

sing

abo

ve

s.d.

T

his

is n

ot th

e ca

se w

ith th

e la

rge b

ubbl

e sh

own

in

Figu

re l

b.

Ill

is to

the

left

of t

he 4

5* l

ine

thro

ugh

the

orig

in a

nd t

here

fore

rep

rese

nts

a po

sitiv

e ex

pect

ed

rate

of c

hang

e of

the

exc

hang

e ra

te,

inco

nsis

tent

w

ith

a cr

edib

le

defe

nse

of th

e up

per

limit

of th

e ta

rget

zon

e. In

this

cas

e th

e in

terv

entio

n rul

e le

ads

to

the

follo

win

g sc

enar

io.

From

Il

l a

one—

off

disc

rete

(st

ock-

shif

t) in

crea

se i

n f

brin

gs

the

syst

em to

11u

' tha

t is

the

poi

nt [

st,

su],

an a

rbitr

arily

sm

all

dist

ance

to th

e ri

ght

of t

he in

ters

ectio

n of

s =

an

d th

e 45

0 lin

e th

roug

h th

e or

igin

. Si

nce

s is

kep

t

cons

tant

at

s =

8n'

ther

e ar

e no

arb

itrag

e pro

fits

. Fr

om 1

l the

syst

em, w

ith A

1 an

d

A2

now

def

ined

by

the

one—

side

d 's

moo

th p

astin

g' c

ondi

tions

giv

en in

(lla

,b),

can

now

agai

n m

ove i

nto

the

inte

rior

of

the

targ

et z

one.

In F

igui

e lb

, the

gra

ph o

f s w

ith A

1 an

d A

2 de

term

ined

by

(ha)

and

(lib

) an

d B

= B

u de

note

d s'

, coi

ncid

es w

ith t

he b

ubbl

e ea

ter

of B

u Fo

r a

slig

htly

larg

er

valu

e of

B

, the

gra

ph o

f s

with

A

1 an

d A

2 de

term

ined

by

tbe

one

-sid

ed s

moo

th

past

ing

cond

ition

s (ha

) and

(lib

) is

deno

ted

s' '.

A la

rger

valu

e of

s

yiel

ds s

'' A

s B

go

es to

infi

nity

the

gra

ph o

f s

appr

oach

es a

vert

ical

line

thro

ugh

t1u

—17

Not

e th

at,

by c

onst

ruct

ion,

we

alw

ays

have

a l

ocal

max

imum

at t

he ta

ngen

cy

poin

t f2

(fo

r fin

ite v

alue

s of

B).

For

pos

itive

unr

egul

ated

real

izat

ions

of

f, s

tand

ard

infi

nite

sim

al re

flec

tion

stop

s th

e re

gula

ted

valu

e of

f f

rom

ris

ing

beyo

nd s.

For

nega

tive r

ealiz

atio

ns of

the

unre

gula

ted f

, the

sys

tem

mov

es (

give

n B

) in

to th

e in

teri

or

of th

e ta

rget

zon

e.

As

in t

he c

ase

of a

sm

all

bubb

le,

a va

riat

ion

in t

he v

alue

of th

e

bubb

le i

n th

e in

teri

or o

f the

targ

et zo

ne re

pres

ents

a ve

rtic

al d

ispl

acem

ent o

f the

gra

ph

of th

e ex

chan

ge r

ate

(with

con

stan

t val

ues

of A

1 an

d A

2).

With

a l

arge

bub

ble,

the

shap

e of

the

grap

h ch

ange

s ev

ery

time

the

exch

ange

rat

e vi

sits

the

bou

ndar

ies a

t a n

ew

valu

e of

B.

Sinc

e th

e bu

bble

gro

ws

expo

nent

ially

(i

n ex

pect

atio

n)

we

expe

ct

mor

e fr

eque

nt

inte

rven

tions

at

the

uppe

r (l

ower

) bo

unda

ry if

the

cur

rent

va

lue

of t

be b

ubbl

e is

posi

tive

(neg

ativ

e) th

an w

ith

a ze

ro b

ubbl

e.

We

cons

ider

the

se r

esul

ts

to b

e qu

ite

intu

itive

. O

ne

wou

ld e

xpec

t an

exc

hang

e ra

te s

ubje

ct t

o a

posi

tive

and

grow

ing

(in

expe

ctat

ion)

bub

ble

to s

pend

a lo

t of

time

at th

e up

per

boun

dary

of

the

targ

et z

one,

with

the

aut

hori

ties

inte

rven

ing

to c

ount

erac

t not

onl

y th

e un

regu

late

d fu

ndam

enta

l

but a

lso

the

bubb

le.

It i

s ap

pare

nt t

hat

whe

n s

reac

hes

5u w

ith

a la

rge

bubb

le,

the

stoc

k—sh

ift

incr

ease

in

f m

ust t

ake

the

syst

em to

a p

ositi

on s

tric

tly t

o th

e ri

ght

of t

he 4

50 l

ine

thro

ugh

the

orig

in.

On

the

450

line

the

expe

cted

rat

e of

dep

reci

atio

n is

zero

. If

we

impo

se t

he 's

moo

th p

astin

g' c

ondi

tions

w

hen

f =

= 5u

' (sm

all)

posi

tive r

ealiz

atio

ns o

f

incr

emen

ts

in t

he u

nreg

ulat

ed f

unda

men

tal

will

be

offs

et w

ith i

nfin

itesi

mal

refl

ectin

g

inte

rven

tions

. Neg

ativ

e re

aliz

atio

ns o

f df

wou

ld b

ring

the

sys

tem

to th

e le

ft o

f the

450

line

thro

ugh

the

orig

in,

that

is

to a

poi

nt w

ith a

pos

itive

exp

ecte

d ra

te o

f ch

ange

of s

.

Thi

s is

of

cour

se i

ncon

sist

ent w

ith

a cr

edib

le d

efen

se o

f the

tar

get

zone

. If

pri

vate

agen

ts

belie

ved

that

the

aut

hori

ties

wer

e st

abili

zing

th

e ex

chan

ge

rate

at 'e

u' u

b a

frie

ndly

spec

ulat

ive

atta

ck o

f in

fini

tesi

mal

size

wou

ld o

ccur

, br

ingi

ng th

e sy

stem

back

to

1u'

Thi

s am

ount

s to

sta

biliz

ing

the

fund

amen

tal

at f

= 5u

Si

nce

—18

Etd

s =

a(s

— f)

th

e ex

chan

ge

rate

wou

ld e

ffec

tivel

y be

exp

ecte

d to

be

fixe

d.

It is

easi

ly se

en th

at a

n ex

chan

ge r

ate

cann

ot re

mai

n fi

xed

for

any

fini

te in

terv

al o

f tim

e if

the

fund

amen

tal

is c

onst

ant

and

ther

e is

a n

on—

zero

spe

cula

tive

bubb

le.

Con

side

r

equa

tion

(3),

for

the

spec

ial

case

in w

hich

f is

kep

t con

stan

t at

s fo

reve

r.

The

sol

utio

n

for

the

exch

ange

rat

e is

giv

en b

y s

= 5u

+ B

whi

ch i

mpl

ies

that

, un

less

B =

0,

the

exch

ange

rat

e ca

nnot

be c

onst

ant.

The

re is

a s

impl

e, un

ifyi

ng c

hara

cter

izat

ion

of th

e in

terv

entio

ns th

at su

stai

n th

e

targ

et z

one.

It

is th

at th

e go

vern

men

t is

cre

dibl

y co

mm

itted

to in

fini

tesi

mal

refl

ectin

g

inte

rven

tions

at th

e bo

unda

ries

s1

an

d T

he v

alue

of

the

fund

amen

tal a

t w

hich

thes

e in

fini

tesi

mal

refl

ectin

g int

erve

ntio

ns at

the

mar

gins

occ

ur a

re d

eter

min

ed by

the

appr

opri

ate

two—

side

d or

one

—si

ded

"no—

arbi

trag

e"

(or

smoo

th

past

ing)

bou

ndar

y

cond

ition

s.

Con

side

r fi

rst t

he c

ase

of a

sm

all p

ositi

ve b

obbl

e sh

own

in F

igur

e la

. T

he v

alue

of th

e fu

ndam

enta

l at w

hich

the

exch

ange

rat

e re

ache

s, s

ay, t

he u

pper

bou

ndar

y u'

in Fi

gure

la

) is

dif

fere

nt fr

om th

e va

lue

of

f th

at c

hara

cter

izes

th

e ta

ngen

cy o

f th

e

bubb

le e

ater

and

the

upp

er b

ound

ary

(4 in

Fig

ure

la)

exce

pt i

n th

e ca

se o

f a

zero

bubb

le.

Sinc

e u

is le

ss t

han

u' th

e di

scre

te (

stoc

k-sh

ift)

jum

p in

the

fund

amen

tal

that

tak

es th

e sy

stem

to t

he re

flec

tion

poin

t Q

ca

n be

int

erpr

eted

as

a fr

iend

ly o

r

sust

aini

ng sp

ecul

ativ

e at

tack

whi

ch is

acc

omm

odat

ed

by th

e au

thor

ities

. *1

11

is

not

a c

redi

ble

eqm

hbrs

um i

f th

e pr

ivat

e se

ctor

kno

ws

the

gove

rnm

ent's

inte

rven

tion

rule

, as

the

re c

an b

e no

traj

ecto

ry s

= g(

f, B

0) +

B1

that

th

e up

per

boun

dary

if th

ere

is in

fini

tesi

mal

refl

ectio

n at

the

poi

nt w

here

s

= g

(f,

B0)

+ B

1 =

The

re is

a c

redi

ble

refl

ectin

g equ

ilibr

ium

at

Ili.

Not

e th

at

Etd

s =

n(su

_41)

dt>

Etd

s =

n(s0

_4)d

t

—19

Sinc

e th

e ex

pect

ed

rate

of

chan

ge o

f s

at

in F

igur

e la

is (

posi

tive

and)

high

er th

an t

he (n

egat

ive)

exp

ecte

d ra

te o

f cha

nge

of s

at

Il ,

the

shift

from

1u

to

can

he i

nter

pret

ed a

s an

inc

reas

e in

the

rela

tive

dem

and f

or h

ome

coun

try

mon

ey

due

to a

dec

reas

e in

the

dom

estic

—fo

reig

n in

tere

st d

iffe

rent

ial.6

T

he a

utho

ritie

s

oblig

ingl

y ac

com

mod

ate

this

frie

ndly

or s

usta

inin

g sp

ecul

ativ

e at

tack

.7 N

ote

that

if at

time

t th

e ex

chan

ge r

ate

reac

hes

s at

f =

fu(B

(rfl

that

is w

hen

the

curr

ent

valu

e

of th

e bu

bble

is t

he s

ame

as t

he v

alue

of

B

at th

e m

ost

rece

nt p

revi

ous

boun

dary

visi

t, th

e si

ze o

f th

e fr

iend

ily s

pecu

lativ

e at

tack

is o

bvio

usly

zer

o, a

nd o

nly

a re

flec

ting

inte

rven

tion o

ccur

s.

With

a la

rge b

ubbl

e (s

ee F

igur

e ib

) th

e st

ock-

-shi

ft i

ncre

ase i

n f

from

f =

f*1

to f

= s

can

also

be

inte

rpre

ted

as t

he p

assi

ve a

ccom

mod

atio

n of

a di

scre

te f

rien

dly

spec

ulat

ive

atta

ck.

The

bub

ble

eate

r of

the

larg

e bu

bble

B

1 in

Fig

ure

lb p

rodu

ces

a

tang

ency

po

int

with

the

upp

er b

ound

ary

at

whi

ch i

s to

the

lef

t of

the

45

line

thro

ugh

the

orig

in.

It th

eref

ore

has

a po

sitiv

e exp

ecte

d ra

te o

f cha

nge

of s,

whi

ch i

s

inco

nsis

tent

with

a c

redi

ble d

efen

se o

f th

e ta

rget

zon

e.

At

11u'

th

e ex

chan

ge r

ate

is

expe

cted

to a

ppre

ciat

e, w

hich

is c

onsi

sten

t with

a cr

edib

le ta

rget

zon

e.

The

pol

icy

rule

w

e us

e to

def

end

the

targ

et z

one

in t

he p

rese

nce

of r

atio

nal

spec

ulat

ive

bubb

les i

s no

t the

onl

y on

e ca

pabl

e of

del

iver

ing

the

surv

ival

of th

e ta

rget

zo

ne.

We

coul

d, f

or in

stan

ce, a

pply

the

one—

side

d sm

ooth

pas

ting

cond

ition

s [(

lla,b

) w

hen

s =

an

d (l

lc,d

) whe

n s

=

even

whe

n B

u >

B(r

) >

B1,

at a

ny e

xoge

nous

ly

give

n va

lue

of

f = u

> 5u

w

hen

s =

s an

d at

any

exo

geno

usly

gi

ven

valu

e of

=

<

whe

n s

= s

. W

e ch

ose

the

part

icul

ar r

ule

give

n in

(lO

a—d)

, (lla

,b)

and

(llc

,d)

both

bec

ause

it i

s qu

ite i

ntui

tive

and

beca

use

it is

as

clos

e as

pos

sibl

e to

the

orig

inal

mod

el.

In p

artic

ular

it i

nclu

des

the

trad

ition

al 's

moo

th p

astin

g" s

olut

ion

as

the

spec

ial

case

whe

n th

ere

is n

o bu

bble

.

—20

W. A

Sur

feit

of B

ubbl

es

Con

side

r ag

ain

the

gene

ral

solu

tion

to e

qnat

ion

(1)

give

n in

(3)

. T

he fi

rst t

erm

on th

e R.H

.S. o

f (3

) is

com

mon

ly c

alle

d th

e fu

ndam

enta

l so

lutio

n, s

1

(14)

5f

0) =

a

Etf(

v) e

(\T

_t)d

v

Con

side

r eq

uatio

n (1

5) b

elow

, w

hich

is a

spec

ial

case

of e

quat

ion

(5).

.\f

If

(15)

s=

f++

A1e

1 +

A2e

2

In th

e te

rmin

olog

y of

McC

allu

m

[198

3] th

is i

s th

e m

inim

al

stat

e so

lutio

n: i

t in

volv

es o

nly

the

stat

e va

riab

le(s

) an

d th

e st

ate

vari

able

s ent

er i

n a

"min

imal

" w

ay.

The

fir

st tw

o te

rms

on t

he R

.II.

S of

(15

) ar

e th

e fu

ndam

enta

l sol

utio

n fo

r th

e

unre

gula

ted f

unda

men

tal.

All

of th

e R

.H.S

. of

(15

), f

or s

ome

give

n (n

on—

zero

) val

ues

of A

1 an

d A

2, is

the

fund

amen

tal s

olut

ion

for t

he r

egul

ated

fund

amen

tal.

For

the

unre

gula

ted

fund

amen

tal,

non—

zero

va

lues

of

A1

and/

or A

2 w

ould

perm

it w

hat

Froo

t an

d O

bstf

eld

[198

9c]

call

intr

insi

c bu

bble

s, w

ithin

the

cla

ss o

f fu

nctio

ns ex

pres

sing

s(

t)

as a

func

tion

of c

urre

nt f

unda

men

tals

(f(

t))

ppjy

.

For

the

regu

late

d fu

ndam

enta

l, th

e ex

iste

nce

of in

trin

sic

bubb

les

wou

ld r

equi

re

that

f

ente

rs t

he s

olut

ion

for

the

exch

ange

rat

e in

the

int

erio

r of

the

targ

et z

one

thro

ugh

term

s oth

er th

an th

e on

es a

ppea

ring

(w

ith A

1 an

d A

2 de

term

ined

by

(lO

a—d)

,

(lla

,b)

or

(llc

,d))

on

the

11.1

1.5.

of

equ

atio

n (1

4)

and

with

no

othe

r va

riab

le(s

) in

clud

ed.

Such

intr

insi

c bu

bble

s ar

e cl

earl

y im

poss

ible

.

The

lite

ratu

re h

as g

ener

ally

cas

t its

dis

cuss

ion

of bu

bble

s fo

r our

cla

ss o

f mod

els

in t

erm

s of

the

beha

vior

of

s as

a f

unct

ion

only

of

curr

ent

f for

dif

fere

nt v

alue

s of A

1 an

d A

2.

For

exam

ple,

in

the

case

of

a fr

eely

flo

atin

g ex

chan

ge r

ate,

unl

ess

A1

= 0,

the

—21

devi

atio

n of

s fr

om f

will

bec

ome

infi

nite

as

I gro

ws

with

out b

ound

(si

nce

A1

> )

and

unle

ss A

2 =

0,

the

devi

atio

n of s

from

f w

ill b

ecom

e in

fini

te a

s f f

alls

with

out b

ound

(sin

ce

A2

< 0

).

A1

= A

2 =

0

rule

s ou

t on

ly

intr

insi

c bu

bble

s.

Man

y ki

nds

of

exch

ange

rat

e bu

bble

s ar

e po

ssib

le u

nder

a f

loat

ing

rate

eve

n if

A1

= A

2 =

0.

With

such

bub

bles

, s c

an d

evia

te fr

om f

(for

any

giv

en va

lue

of f

) by

even

tual

ly un

boun

ded

amou

nts.

It is

rec

ogni

zed

in t

he l

itera

ture

(Fr

oot

and

Obs

tfel

d [1

989b

] are

an

exam

ple)

,

that

the

choi

ce o

f A

1 an

d A

2 on

ly r

estr

icts

the

rela

tions

hip b

etw

een

s an

d f,

but

that

stat

e va

riab

les o

ther

than

I m

ay e

nter

the

solu

tion

for

s th

roug

h bu

bble

s.

In g

ener

al,

bubb

les

can

intr

oduc

e on

e or

mor

e ad

ditio

nal s

tate

var

iabl

es i

nto

the

solu

tion

for

s.

Subj

ect

to th

e co

nstr

alnt

that

the

se b

ubbl

es o

bey

(4),

th

ey c

an c

ause

s t

o de

viat

e in

alm

ost a

ny c

once

ivab

le m

anne

r fro

m g

(f,

.),

for

any

valu

es o

f A1

and

A2.

Wha

t is

not,

we

belie

ve,

reco

gniz

ed a

s cl

earl

y is

tha

t s

can

depe

nd o

n fi

n w

ays

othe

r th

an g

iven

by

g(f,

•)

(or

the

R.H

.S.

of (1

5)),

pro

vide

d s

also

dep

ends

(th

roug

h

B)

on a

t lea

st o

ne o

ther

sta

te v

aria

ble.

We

show

bel

ow th

at t

he b

ubbl

e at

tim

e t,

B(t

)

can

be a

fun

ctio

n of

the

unre

gula

ted

fund

amen

tal

f(t)

, pr

ovid

ed

it al

so d

epen

ds o

n

som

e ot

her

stat

e va

riab

le (

or st

ate

vari

able

s) z

(t),

with

P1.

1 =

0, w

hich

ens

ures

that

01

(t)

Etd

B =

csB

dt.

Thi

s im

plie

s th

at w

ithin

the

targ

et z

one,

dB

ca

n be

a fu

nctio

n of

df.

z(t)

ca

n (b

ut n

eed

not)

be

a fu

nctio

n ex

clus

ivel

y of

pas

t an

d/or

ant

icip

ated

fut

ure

valu

es

of t

he u

nreg

ulat

ed f

unda

men

tal.

We

call

this

cla

ss o

f ge

nera

lized

in

trin

sic

bubb

les "

fund

amen

tal-

.dep

ende

nt" b

ubbl

es.

(a) B

ubbl

es th

at d

on't

burs

t

We

begi

n by

con

side

ring

spe

cula

tive

bubb

les,

bot

h de

term

inis

tic an

d st

ocha

stic

,

that

don

't co

llaps

e.

Exa

mpl

es o

f non

—co

llaps

ing b

ubbl

e pr

oces

ses

satis

fyin

g (4

) in

clud

e

the

sim

ple

exog

enou

s de

term

inis

tic p

roce

ss

give

n in

(16

),

whe

re B

(0)

= B

0 is

an

— 2

2 —

arbi

trar

y in

itial

val

ue fo

r the

B p

roce

ss.

(16)

B

(t) =

B0e

4.

An

exam

ple

of a

si

mpl

e ex

ogen

ous

(bac

kwar

d—lo

okio

g) s

toch

astic

bub

ble

satis

fyin

g (4

) is

give

n in

equ

atio

ns (1

7a,b

), w

ith B

0 ar

bitr

ary.

(17a

) dB

(t)

= n

B(t

)dt +

abd

Wb

(llb

) B

(t) =

j.t e

abdW

b(v)

+

e'B

0.

The

par

amet

er

cb is

the

ins

tant

aoeo

us va

rian

ce o

f th

e bu

bble

pro

cess

. N

ote

that

5b

co

uld

be a

fuo

ctio

n of

s,

f,

t or

any

oth

er s

et o

f va

riab

les

with

out t

his

affe

ctin

g th

e fa

ct t

hat

equa

tion

(17a

) sat

isfi

es

(4).

Fo

r in

stan

ce, i

f U

B

is a

pos

itive

cons

tant

, th

e bu

bble

can

"ch

ange

sig

n".

With

geo

met

ric

Bro

wni

an

mot

ion

prop

ortio

nal t

o B

), s

ign

reve

rsal

s of

the

bub

ble

are

rule

d ou

t, al

thou

gh B

ca

n ch

ange

dIre

ctio

n.

Impo

rtan

t fo

r ou

r an

alys

is,

the

bubb

le p

roce

ss

can

itsel

f be

a

func

tion

of

curr

ent,

past

or

antic

ipat

ed f

utur

e va

lues

of

the

unre

gula

ted

fund

amen

tal

f.

One

exam

ple

is g

iven

in e

quat

ions

(18

a,b)

bel

ow.

In w

hat f

ollo

ws

the

vari

able

z(

t)

is

eith

er s

tric

tly d

eter

min

istic

or f

ollo

ws

a di

ffus

ion

proc

ess.

T

he z

(t)

proc

ess

is a

twic

e

cont

inuo

usly

dif

fere

ntia

ble

func

tion

of it

s ar

gum

ents

. T

he k

ey d

efin

ing

prop

erty

of

z(t)

is

tha

t it d

oes

not v

ary

with

f(t

), i

.e.

= 0.

Fo

r the

exa

mpl

e giv

en be

low

, U

f(t)

th

e z

proc

ess

give

n in

equ

atio

n (l

Sb)

(a b

ackw

ard—

look

ing

dete

rmin

istic

pro

cess

)

gene

rate

s, to

geth

er w

ith e

quat

ion

(18a

), a

bubb

le p

roce

ss th

at sa

tisfi

es e

quat

ion

(4).

—23

-—

(isa)

B

(t) =

kf(

t) +

z(t

) k

# —

1

(18b

) z(

t) =

et

_[k(

of(v

) —

#)]d

v +

ez0

is a

n ar

bitr

ary

initi

al v

alue

for

the

z pr

oces

s.

Not

e th

at t

he b

ound

ary

cond

ition

s

(iO

a—d)

ca

nnot

be

satis

fied

if

k =

—1

and

that

, gi

ven

the

boun

dary

cond

ition

s (i

ia,b

)

or (

lied)

, A1

and

A2

are

inde

term

inat

e if k

= —

1.

Mor

e ge

nera

lly, e

quat

ion

(18a

) will

hol

d pr

ovid

ed z

sat

isfi

es

(19)

E

tdz(

t) =

[k(

of(t

) - j)

+ o

z(t)

]dt.

The

z p

roce

ss

give

n in

(1

9)

(and

th

eref

ore

the

B

proc

ess

itsel

f) c

an b

e

forw

ard—

look

ing,

bac

kwar

d—lo

okin

g (o

r a

linea

r co

mbi

natio

n of

the

for

war

d—lo

okin

g

and

back

war

d—lo

okin

g so

lutio

ns),

it

can

be s

tric

tly d

eter

min

istic

or st

ocha

stic

.

An

exam

ple

of

a fo

rwar

d—lo

okin

g so

lutio

n fo

r z

(now

tr

eate

d as

a

non—

pred

eter

min

ed v

aria

ble)

whi

ch s

atis

fies

equ

atio

ns

(isa

) an

d (1

9)

(and

of

cou

rse

equa

tion

(4))

is g

iven

in e

quat

ion

(20)

.

(20)

z(

t) =

- J e(

t)E

tk(a

i(v)

- ji)

dv +

R(t

).

Her

e R

(t)

is a

ny

stri

ctly

det

erm

inis

tic

or s

toch

astic

pro

cess

w

hich

sat

isfi

es

Btd

R(t

)dt

= c

sR(t

)dt.

It c

an b

e pr

edet

erm

ined

or

non

—pr

edet

erm

ined

. O

ne p

ossi

ble

choi

ce f

or

R(t

) w

ould

be

the

proc

ess

give

n on

the

rig

ht—

hand

sid

e of

(iS

b).

Thi

s

wou

ld m

ake

the

bubb

le

B(t

) in

(is

a) a

func

tion

of c

urre

nt,

past

and

ant

icip

ated

futu

re v

alue

s of

the

unre

gula

ted f

unda

men

tal.

Ano

ther

inte

rest

ing

bubb

le p

roce

ss i

nvol

ving

f

is th

e fo

llow

ing:

—24

(21a

) B

(t) =

fO)z

(t)

(21b

) dz

(t) =

f(t)

[ai(t

) — p]

z(t)

dt +

adW

5 If

0 z(

t)=

0 1=

0

(dW

and

dW1

are

assu

med

to

be co

ntem

pora

neou

sly

inde

pend

ent)

.

An

exam

ple

of a

cla

ss o

f ra

tiona

l bub

bles

th

at h

as t

he p

rope

rty

that

-

01(t

) de

pend

s on

1(

t) is

giv

en in

(22a

—b)

.

(22a

) B

(t) =

f(t)

2 +

z(t)

(22b

) E

tdz(

t) =

[o(f

(t)2

+ z

(t))

— 2

f(t)

p — o]

dt.

A s

impl

e bac

kwar

d—lo

okin

g de

term

inis

tic

solu

tion

for

a th

at sa

tisfi

es (

22b)

is

(23)

z(

t) =

j eo(

t_v)

[a(v

)2 —

2f(

v) —

a]dt

+ e

ulz0

.

If B

ca

n de

pend

on

the

regu

late

d va

lue

of th

e fu

ndam

enta

l as

wel

l as

on

the

unre

gula

ted

fund

amen

tal,

we

can

gene

rate

the

very

unu

sual

bub

ble

give

n in

equ

atio

n

(24)

.

(24)

B

(t) =

- g(f

(t),

B(r

)) +

kf(

t) +

z(t

).

Thi

s bu

bble

is c

onsi

sten

t with

the

law

of

mot

ion

(4),

pro

vide

d th

at (2

5) h

olds

:

— 2

5 —

(25)

Etd

z(t)

= {—

a[g

(f(t

),.)

— kf

(t)

— z(

t)] +

[g'(f

(t),

.) —

k]p

+ g

' '(f(

t),.)

a}dt

Equ

atio

n (2

6) i

s an

exa

mpl

e of

a s

impl

e st

ocha

stic

proc

ess

for

z th

at

satis

fies

(25)

(26)

z(

t) =

eQ

(t) [{

-a[g

(f(v

),.)

- kf(

v)]

+ [g

' (f(

v),.)

—k]

p +

g

/ (fN

),.)

a}dv

+ cz

dWz(

v)] dv

+ e

z0.

Whi

le th

is b

ubbl

e is

con

sist

ent w

ith t

he l

aw o

f m

otio

n (4

) w

ithin

the

targ

et

zone

, it

does

not

per

mit

the

boun

dary

cond

ition

s to

be

satis

fied

. W

ith th

is p

roce

ss f

or

B(t

) th

e ex

chan

ge r

ate

is g

iven

by

(27)

s(

t) =

kf(

t) +

z(t)

.

Not

e th

at in

this

cas

e s

incr

ease

s or

dec

reas

es l

inea

rly

with

f

(if k

# 0

), t

hus

mak

ing

"sm

ooth

pas

ting"

bo

unda

ry c

ondi

tions

im

poss

ible

to

appl

y.

Whe

n k

= 0

, 5(

t)

is in

depe

nden

t of

f(t)

, al

thou

gh

z(t)

of

cou

rse

depe

nds

on p

ast v

alue

s of

f

(and

of f

l

(b)

Bub

bles

that

bur

st

Bla

ncha

rd [

1979

] pr

ovid

es

an i

nter

estin

g ex

ampl

e of

a r

atio

nal

spec

ulat

ive

bubb

le w

hose

exp

ecte

d du

ratio

n (t

ime

until

the

mom

ent a

col

laps

e oc

curs

) is

fini

te.

It

is

mos

t ea

sily

mot

ivat

ed i

n di

scre

te t

ime.

E

quat

ion

(28)

is

the

disc

rete

ana

log

of

equa

tion

(1).

(28)

5t

= 0

(1 +

u)f

t + (1

+ a

)'Ets

t+r

0> 0

— 2

6

The

sol

utio

n of

(28

) co

o±tio

n on

cur

rent

and

exp

ecte

d fu

ture

vue

s of

1t i

s

(29)

s

= o

(i +

0)_

i E0

(1 +

or'

Etf

t+ +

Bt

whe

re

Bt

satis

fies

(30)

E

tBt+

i = (1

+ o)

Bt.

Con

side

r th

e fo

llow

ing

proc

ess

for B

(31)

B

t+i =

(1 —

irr'(

i + o

)Bt

+ +

i w

ith p

roba

bilit

y 1

Bt+

i = 6t

+i

with

pro

babi

lity

r

whe

re

0< r

< 1

an

d

(32)

E

tct+

i = E

tct+

i = 0.

It is

eas

ily c

heck

ed t

hat

(31)

an

d (3

2) s

atis

fy (

30).

E

quat

ions

(31)

an

d (3

2)

defi

ne a

bub

ble

(sto

cbas

tic if

t

or

is a

ran

dom

var

iabl

e)

for

whi

ch th

ere

is a

cons

tant

pro

babi

lity

of co

llaps

e,

ir,

each

per

iod.

If

a c

olla

pse

occu

rs i

n pe

riod

t+

l, th

e ex

chan

ge r

ate

retu

rns

to it

s fu

ndam

enta

l val

ue if

=

0.

Mor

e ge

nera

lly, i

f

is r

ando

m,

the

expe

cted

po

st—

colla

pse

valu

e of

the

exc

hang

e ra

te i

s its

fund

amen

tal v

alue

, but

the

rea

lized

val

ue o

f th

e ex

chan

ge r

ate

can

diff

er f

rom

this

expe

ctat

ion b

y a

zero

mea

n fo

reca

st e

rror

. If

the

bubb

le c

olia

pses

in

peri

od

t+1

and

—27

is

non—

zero

, a

new

bub

ble

star

ts i

n pe

riod

t+

1 w

hicb

fol

low

s a

law

of

mot

ion

like

(31)

and

(32)

, bu

t pos

sibl

y w

itb a

dif

fere

nt v

alue

of

ir a

nd d

iffe

rent

(al

thou

gh st

ill

zero

mea

n) ra

ndom

dis

turb

ance

term

s c

and

c.

In co

ntin

uous

tim

e th

e sa

me

idea

is c

aptu

red

as f

ollo

ws.

ir

now

is t

he c

onst

ant

inst

anta

neou

s pr

obab

ility

of

colla

pse

of t

he b

ubbl

e.

Con

ditio

nal

on t

he b

ubbl

e no

t

havi

ng c

olla

psed

by

time

t, th

e pr

obab

ility

of

the

bubb

le l

astin

g til

l tim

e t +

At

is

e_iA

t (A

t 0)

. W

e co

nsid

er t

he f

ollo

win

g pr

oces

s fo

r B.

As

befo

re,

Wb

and

Wb

are

stan

dard

Wie

ner p

roce

sses

. C

ondi

tiona

l on

the

bubb

le h

avin

g la

sted

till

time

t, w

e

have

(33)

B

(t+

At)

= e

( r)

AtB

(t) +

Cb Jl

e(X

WbN

) w

ith p

roba

bilit

y

B(t

+A

t) =

b Je

b(v)

-i

rAt

with

pro

babi

lity

1 —

e

It fo

llow

s th

at t

he e

xpec

ted

rate

of

chan

ge o

f B

per

uni

t tim

e ov

er t

he in

terv

al

[t,

t+A

t] is

B B

(t+

At)

— B

(t)

oAt

(34

____

____

____

__ _

(e

—_1

)Bt

At

At

In t

he li

mit

as

At

0 th

is b

ecom

es

Etd

B(t

) =

oB

(t)d

t, as

req

uire

d. W

hile

the

bubb

le l

asts

, the

inst

anta

neou

s rat

e of

cha

nge

of t

he b

ubbl

e is

giv

en by

—28

dB(t

) (a

+ ir

)B(t

)dt +

cbdW

b(t)

.

Whi

le th

e ex

pect

ed r

ate

of c

hang

e of

the

bubb

le i

s E

tdB

= a

Bdt

, th

e ex

pect

ed

rate

of v

aria

tion

of

B

cond

ition

al

on th

e bu

bble

sur

vivi

ng t

he n

ext i

nsta

nt, i

s gi

ven

by (a

+ ir

)Bdt

.

Let

tc be

the

(ran

dom

) tim

e w

hen

the

bubb

le c

olla

pses

. If

the

valu

e of

B

at

tc

(a-i

-r)(

t —

v)

the

time

of c

olla

pse,

B

(tc)

= b

e

c dW

b(v)

is n

on—

zero

, a

new

bub

ble

will

sta

rt a

t tc

pos

sibl

y w

ith a

dif

fere

nt v

alue

of

ii,

diffe

rent

whi

te n

oise

pro

cess

es

dWb

and

dWb

and

diff

eren

t va

lues

for

ab

and

A B

lanc

hard

bubb

le i

s pe

rfec

tly

com

patib

le w

ith o

ur t

arge

t zon

e m

odel

. W

hen

it co

llaps

es t

here

will

be

a di

scre

te c

hang

e in

the

leve

l of

the

exch

ange

rat

e, w

hich

lie

s

on t

he t

raje

ctor

y ap

prop

riat

e to

the

new

(po

st—

colia

pse)

bub

ble.

Si

nce

this

cha

nge

is

unex

pect

ed,

no a

rbitr

age o

ppor

tuni

ties a

rise

.

zone

s w

ith fu

ndam

enta

l—de

pend

ent

bubb

les

If th

e va

lue

of th

e bu

bble

at t

ime

t is

a fu

nctio

n of

the

cont

empo

rane

ous

valu

e

of t

he u

nreg

ulat

ed f

unda

men

tal,

then

with

in t

he t

arge

t zo

ne

the

resp

onse

of

the

exch

ange

rat

e to

the

fund

amen

tal w

ill b

e af

fect

ed r

elat

ive t

o th

e ca

se o

f an

exo

geno

us

bubb

le (

and,

a—

fort

iori

, rel

ativ

e to

the

no—

bubb

le

case

). T

his

is s

trai

ghtf

orw

ard,

sinc

e

in th

is c

ase

= 1

. A

s lo

ng a

s j —

1,

the

tang

ency

cond

ition

s (b

c—cl

), (

llb)

and

(lld

) are

the

sam

e as

in th

e ca

se o

f an

exo

geno

us b

ubbl

e.

—29

V.

The

Dis

trib

utio

n Fun

ctio

n of

the

Exc

hang

e Rat

e in

a T

arge

t Zon

e in

the

Pres

ence

of a

Rat

iona

l Spe

cula

tive B

ubbl

e

By

Ito'

s lem

ma,

the

exch

ange

rat

e in

the

targ

et z

one

is a

dif

fusi

on p

roce

ss w

ith

stoc

hast

ic di

ffer

entia

l

(35)

ds

(t)

= [g

f(f,.

)p +

gff(

f,.)

4]dt

+ gf

(f,.)

cfdW

f + dB

.

Let

V

artx

(t')

deno

te t

he v

aria

nce

of

x(t')

con

ditio

nal

on t

he in

form

atio

n av

aila

ble

at t

ime

t, th

at i

s V

artx

(t')

z E

t[x(

t') —

E

tx(t

')]2.

T

he in

stan

tane

ous

varia

nce o

f the

bub

ble

proc

ess

is g

iven

by

o(f,

f, s

, B

, .)

S

ince

the

only

res

tric

tion

on t

he b

ubbl

e pr

oces

s is

tha

t E

tdB

= nB

dt,

it is

cer

tain

ly p

erm

issi

ble

for

the

inst

anta

neou

s va

rian

ce o

f th

e bu

bble

pr

oces

s to

dep

end

on

f, f

, s,

B

or

oth

cr

vari

able

s.

As

befo

re, r

ecog

nizi

ng t

hat

B m

ay d

epen

d on

f,

we

wri

te

- .

,9z

B=

b(f,

z) w

ith

Usi

ng s

elf—

expl

anat

ory

nota

tion w

e ob

tain

(36a

) E

ds(t

) =

(gf+

bf)p

+ b

Etd

z(t)

+ {(

gff+

bff)

o +

bzzU

+ 2b

zfPf

rfff

Cz]

dl

and

(36b

) V

artd

s(t)

= [(

gf+

bf)2

o + b

a + 2

(gf +

bf)b

zPfz

afcz

] dt

.

For

a bu

bble

pr

oces

s to

be

cons

iste

nt w

ith a

targ

et z

one

unde

r th

e sp

ecif

ied

inte

rven

tion

rule

, it

mus

t be

th

e ca

se t

hat,

for

any

size

bu

bble

, th

e bo

unda

ry

—30

cond

ition

s ch

arac

teri

ze

tang

ency

po

ints

tha

t ar

e lo

cal

max

ima

(min

ima)

if

the

tang

ency

is

at t

he u

pper

(lo

wer

) bo

unda

ry.

For

a bu

bble

tha

t is

con

sist

ent w

ith t

he

law

of m

otio

n in

(4)

but d

oes

not p

erm

it sm

ooth

pas

ting

acco

rdin

g to

(lo

ad),

we r

efer

to th

e bu

bble

in

(18a

,b)

with

k =

—1

or th

e bu

bble

giv

en in

(24)

and

(25

).

Not

e th

at t

he i

nsta

ntan

eous

cond

ition

al v

aria

nce

of th

e ch

ange

in

the

exch

ange

rate

no

long

er

in g

ener

al

goes

to

zero

as

the

boun

dari

es

of t

he t

arge

t zo

ne

are

appr

oach

ed.

As

the

exch

ange

ra

te

appr

oach

es

the

boun

dary

, the

var

ianc

e of

the

exog

enou

s co

mpo

nent

of

the

bubb

le n

eed

not g

o to

zer

o no

r nee

d gf

+ bf

tend

to z

ero.

If,

in t

he

spir

it of

O

bstf

eld—

Rog

off

and

Dib

a—G

rnss

man

, w

e ac

cept

th

e

prop

ositi

on th

at u

nder

fre

e fl

oatin

g bu

bble

s ca

nnot

occ

ur,

the

cnnd

ition

al

expe

ctat

ion

and

vari

ance

of th

e ch

ange

in

the

exch

ange

rat

e w

ith a

free

flo

at a

re (

letti

ng s

den

ote

the

exch

ange

rat

e un

der a

free

flo

at):

Ft

=

and

ds(t

2

Var

1 --

-- =

Triv

ially

, sin

ce b

ubbl

es c

an e

xist

in

a ta

rget

zon

e, t

he in

stan

tane

ous c

ondi

tiona

l

vari

ance

of

chan

ges

in t

he e

xcha

nge

rate

may

be

less

und

er fr

ee f

loat

ing

than

und

er a

targ

et z

one.

If

unco

vere

d in

tere

st p

arity

hol

ds, it

mus

t be

the

case

tha

t th

e in

tere

st

diff

eren

tial

S is

giv

en b

y

8(t)

e i(

t) —

i*(t

) = F

t ds

(t)/

dt.

It f

ollo

ws

that

the

inst

anta

neou

s con

ditio

nal

mea

n an

d va

rian

ce of

the

chan

ge in

the i

nter

est d

iffe

rent

ial in

the

targ

et z

one

are

give

n by

— 3

1 —

(37a

) E

t4 =

o{(g

f+bf

-1)P

+b5

Ftd

z +

(37b

) V

art4

.l = [o

(gf+

bç—

1)]2

a +

(Obz

)2O

+ 2o

2(gf

+bf

—1)

bpfc

fc5.

As noted in Svensson [1989], if

= 0

(a

nd therefore also fz = 0

) eq

uatio

ns

(36b

) and (37b) imply that (denoting the conditional standard deviation by SD)

ds

—1

dS

SDt() —

SD

t() =

Cf.

Thi

s fi

ndin

g of

a c

onst

ant t

rade

—of

f be

twee

n th

e in

stan

tane

ous v

aria

bilit

y of

the

chan

ge i

n th

e ex

chan

ge

rate

and

the

ins

tant

aneo

us

vari

abili

ty o

f the

cba

nge

in t

he

inte

rest

dif

fere

ntia

l is

no l

onge

r au

tom

atic

ally

valid

whe

n th

ere

is a

bub

ble,

as

can

be

veri

fied

by

insp

ectio

n of

(36b

) an

d (3

7b).

The

co

nditi

onal

in

stan

tane

ous

mea

n an

d va

rian

ce of

the

chan

ge i

n th

e in

tere

st

diff

eren

tial u

nder

a fr

ee f

loat

, dS

, ar

e: Etd

b(t)

= 0

and

Var

tds(

t) =

0.

With

or w

ithou

t a b

ubbl

e, t

he fr

ee f

loat

the

refo

re a

lway

s de

liver

s m

ore

stab

ility

in c

hang

es i

n th

e in

tere

st d

iffe

rent

ial

than

doe

s th

e ta

rget

zon

e.

The

rel

atio

nshi

p be

twee

n th

e le

vel

of t

he e

xcha

nge

rate

an

d th

e in

tere

st

diff

eren

tial in

the

targ

et z

one

is f

ound

by

notin

g th

at

95

lB

= o(

gf +

1)

—32

and

DB

=

f +

It fo

llow

s th

at

Os

gf+

-3-f

-

atgf

+

Whe

n the

re is

no

bubb

le a

nd t

he ra

nge o

f pe

rmis

sibl

e va

lues

of

f is

f1 �

f � 1

(t

hat i

s th

e va

lues

of

f co

rres

pond

ing

to t

he u

pwar

d—sl

opin

g par

t of

the

bubb

le e

ater

for

B =

0)

we

have

, sin

ce 0

� gf

< 1

fo

r f1

< f <

f0

, a

dow

nwar

d—sl

opin

g re

latio

n-

ship

bet

wee

n th

e le

vel of

the

exch

ange

rat

e an

d th

e in

tere

st d

iffe

rent

ial.

Thi

s re

latio

n-

ship

is

qu

alita

tivsl

y un

affe

cted

by

the

bubb

le i

f th

e cu

rren

t va

lue

of th

e bu

bble

is

inde

pend

ent o

f th

e cu

rren

t val

ue o

f th

e fu

ndam

enta

l, if

we

agai

n re

stri

ct th

e an

alys

is t

o

the

upw

ard—

slop

ing

part

of

the

h(f,

B(r

), B

(t))

= g

(f)

+ B

(t)

func

tion.

T

here

are

,

how

ever

, no

goo

d re

ason

s fo

r th

is

rest

rict

ion

whe

n th

ere

are

bubb

les,

so

the

rela

tions

hip

betw

een

the

leve

l of

the

exch

ange

rat

e an

d th

e in

tere

st d

iffe

rent

ial c

an

chan

ge s

ign.

T

his

is s

how

n in

Fig

ure

2 fo

r an e

xoge

nous

bub

ble.

Har

riso

n [1

985,

pp.

89—

92]

show

s th

at t

he s

tead

y st

ate

or u

ncon

ditio

nal d

ensi

ty

func

tion

ir

of B

row

nian

mot

ion

f w

ith i

nsta

ntan

eous

var

ianc

e cr

an

d dr

ift

p w

hich

is re

gula

ted

on th

e cl

ose

inte

rval

[f1,

is

giv

en by

:

71ff

) =

[I0 —

f1]

if

p =

0 (t

he u

nifo

rm d

ensi

ty)

= ee

if

p

0 (t

runc

ated

exp

onen

tial d

ensi

ty)

— 3

3 —

-2

0 =

2/w

f

With

out a

bub

ble

it fo

llow

s th

at, s

ince

s

= g

(f)

is a

str

ictly

incr

easi

ng fu

nctio

n

of

f w

ith c

ontin

uous

fi

rst

deri

vativ

e on

the

open

int

erva

l f0

) an

d th

e im

plic

it

(inv

erse

) fu

nctio

n f =

g(s

) th

eref

ore

exis

ts o

n th

e ta

rget

zon

e (e

xcep

t at

the

end

poin

ts),

the

stea

dy s

tate

den

sity

func

tion

of s,

cc(

s)

say,

is g

iven

by:

(5)7

r( j

s))

Sj<

5<5

g'(f

(s))

u

Sven

sson

[1

989]

sho

ws

that

th

e re

sulti

ng d

istr

ibut

ion

of e

xcha

nge

rate

s is

U—

shap

ed.

With

a b

ubbl

e, h

owev

er,

the

long

—ru

n be

havi

or o

f the

exc

hang

e ra

te i

s qu

ite

diff

eren

t. Fi

rst

of a

ll, o

ne o

f the

sta

te v

aria

bles

, th

e bu

bble

, is

non—

stat

iona

ry a

nd

unre

gula

ted.

It

ther

efor

e do

es n

ot p

osse

ss a

ste

ady

stat

e di

stri

butio

n. T

he o

ther

sta

te

vari

able

, f,

is

of c

ours

e al

so n

on—

stat

iona

ry w

hen

unre

gula

ted,

bu

t re

gula

tion

may

st

ill e

nsur

e th

at it

has

a s

tead

y st

ate

dist

ribu

tion.

Not

e, h

owev

er t

hat

boun

ds s

et b

y

the

cont

rolle

r are

bou

nds

on s

, no

t on

f.

It is

cle

arly

pos

sibl

e fo

r f

to b

ecom

e a

nega

tive

num

ber s

mal

ler

than

the

val

ue of

f1 as

soci

ated

with

B =

U

(for

inst

ance

if a

posi

tive

bubb

le g

row

s st

eadi

ly a

nd t

here

is a

long

seq

uenc

e of

, on

ave

rage

, ne

gativ

e re

aliz

atio

ns o

f df

with

in th

e ta

rget

zon

e).

Sim

ilarl

y, f

can

bec

ome

a po

sitiv

e num

ber

larg

er th

an t

he v

alue

of f0

asso

ciat

ed w

ith

B =

0

if a

neg

ativ

e bu

bble

ste

adily

gro

ws

in s

ize

and

ther

e is

a lo

ng r

un o

f, o

n av

erag

e, p

ositi

ve re

aliz

atio

ns o

f df

.8

Whi

le th

e de

riva

tion

of th

e st

eady

-sta

te d

istr

ibut

ion o

f s

for m

ost i

nter

estin

g ca

ses

(inc

ludi

ng

the

exog

enou

s st

ocha

stic

bub

ble

and

the

stoc

hast

ic b

ubbl

e th

at

depe

nds

on

f) r

emai

ns to

be

done

, a

few

gen

eral

rem

arks

can

be

mad

e.

Con

side

r fi

rst t

he ca

se o

f an

exog

enou

s, n

on-s

toch

astic

bubb

le.

The

asy

mpt

otic

—34

)

dist

ribu

tion

of th

e ex

chan

ge r

ate

will

, if

the

bubb

le i

s po

sitiv

e,

put

all

its p

roba

bilit

y

mas

s ne

ar i

ts

uppe

r bo

unda

ry

valu

e s.

In t

his

very

sim

ple

case

, th

e ta

rget

zon

e

does

n't

kill

the

bubb

le, b

ut t

he b

ubbl

e ef

fect

ivel

y ki

lls t

he t

arge

t zo

ne a

nd t

urns

it,

asym

ptot

ical

ly, in

to a

fixe

d ex

chan

ge r

ate

regi

me.

Now

con

side

r the

cas

e w

here

the

bubb

le i

s st

ocha

stic

. If

we

have

the

exo

geno

us

geom

etri

c bub

ble

dB =

caB

dt +

cTB

BdW

B,

tbe

bubb

le c

anno

t cha

nge

sign

, al

thou

gh it

can

chan

ge d

irec

tion.

T

ake

the

case

with

B

(O)

> 0

. If

the

bub

ble

ever

rea

ches

B =

0, it

dies

and

we

are

back

hen

cefo

rth

in th

e ca

se s

tudi

ed b

y Sv

enss

oo.

Cle

arly

we

expe

ct t

o ha

ve a

non—

dege

nera

te d

istr

ibut

ion

of

s in

thi

s ca

se,

but

still

with

mor

e

mas

s at

s th

an w

ithou

t a b

ubbl

e.

u

If t

he e

xoge

nous

bu

bble

pr

oces

s is

giv

en in

stea

d by

dB

= o

Bdt

+ C

BdW

B

with

>

0

but c

onst

ant,

B

will

, in

a re

pres

enta

tive h

isto

ry, a

ssum

e bo

th p

ositi

ve

and

nega

tive

valu

es.

The

bub

ble

neve

r die

s in

this

cas

e.

Rel

ativ

e to

the

case

stu

died

by S

vens

son

we

wou

ld e

xpec

t the

lon

g—ru

n de

nsity

fun

ctio

n to

hav

e m

ore

mas

s at

the

two

end

poin

ts b

ut w

ith n

o pa

rtic

ular

bia

s to

war

ds ei

ther

s0

or s

. A

sym

ptot

ical

ly,

the

infl

uenc

e of

the

initi

al c

ondi

tions

fo

r B

on

the

den

sity

func

tion

for

s sh

ould

vani

sh.

We

hope

to b

ack

up th

ese

intu

ition

s with

sol

id a

naly

sis

in th

e ne

ar fu

ture

.

VT

. Bub

bles

, tar

get z

ones

an

d in

terv

entio

n rul

es: w

hat g

ives

whe

n th

ey a

re in

cons

iste

nt?

Our

vie

w o

f wha

t hap

pens

to

spec

ulat

ive

bubb

les

unde

r a c

redi

ble

targ

et z

one

is

rath

er d

iffe

rent

fro

m th

at o

f M

iller

, W

elle

r an

d W

illia

mso

n [198

91 (

henc

efor

th M

WW

).

The

for

mal

ana

lysi

s in

the

ir p

aper

is c

ast

in t

erm

s of

Bla

ncha

rd b

ubbl

es,

but

wou

ld

appl

y a—

fort

iori

to

the

non

—co

llaps

ing

bubb

les

we

cons

ider

ed.

MW

W p

ut

thei

r

argu

men

t su

ccin

ctly

:

If al

l mar

ket p

artic

ipan

ts be

lieve

that

, whe

n th

e ex

chan

ge ra

te

hits

the

edg

e of

the

ban

d, t

he a

utho

ritie

s will

def

end

the

rate

bp

sud

den

inte

rven

tion

(des

igne

d to

pro

duce

a

jum

p in

the

ra

te),

th

is m

ust

caus

e th

e bu

bble

to

bur

st.

But

thi

s is

—35

inco

nsis

tent

with

the

exi

sten

ce o

f a b

ubbl

e in

the

fir

st p

lace

: If

all k

now

this

col

laps

e is

cer

tain

to o

ccur

at t

ime

t, al

l w

ish

to s

ell t

he c

urre

ncy a

t t—

e. B

ut th

en c

olla

pse

will

occ

ur at

t—e.

R

ep ea

ting

this

arg

umen

t we

find

that

col

laps

e m

ust

occu

r at

time

zero

that

is, a

ll su

ch bu

bble

s ar

e "s

tran

gled

at b

irth

".

(Mill

er,

Wel

ler

and

Will

iam

son

1g8g

]).

It sh

ould

be

reco

gniz

ed

that

in

this

pap

er w

e ha

ve

chan

ged

the

inte

rven

tion

rule

s fo

llow

ed b

y th

e au

thor

ities

from

wha

t was

ass

umed

by

MW

W a

nd b

y th

e ot

her

cont

ribu

tors

to th

is li

tera

ture

. In

add

ition

to th

e st

anda

rd re

flec

tion

polic

ies,

we

allo

w

the

acco

mm

odat

ion

by

the

auth

oriti

es o

f su

stai

ning

sp

ecul

ativ

e at

tack

s.

The

inte

rven

tion

polic

ies

cons

ider

ed by

MW

W an

d in

the

rest

of t

his l

itera

ture

, do

not h

ave

this

fea

ture

. It

sho

uld

ther

efor

e co

me

as n

o su

rpri

se t

hat

from

dif

fere

nt a

ssum

ptio

ns

we

reac

h di

ffer

ent

conc

lusi

ons.

Giv

en th

e as

sum

ptio

n th

at t

he o

nly

form

of

inte

rven

tion

is r

egul

atio

n at

the

boun

dari

es

thro

ugh

infi

nite

sim

al r

efle

ctin

g op

erat

ions

, th

e co

nclu

sion

th

at

this

conf

igur

atio

n is

inc

onsi

sten

t w

ith r

atio

nal b

ubbl

es f

ollo

ws.

O

ne c

omm

on re

spon

se t

o

this

inc

onsi

sten

cy i

s to

trea

t the

bub

ble

as a

resi

duaL

T

he re

ason

ing

goes

as

follo

ws:

at

*1

a po

int l

ike

LI

in F

igur

e la

ther

e w

ould

be

a di

scon

tinuo

us

antic

ipat

ed fa

ll in

the

va

lue

of s

fr

om

5u

to

;, th

e va

lue

of t

he e

xcha

nge

rate

at

f = u

' on t

he b

ubbl

e

eate

r fo

r B

= 0,

gi

ven

by

s =

g(f

, B

0) +

B0,

int

erpr

etin

g B

° to

equ

al z

ero.

T

his

viol

ates

the

no—

arbi

trag

e co

nditi

on.

Wha

t is

pres

ente

d in

the

pre

viou

s pa

ragr

aph

as a

"re

al—

time"

eve

nt, t

he fa

ll of

th

e ex

chan

ge r

ate

from

5u to

s,

can

in fa

ct b

e no

mor

e th

an th

e de

sign

atio

n of

s as

the

only

val

ue o

f th

e ex

chan

ge

rate

rea

ched

whe

n =

tu ,

tha

t is

con

sist

ent w

ith t

he

inte

rven

tion

rule

. E

quiv

alen

tly, B

a 0

is t

he o

nly

valu

e of

the

bubb

le c

onsi

sten

t with

*1

th

e in

terv

entio

n ru

le.

If th

e sy

stem

real

ly w

ere

(som

ehow

) at

12

, th

ere

wou

ld b

e an

inco

nsis

tenc

y.

We

cann

ot d

educ

e ho

w t

he e

xcha

nge

rate

wou

ld b

ehav

e st

artin

g fr

om

an in

cons

iste

nt po

sitio

n.

Not

e th

at i

t is

not j

ust

the

assu

mpt

ion

of a

cre

dibl

e ex

chan

ge r

ate

targ

et z

one

—36

that

pr

oduc

es th

e in

cons

iste

ncy,

bu

t th

e co

mbi

natio

n of

a c

redi

ble

targ

et z

one

and

a

part

icul

ar in

terv

entio

n ru

le t

hat

can

only

sup

port

the

targ

et z

one

if n

o bu

bble

exi

sts.

Our

alte

rnat

ive

inte

rven

tion

rule

, w

hich

has

a d

iscr

ete

(sto

ck—

shif

t) in

crea

se in

f

at

*1

1 *1

Il

fr

om f1

1 to

f11

and

a co

rres

pond

ing

mov

emen

t of

the

equi

libri

um fr

om

11

to

fli,

cons

iste

ntly

com

bine

s a

cred

ible

tar

get z

one

and

ratio

nal b

ubbl

es.

An

obvi

ous

ques

tion

wou

ld s

eem

to

be:

Why

wou

ld t

he a

utho

ritie

s ad

opt a

rule

that

per

mits

the

pres

ence

of

bubb

les

rath

er th

an a

rule

that

pre

clud

es t

heir

exi

sten

ce?

Putti

ng t

he q

uest

ion

this

way

pr

ejud

ges

the

ausw

er,

beca

use

it as

sum

es t

hat,

give

n an

inc

onsi

sten

cy b

etw

een

havi

ng a

cred

ible

tar

get z

one,

a ra

tiona

l bub

ble

and

a

part

icul

ar i

nter

vent

ion

polic

y,

the

inco

nsis

tenc

y m

ust

be re

solv

ed

by d

ropp

ing

the

bubb

le,

rath

er th

an b

y dr

oppi

ng th

e ta

rget

zon

e or

the

inte

rven

tion

polic

y.

We

pref

er

to v

iew

the

exis

tenc

e of

a b

ubbl

e as

a

: A

bub

ble

eith

er e

xist

s or

it

does

n't,

and

the

auth

oriti

es m

ust

desi

gn

thei

r in

terv

entio

n po

licy

acco

rdin

gly,

if

the

y w

ish

to

supp

ort t

he ta

rget

zon

e.

A p

artic

ular

inte

rven

tion

rule

(e.

g. i

nfin

itesi

mal

refl

ectin

g in

terv

entio

ns

at th

e

boun

dari

es

with

out

acco

mm

odat

ion

of f

rien

dly

spec

ulat

ive

atta

cks)

may

of

cour

se b

e

inco

nsis

tent

with

hav

ing

a cr

edib

le

targ

et z

one

in t

he p

rese

nce

of b

ubbl

es.

Thi

s

impl

ies,

in

our

view

, a

colla

pse

eith

er o

f th

e ta

rget

zon

e or

of

the

inte

rven

tion r

ule,

but

not

of t

he b

ubbl

e,

whi

ch i

s no

t an

obj

ect

of c

hoic

e at

som

e in

itial

dat

e, n

ot e

ven

a

colle

ctiv

e on

e.

Giv

en th

e ex

iste

nce

of a

bub

ble,

a cr

edib

le t

arge

t zon

e re

quir

es a

polic

y

rule

spe

cifi

catio

n th

at p

erm

its th

e bu

bble

and

the

targ

et z

one

to c

oexi

st.

Of c

ours

e it

mus

t als

o be

cap

able

of h

andl

ing t

he n

o—bu

bble

case

.

Our

rul

e is

an

exam

ple

of s

uch

a co

nsis

tent

polic

y.9

VII

. Con

clus

ions

The

the

ory

of e

xcha

nge

rate

beh

avio

r w

ithin

a t

arge

t zon

e as

dev

elop

ed i

n th

e

—37

rece

nt

liter

atur

e ho

lds

that

exc

hang

e ra

tes

unde

r a

curr

ency

ba

nd r

egim

e ar

e le

ss

resp

onsi

ve t

o fu

ndam

enta

l sho

cks

than

exc

hang

e ra

tes

unde

r fre

e fl

oat,

prov

ided

that

the

inte

rven

tion

rule

s of

the

Cen

tral

Ban

k(s)

are

com

mon

kno

wle

dge.

M

oreo

ver,

the

re

alw

ays

exis

ts a

tra

de—

off

betw

een

the

inst

anta

neou

s vo

latil

ity o

f ch

ange

s in

the

exch

ange

rat

es a

nd c

hang

es i

n th

e in

tere

st r

ate

diff

eren

tial,

inde

pend

ently

of th

e si

ze o

f

the

band

and

the d

egre

e of

cred

ibili

ty of

the

targ

et z

one.

The

se r

esul

ts a

re d

eriv

ed a

fter

havi

ng

assu

med

a

prio

ri th

at

"rat

iona

l ex

cess

vol

atili

ty"

(due

to

so

calle

d ra

tiona

l

bubb

les)

doe

s no

t occ

ur in

the

for

eign

exc

hang

e m

arke

t. Im

plic

itly

or e

xplic

itly,

it

is

assu

med

th

at s

pecu

lativ

e bu

bble

s ar

e in

com

patib

le

with

the

exi

sten

ce a

nd t

he

pers

iste

nce o

f a c

redi

ble t

arge

t zon

e, s

o th

at th

ey n

ever

mat

eria

lize.

We

cons

ider

inst

ead

a se

tup

in w

hich

the

exis

tenc

e of

spe

cula

tive

beha

vior

is

a

datu

m t

he C

entr

al B

ank

has

to d

eal

with

. W

e sh

ow t

hat

ther

e is

no

inco

mpa

tibili

ty

betw

een

the

exis

tenc

e of a

tar

get z

one

and

the

pres

ence

of

ratio

nal b

ubbl

es.

Rat

her,

th

ere

are

inte

rven

tion

rule

s th

at

shou

ld

be

follo

wed

by

th

e C

entr

al B

ank

whe

n

spec

ulat

ive

bubb

les

aris

e, a

nd th

ese

sam

e ru

les

incl

ude

as a

spec

ial

case

the

trad

ition

al

polic

ies

for d

efen

ding

an

exch

ange

rat

e ba

nd w

hen

spec

ulat

ive

bubb

les d

o no

t occ

ur.

In

the

stan

dard

m

odel

, th

e de

fens

e of

a

targ

et

zone

is

gu

aran

teed

by

inte

rmitt

ent

vari

atio

ns

in d

omes

tic c

redi

t (s

ay

thro

ugh

open

m

arke

t op

erat

ions

)

and/

or n

on s

teri

lized

fore

ign

mar

ket

inte

rven

tions

th

at t

ake

plac

e w

hen

the

exch

ange

rate

hits

one

of

the

limits

of

the

band

. For

inst

ance

, whe

n th

e ex

chan

ge r

ate

reac

hes

the

uppe

r bo

unda

ry, t

he s

tock

of f

orei

gn r

eser

ves

(or

the

stoc

k of

dom

estic

cre

dit)

is

redu

ced

in o

rder

to p

reve

nt th

e ex

chan

ge r

ate

from

ral

sing

fur

ther

. H

euri

stic

ally

, th

ese

oper

atio

ns

can

be c

hara

cter

ized

as i

nfin

itesi

mal

re

flec

ting

inte

rven

tions

. A

lthou

gh th

e

size

of

th

e re

flec

ting

oper

atio

ns

may

be

qua

ntita

tivel

y lim

ited,

th

e in

duce

d

expe

ctat

ions

stab

ilize

the

exch

ange

rat

e ev

en b

efor

e th

e up

per

or lo

wer

lim

it is

reac

hed.

We

show

tha

t in

the

cum

bub

ble

setu

p re

flec

ting

inte

rven

tions

are

insu

ffic

ient

.

In

fact

, w

hen

spec

ulat

ive

bubb

les

aris

e,

the

com

pani

on

phen

omen

a of

spe

cula

tive

— 3

8 —

atta

cks

on t

he t

arge

t zo

ne

mus

t oc

cur

as

wel

l. T

hese

sp

ecul

ativ

e at

tack

s ar

e

stoc

k—sh

ift r

eshu

ffle

s of

pri

vate

age

nts'

fina

ncia

l po

rtfo

lios

led

by ra

tiona

l exp

ecta

tions

of ch

ange

s in

the

rat

e of

exc

hang

e ra

te d

epre

ciat

ion.

A

s an

exa

mpl

e, if

the

rate

of

appr

ecia

tion

of th

e ex

chan

ge r

ate

wer

e su

dden

ly e

xpec

ted

to i

ncre

ase,

rat

iona

l pri

vate

agen

ts w

ould

in

crea

se

thei

r de

man

d fo

r ho

me

coun

try

mon

ey

and/

or d

ecre

ase

thei

r

dem

and

for f

orei

gn c

ount

ry m

oney

due

to

a de

crea

se i

n th

e do

mes

tic—

fore

ign

inte

rest

rate

dif

fere

ntia

l.

We

show

tha

t th

e de

fens

e of

the

targ

et z

one

in t

he p

rese

nce

of b

ubbl

es is

guar

ante

ed if

the

Cen

tral

Ban

k ac

com

mod

ates

spe

cula

tive

atta

cks

wbe

n th

e la

tter

are

frie

ndly

, th

at is

whe

n th

ey a

re c

onsi

sten

t with

the

sur

viva

l of

the

targ

et z

one

itsel

f

(gtv

en

the

Cen

tral

Ban

k's

role

).

Thi

s in

tuiti

ve p

olic

y ru

le

is c

ompa

tible

with

self

—fu

lfill

ing e

xpec

tatio

ns: a

fri

endl

y at

tack

occ

urs

only

if a

gent

s kno

w th

at i

t will

be

acco

mm

odat

ed,

and

exac

tly b

ecau

se

of t

he p

assi

ve

acco

mm

odat

ion

the

spec

ulat

ive

atta

ck s

nsta

in8

the

targ

et z

one.

H

ence

, th

e po

licy

rule

is

sum

mar

ized

by

the

max

im

"Ref

lect

w

hen

this

is s

uffi

cien

t; ac

com

mod

ate

whe

n th

is is

nec

essa

ry".

We

show

tha

t w

hen

the

exch

ange

ra

te h

its o

ne o

f th

e bo

unda

ries

w

ith

a

non—

zero

bub

ble,

an

acco

mm

odat

ed

frie

ndly

spe

cula

tive

atta

ck o

ccur

s, f

ollo

wed

by

refl

ectin

g in

terv

entio

ns.

Man

y of

the

con

clus

ions

re

ache

d in

the

exi

stin

g lit

erat

ure

do

not

appe

ar

suff

icie

ntly

rob

ust w

hen

spec

ulat

ive

bubb

les

are

cons

ider

ed a

s w

ell.

Firs

t, it

is n

ot tr

ue

anym

ore

that

the

ins

tant

aneo

us v

olat

ility

of

exch

ange

ra

tes

with

in a

tar

get

zone

is

alw

ays

less

tha

n th

e in

stan

tane

ous v

olat

ility

of e

xcha

nge

rate

und

er fr

ee f

loat

, pro

vide

d

that

, for

fam

iliar

reas

ons,

in

free

flo

at n

o sp

ecul

ativ

e bu

bble

ari

ses.

Se

cond

, the

find

ing

of a

con

stan

t tr

ade—

off

betw

een

the

inst

anta

neou

s va

riab

ility

of

the

chan

ge i

n th

e

exch

ange

ra

te a

nd t

he i

nsta

ntan

eous

var

iabi

lity

of t

he c

hang

e in

the

inte

rest

rat

e p

diff

eren

tial

is n

o lo

nger

auto

mat

ical

ly va

lid w

hen

ther

e is

a b

ubbl

e.

Mor

eove

r, t

he s

tand

ard

theo

ry c

hara

cter

izes

a s

tabl

e rel

atio

n be

twee

n th

e le

vel

—39

of th

e ex

chan

ge r

ate

and

the

expe

cted

rate

of

depr

ecia

tion

(equ

al to

the

int

eres

t rat

e

diff

eren

tial

if u

ncov

ered

int

eres

t par

ity ho

lds)

. A

ccor

ding

to

this

mod

el,

the

high

er th

e

exch

ange

ra

te,

the

low

er

the

inte

rest

ra

te d

iffe

rent

ial.

Mor

eove

r th

e in

tere

st r

ate

diff

eren

tial

is a

lway

s ne

gativ

e in

the

nei

ghbo

rhoo

d of

the

uppe

r bo

unda

ry o

f th

e

exch

ange

ra

te

band

; th

e op

posi

te r

esul

t ho

lds

in t

he n

eigh

borh

ood

of t

he l

ower

boun

dary

. W

e sh

ow th

at t

his

rela

tion

is n

o lo

nger

sta

ble

whe

n bu

bble

s ari

se.

—40

I

NO

TE

S

'Con

side

r a

dyna

mic

lin

ear

ratio

nal

expe

ctat

ions

m

odel

w

ith

cons

tant

coef

fici

ents

th

at h

as

the

usua

l sa

ddle

poin

t co

nfig

urat

ion

(as

man

y pr

edet

erm

ined

vari

able

s, n

1 sa

y,

as

stab

le c

hara

cter

istic

root

s an

d as

m

any

non—

pred

eter

min

ed

vari

able

s, n

2 sa

y, a

s un

stab

le c

hara

cter

istic

ro

ots)

. T

rans

form

the

syst

em to

can

onic

al

vari

able

s, b

y di

agon

aliz

ing

it or

by

usin

g Jo

rdan

's c

anon

ical

for

m.

Gro

up t

oget

her t

he

n2

dyna

mic

equ

atio

ns

cont

aini

ng

the

cano

nica

l va

riab

les

who

se

hom

ogen

eous

equa

tions

are

gov

erne

d by

th

e un

stab

le r

oots

. T

he

gene

ral

solu

tion

for

thes

e

non—

pred

eter

min

ed

cano

nica

l va

riab

les

can

cont

ain

an n2

—di

men

sion

al bu

bble

pro

cess

that

mus

t sa

tisfy

the

hom

ogen

eous

eq

uatio

n sy

stem

of

the

n2 n

on—

pred

eter

min

ed

cano

nica

l va

riab

les.

B

ubbl

e pr

oces

ses

will

th

eref

ore

be n

on—

stat

iona

ry

(in

expe

ctat

ion)

. T

he s

tate

var

iabl

es o

f th

e m

odel

, w

hich

are

linea

r co

mbi

natio

ns

of t

he

cano

nica

l va

riab

les,

will

, if

ther

e is

a

bubb

le,

be n

on—

stat

iona

ry.

If th

ere

are

mor

e

non—

pred

eter

min

ed

stat

e va

riab

les

than

uns

tabl

e ro

ots,

st

atio

nary

bub

bles

w

ill

of

cour

se b

e po

ssib

le e

ven

in li

near

mod

els.

2Sis

nila

r ar

gum

ents

ca

n be

m

ade

for

non—

stat

iona

ry

bubb

les

(and

th

e

non—

stat

iona

ry

beha

vior

of

ke

y en

doge

nous

va

riab

les

freq

uent

ly a

ssoc

iate

d w

ith

non—

stat

iona

ry

bubb

les)

in

cert

ain

non—

linea

r mod

els.

Fo

r in

stan

ce, i

n ov

erla

ppin

g

gene

ratio

ns

(OL

G)

mod

els

of a

co

mpe

titiv

e ec

onom

y w

ith

a si

ngle

pe

rish

able

com

mod

ity

and

with

no

n—in

tere

st—

bear

ing o

utsi

de m

oney

as

the

only

sto

re o

f val

ue,

ratio

nal d

efla

tiona

ry b

ubbl

es (

i.e.

non—

stat

iona

ry b

ubbl

es w

ith th

e pr

ice l

evel

dec

linin

g

to z

ero)

hav

e be

en s

how

n to

be

infe

asib

le.

Tak

e for

inst

ance

the

case

of

a tw

o—pe

riod

OL

G m

odel

with

a c

onst

ant p

opul

atio

n in

whi

ch th

e no

min

al m

oney

sto

ck is

con

stan

t,

—41

all m

oney

is

held

by

the

old

and

only

the

you

ng r

ecei

ve a

con

stan

t bou

nded

end

owm

ent

stre

am o

f th

e si

ngle

com

mod

ity. I

f the

pri

ce le

vel

wer

e to

fall

with

out b

ound

, rea

l cas

h

bala

nces

wou

ld b

e in

crea

sing

with

out b

ound

and

the

dem

and f

or th

e co

mm

odity

by t

he

old

gene

ratio

n w

ould

incr

ease

with

out b

ound

. Si

nce

the

supp

ly of

the

good

eac

h pe

riod

is b

ound

ed a

bove

by

the

sum

of

the

end

owm

ents

of

the

you

ng,

soon

er

or l

ater

the

dem

and

for

the

com

mod

ity m

ust

outs

trip

the

sup

ply.

A

seq

uenc

e of

mon

ey

pric

es

falli

ng t

o ze

ro c

an th

eref

ore

not

be r

atio

nal e

xpec

tatio

ns

equi

libri

um (s

ee e

.g.

Hah

n

[198

2, p.

10]

).

Unl

ess

very

st

rong

re

stri

ctio

ns a

re p

lace

d on

tbe

pri

vate

util

ity f

unct

ions

,

ratio

nal i

nfla

tiona

ry b

ubbl

es (w

ith t

he p

rice

lev

el i

ncre

asin

g w

ithou

t bou

nd d

espi

te a

cons

tant

nom

inal

mon

ey s

tock

) can

exi

st in

suc

h an

eco

nom

y.

The

seq

uenc

e of

risi

ng

pric

es w

ould

, in

the

limit,

dri

ve t

he r

eal v

alue

of m

oney

to

zero

. T

he s

tead

y st

ate

to

whi

ch s

uch

a m

odel

co

nver

ges

is t

hat

of a

non

—m

onet

ary

econ

omy.

T

o ru

le

out

infl

atio

nary

bub

bles

as

w

ell,

Obs

tfel

d an

d R

ogof

f [1

983]

, in

an

inf

inite

—liv

ed

repr

esen

tativ

e ag

ent

mod

el,

impo

sed

a po

litic

al—

tech

nolo

gica

l re

stri

ctio

n on

th

e

term

inal

val

ue

of m

oney

: th

e go

vern

men

t fr

actio

nally

ba

cks

the

curr

en,c

y by

guar

ante

eing

a m

inim

al r

eal r

edem

ptio

n va

lue

for

mon

ey.

Eve

n if

pri

vate

age

nts

are

not

com

plet

ely

cert

ain

that

th

ey c

an r

edee

m t

heir

mon

ey i

n an

y gi

ven

peri

od, t

his

suff

ices

to

rule

out

spe

cula

tive h

yper

infl

atio

ns.

Whi

le th

is a

ssum

ptio

n on

gove

rnm

ent

beha

vior

seem

s qu

ite a

d—ho

c,

it is

oft

en c

ited

as t

he s

econ

d bl

ade

of th

e sc

isso

rs t

hat

cut

the

lifel

ine

of n

on—

stat

iona

ry r

atio

nal

spec

ulat

ive

bubb

les,

bo

th d

efla

tiona

ry

and

infl

atio

nary

. D

iba

and

Gro

ssm

an

[198

8] a

lso

deri

ve s

uffi

cien

t co

nditi

ons

for r

ulin

g ou

t

non—

stat

iona

ry i

nfla

tiona

ry bu

bble

s.

3Sar

gent

an

d W

alla

ce's

"U

nple

asan

t Mon

etar

ist A

rith

met

ic"

mod

el

(Sar

gent

and

Wal

lace

[19

81])

can

exh

ibit

stat

iona

ry r

atio

nal

bubb

les.

E

ven

a fi

rst

orde

r

dete

rmin

istic

non

—lin

ear

diff

eren

ce

equa

tion

may

ex

hibi

t va

riou

s ki

nds

of p

erio

dic

— 4

2 —

solu

tions

or c

haot

ic b

ehav

ior

(see

e.g

. B

enha

bib

and

Day

[19

81,

1982

] an

d G

rand

mon

t

[198

5]).

St

atio

nary

bub

bles

ar

e ea

sily

gen

erat

ed

by s

uch

mod

els

(Aza

riad

is

[198

1],

Aza

riad

is a

nd G

uesn

erie

[1

986]

, C

hiap

pori

an

d G

uesn

erie

[1

988]

, W

oodf

ord

[198

7],

Farm

er a

nd

Woo

dfor

d [1

986]

).

Seco

nd o

rder

det

erm

inis

tic n

on—

linea

r di

ffer

entia

l

equa

tions

can

gen

erat

e lim

it cy

cles

and

thi

rd o

rder

non

—lin

ear

diff

eren

tial

equa

tions

can

exhi

bit

chao

tic b

ehav

ior.

A

gain

, su

ch m

odel

s ca

n su

ppor

t st

atio

nary

rat

iona

l

spec

ulat

ive b

ubbl

es.

4Tiii

s in

terp

reta

tion

of f

com

es f

rom

the

tw

o—co

untr

y m

ini—

mod

el o

utlin

ed

belo

w.

All

vari

able

s exc

ept

inte

rest

rat

es a

re in

nat

ural

log

arith

ms.

m

is

the

hom

e

coun

try

nom

inal

mon

ey s

tock

, p

the

hom

e co

untr

y pr

ice

leve

l, y

hom

e co

untr

y re

al

GD

P, i

th

e ho

me c

ount

ry sh

ort

nom

inal

int

eres

t ra

te a

nd

s th

e sp

ot p

rice

of

fore

ign

exch

ange

. C

orre

spon

ding

for

eign

cou

ntry

var

iabl

es a

re s

tarr

ed.

— p

= k

y —

i, k

, .1

> 0

(H

ome

mon

etar

y eq

uilib

rium

)

* *

* *

— p

= k

y —

X

i (F

orei

gn m

onet

ary

equi

libri

um)

* p

= p

+

(P

urch

asin

g Pow

er P

arity

)

Etd

s(t)

= [i

(t) —

i*(t

)]dt

(U

ncov

ered

Int

eres

t Par

ity)

From

this

ver

y si

mpl

e m

odel

we

obta

in t

he fo

llow

ing

rela

tion:

s(t)

dt =

]m(t

) — m

*(t)

— k(

y(t)

— y*

(t))

]dt +

AE

tds(

t).

Thi

s co

rres

pond

s to

equ

atio

n (1

) w

ith .

1 =

o (the

int

eres

t sem

i—el

astic

ity o

f

—43

* *

mon

ey d

eman

d) a

ndf=

m—

m

—k(

y—y

).

5Equ

atio

n (5

) is

of c

ours

e eq

uiva

lent

to

the

perh

aps

mor

e fa

mili

ar fo

rm

Af(

t)

A 1(

t)

s(t)

=fO

)+B

(t)+

+A

1(t)

e 1

+A

2(t)

e 2

whe

re A

1 an

d A

2 ar

e co

nsta

nt a

s lo

ng a

s s

is in

the

inte

rior

of t

he ta

rget

zon

e, b

ut

can

chan

ge w

hen

s is

at o

ne o

f its

bou

ndar

ies.

Fo

r m

any

bubb

les,

incl

udin

g th

e

exog

enou

s on

es, t

he r

epre

sent

atio

n gi

ven

in (

5) i

s at

trac

tive

beca

use i

t can

bri

ng n

ut

clea

rly

the

hori

zont

al sh

ift

nf t

he b

ubbl

e ea

ter

(def

ined

bel

ow)

whe

n th

e m

agni

tude

of

the

bubb

le v

arie

s.

6Whi

le t

he e

xpec

ted

rate

of

chan

ge o

f s

at th

e re

flec

tion

poin

ts o

n th

e up

per

boun

dary

(po

ints

like

or

flu

) is

al

way

s le

ss t

han

that

at

poin

ts l

ike

111,

the

*1

expe

cted

rate

of c

hang

e at

II

can,

for

ver

y sm

all

bubb

les,

be n

egat

sve.

7The

no

tion

of a

frie

ndly

or

sust

aini

ng

spec

ulat

ive

atta

ck i

s re

late

d to

the

"sus

tain

ing"

mon

ey

dem

and

by a

rbitr

ageu

rs in

one

of

the

solu

tions

to

the

"go

ld

stan

dard

pa

rado

x" p

ropo

sed

by B

uite

r an

d G

rilli

[1

989]

. Se

e al

so

Kru

gman

an

d

Rot

embe

rg 11

9901

.

8The

inf

imum

of f

is

whe

re th

e do

wnw

ard—

slop

ing p

art

of th

e bu

bble

eat

er f

or

Bu

cuts

the

upp

er b

ound

ary.

T

he s

upre

mum

of f

is

whe

re th

e do

wnw

ard—

slop

ing

part

of th

e bu

bble

eat

er fo

r B

1 cu

ts th

e lo

wer

bou

ndar

y.

sepa

rate

arg

umen

t th

at m

ight

be

mad

e ag

ains

t bu

bble

equ

ilibr

ia ta

kes

aim

—44

at it

s m

ost s

trik

ing

feat

ure:

th

e po

ssib

ility

of v

ery

freq

uent

int

erve

ntio

ns at

the

edge

s

of th

e zo

ne.

Con

side

r for

sim

plic

ity

a de

term

inis

tic bu

bble

tha

t st

arts

fro

m a

pos

itive

initi

al

valu

e.

It m

ight

be

argu

ed t

hat

the

inte

rven

tions

in

the

fund

amen

tal th

at a

re re

quir

ed

to o

ffse

t su

ch a

non

—st

atio

nary

bub

ble

are

not s

usta

inab

le be

caus

e, a

s th

e m

agni

tude

of

the

bubb

le g

row

s ov

er t

ime,

inte

rven

tions

at

the

uppe

r bou

ndar

y ca

n be

exp

ecte

d to

beco

me

mor

e an

d m

ore

freq

uent

. Fi

nite

inte

rnat

iona

l res

erve

s (a

ssum

ing

tbes

e ar

e th

e

inte

rven

tion m

ediu

m) a

re b

ound

to b

e ex

haus

ted i

n du

e co

urse

.

It s

houl

d be

not

ed th

at a

very

sim

ilar a

rgum

ent c

an b

e m

ade

even

if th

ere i

s no

bubb

le,

as l

ong

as th

ere

is p

ositi

ve dr

ift i

n th

e fu

ndam

enta

l. W

hile

the

expe

cted

va

lue

of t

he f

unda

men

tal g

row

s lin

earl

y (a

t a c

onst

ant r

ate

p) r

athe

r th

an e

xpon

entia

lly

as

the

bubb

le d

oes,

the

drif

t of

the

unre

gula

ted f

unda

men

tal w

ill a

lso

in fi

nite

tim

e ca

use

any

fini

te s

tock

of r

eser

ves

to b

e ex

haus

ted

with

pro

babi

lity

one.

Ind

eed,

eve

n w

ithou

t

drif

t in

the

fun

dam

enta

l (an

d w

ithou

t a

bubb

le),

a

stoc

hast

ic fu

ndam

enta

l pr

oces

s

driv

en b

y B

row

nian

mot

ion

will

bri

ng a

ny fi

nite

stoc

k of

res

erve

s do

wn

to a

ny p

ositi

ve

low

er b

ound

in fi

nite

tim

e w

ith p

roba

bilit

y one

(se

e B

uite

r [19

89])

.

The

pro

blem

s of

inte

rnat

iona

l res

erve

exh

aust

ion

crea

ted

by t

he b

ubbl

e ar

e of

cour

se

less

se

vere

if

the

bub

ble

is

a B

lanc

hard

bub

ble,

whi

ch h

as f

inite

exp

ecte

d

dura

tion.

Not

e al

so

that

in

the

mos

t co

mm

on

inte

rpre

tatio

n of

ou

r m

odel

, th

e

fund

amen

tal

f st

ands

for

rel

ativ

e ho

me

coun

try

mon

ey s

uppl

y m

inus

rel

ativ

e re

al

inco

me

— r

elat

ed m

oney

dem

and.

Tak

ing

rela

tive n

omin

al m

oney

sto

cks

as t

he o

bjec

t

of r

egul

atio

n, th

ere

is n

othi

ng

in t

he l

ogic

of

the

mod

el t

hat

requ

ires

in

tern

atio

nal

rese

rves

to

be u

sed

to re

gula

te th

e m

oney

sto

cks.

D

omes

tic c

redi

t ex

pans

ion

(whe

ther

refl

ectin

g op

en m

arke

t ope

ratio

ns o

r m

onet

ary

fina

ncin

g of

gov

ernm

ent

budg

et d

efic

its)

can

achi

eve t

he s

ame

mon

etar

y ob

ject

ives

.

—45

RE

FE

RE

NC

ES

Ave

sani

, R

enzo

C.

[199

0],

"End

ogen

ousl

y D

eter

min

ed

Tar

get

Zon

e an

d C

entr

al B

ank

Opt

imal

Pol

icy

With

Lim

ited R

eser

ves"

, m

imeo

, U

nive

rsity

of T

rent

o, J

une.

Aza

riad

is, G

. [1

9811

, "S

elf—

fulf

illin

g Pr

ophe

sies

",

Jour

nal

of E

cono

mic

T

heor

y, 2

5, p

p.

380—

396.

____

____

and

R. G

uesn

erie

[19

86],

"Su

nspo

ts a

nd C

ycle

s",

Rev

iew

of Ec

onom

ic S

tudi

es,

53,

pp. 7

87—

806.

Ben

habi

b, J

. an

d R

. D

ay

[198

1],

Rat

iona

l Cho

ice

and

Err

atic

Beh

avio

ur",

R

evie

w of

E

cono

mic

Stud

ies,

48,

pp.

459

—71

.

____

____

and

___

____

_ [1

982]

, A

Cha

ract

eriz

atio

n of

Err

atic

Dyn

amic

s in

the

Ove

rlap

ping

G

ener

atio

ns M

odel

", Jo

urna

l of E

cono

mic

Dyn

amic

s an

d C

ontr

ol,

, pp.

37—

55.

Ber

tola

, G

iuse

ppe

and

Ric

ardo

J.

Cab

alle

ro

[198

9],

"Tar

get

Zon

es a

nd

Rea

lignm

ents

",

Mim

eo,

Dec

embe

r.

____

____

and

___

____

[19

90],

"Res

erve

s an

d R

ealig

nmen

ts in

a T

arge

t Z

one"

, m

imeo

, Ju

ne.

Bla

ncha

rd,

Oliv

ier

J. [

1979

], "

Spec

ulat

ive

Bub

bles

, C

rash

es a

nd R

atio

nal E

xpec

tatio

ns",

E

cono

mic

Let

ters

, pp.

3 87

—9.

____

____

and

Sta

nley

Fis

cher

[19

89],

L

ectu

res

on M

acro

econ

omic

s, M

IT P

ress

, Cam

brid

ge,

Mas

sach

uset

ts.

Bui

ter,

Will

em II

. [19

89],

"A

Via

ble

Gol

d St

anda

rd R

equi

res

Flex

ible

Mon

etar

y an

d Fi

scal

Po

licy"

, Rev

iew

of E

cono

mic

Stu

dies

, 56

, Jan

uary

, pp.

101

—18

.

____

____

and

Vitt

orio

Gri

lli [

1989

], "

The

'Gol

d St

anda

rd P

arad

ox' a

nd I

ts R

esol

utio

n",

NB

ER

Wor

king

Pa

per N

o. 3

178,

Nov

embe

r.

Chi

appo

ri,

P.

and

R.

Gue

sner

ie

[198

6],

"End

ogen

ous

Fluc

tuat

ions

Und

er

Rat

iona

l E

xpec

tatio

ns",

Eur

opea

n Eco

nom

ic R

evie

w,

32, p

p. 3

89—

97.

Del

gado

, Fr

anci

sco

and

Ber

nard

Dum

as [

1990

], "

Mon

etar

y C

ontr

actin

g B

etw

een

Cen

tral

B

anks

And

The

Des

ign

of S

usta

inab

le

Exc

hang

e—R

ate

Zon

es",

mim

eo,

Mar

ch.

Dib

a, B

. T

. an

d H

. I.

Gro

ssm

an

[198

8],

"Rat

iona

l In

flat

iona

ry

Bub

bles

", J

ourn

al o

f M

onet

ary

Eco

nom

ics,

21,

pp.

35—

46.

Dix

it, A

vina

sh [

1988

], "

A S

impl

ifie

d E

xpos

ition

of

Som

e R

esul

ts

Con

cern

ing

Reg

ulat

ed

Bro

wni

an M

otio

n", W

orki

ng P

aper

, Pri

ncet

on U

nive

rsity

, A

ugus

t.

Dor

nbus

ch,

Rud

iger

[19

76],

"E

xpec

tatio

ns a

nd

Exc

hang

e R

ate

Dyn

amic

s",

Jour

nal

of

Polit

ical

Eco

nom

y, 8

, pp

. 11

61—

76.

46 —

Dum

as, B

erna

rd [1

989]

, Sup

er C

onta

ct a

nd R

elat

ed O

ptim

ality

Con

ditio

ns:

A S

uppl

emen

t to

Avi

nash

Dia

lt's

"A S

impl

ifie

d E

xpos

ition

of

Som

e R

esul

ts

Con

cern

ing

Reg

ulat

ed

Bro

wni

an M

otio

n", N

BE

R T

echn

ical

Wor

king

Pap

er N

o. 7

7, A

pril.

Farm

er, R

. and

M. W

oodf

ord

[198

6], "

Self

—fu

lfill

ing

Prop

hesi

es A

nd T

he B

usin

ess

Cyc

le",

C

AR

ESS

Dis

cuss

ion

Pape

r.

Floo

d, R

ober

t P. a

nd P

eter

M. G

arbe

r [19

89],

"T

he L

inka

ge B

etw

een

Spec

ulat

ive A

ttack

an

d T

arge

t Zon

e M

odel

s of

Exc

hang

e R

ates

", N

BE

R W

orki

ng P

aper

No.

291

8.

Froo

t, K

enne

th A

. an

d M

auri

ce

Obs

tfel

d [1

989a

],

"Sto

chas

tic P

roce

ss

Switc

hing

: So

me

Sim

ple

Solu

tions

", N

BE

R W

orki

ng P

aper

No.

299

8, J

une.

____

____

_ an

d __

____

___

[198

9b[,

"E

xcha

nge R

ate

Dyn

amic

s un

der

Stoc

hast

ic R

egim

e Sh

ifts

: A

Uni

fied

App

roac

h. m

imeo

, Ju

ne.

____

____

_ an

d __

____

__ [1

989c

], "I

ntri

nsic

Bub

bles

: T

he C

ase

of S

tock

Pr

ices

", N

BE

R

Wor

king

Pa

per N

o. 3

091,

Sep

tem

ber.

Gra

ndm

ont,

J. [

1985

], "

On

End

ogen

ous

Com

petit

ive

Bus

ines

s C

ycle

s",

Eco

nom

etri

ca,

53,

pp. 9

95—

105.

Hah

n, F

rank

[19

82],

Mon

ey a

nd I

nfla

tion,

Bas

il B

lack

wel

l Pu

blis

hers

, Oxf

ord.

Har

riso

n, J

. M

icha

el [

1985

], B

row

nian

Mot

ion

and

Stoc

hast

ic F

low

Sys

tem

s, J

ohn

Wile

y an

d So

ns,

New

Yor

k.

Ichi

kaw

a,

M. M

.H. M

iller

and

A.

Suth

erla

nd [1

990]

, "E

nter

ing

a Pr

eann

ounc

ed

Cur

renc

y B

and"

, War

wic

k E

cono

mic

Res

earc

h Pa

pers

, No.

347

, M

arch

.

Kar

atza

s I.

an

d S.

E

. Sh

reve

[1

988]

, B

row

nian

m

otio

n an

d st

ocha

stic

cal

culu

s,

Spri

nger

—V

erla

g, N

ew Y

ork.

Kle

in,

Mic

hael

W.

1989

[, "

Play

ing

With

The

Ban

d: D

ynam

ic E

ffec

ts O

f Tar

get Z

ones

In

An

Ope

n E

cono

my'

, mim

eo,

Aug

ust.

Kru

gman

, Pa

ul [

1987

], "

Tri

gger

Str

ateg

ies

and

Pric

e D

ynam

ics

in E

quity

and

For

eign

E

xcha

nge

Mar

kets

", N

BE

R W

orki

ng P

aper

No.

245

9, D

ecem

ber.

____

____

[198

9],

"Tar

get Z

ones

with

Lim

ited

Res

erve

s",

Dis

cuss

ion

Pape

r, M

IT.

[199

0],

"Tar

get

Zon

es a

nd

Exc

hang

e R

ate

Dyn

amic

s",

Qua

rter

ly J

ourn

al o

f E

cono

mic

s, f

orth

com

ing.

Kru

gman

, P.

and

J. R

otem

berg

[1

990]

, "T

arge

t Z

ones

With

Lim

ited

Res

erve

s",

mim

eo,

July

.

Lew

is, K

aren

K.

[199

0], O

ccas

iona

l In

terv

entio

ns to

Tar

get R

ates

with

a F

orei

gn E

xcha

nge

App

licat

ion"

, mim

eo,

N.Y

.U.

June

.

McC

aJlu

m,

Ben

nett

T.

ç198

3[,

"On

Non

—un

ique

ness

in R

atio

nal E

xpec

tatio

ns M

odel

s: A

n A

ttem

pt a

t Pe

rspe

ctiv

e,

' Jo

urna

l of M

onet

ary

Eco

nom

ics,

11, M

arch

, pp

. 139

—68

.

—47

Maf

fiar

is, A

. G

. an

d W

. A

. B

rock

[19

821,

St

ocha

stic

Met

hods

in E

cono

mic

s an

d Fi

nanc

e,

Nor

th—

Hol

land

, A

mst

erda

m.

Mill

er,

Mar

cus,

H.

and

Paul

Wel

ler

[198

8a],

"So

lvin

g St

ocha

stic

Sad

dlep

oint

Sys

tem

s:

A

Qua

litat

ive

Tre

atm

ent

with

Eco

nom

ic A

pplic

atio

ns",

W

arw

ick

Eco

nom

ic

Res

earc

h Pa

per

No.

309

(C

oven

try,

Eng

land

: Uni

vers

ity of

War

wic

k).

____

____

and

___

____

_ [1

988b

}, "T

arge

t Z

ones

, C

urre

ncy O

ptio

ns a

nd M

onet

ary

Polic

y",

snim

eo,

July

.

____

____

and

[1

989]

, "A

Qua

litat

ive

Ana

lysi

s of

Sto

chas

tic S

addl

epat

hs

and

its

App

licat

ion t

o E

xcha

nge R

ate R

egim

es",

snim

eo,

Uni

vers

ity of

War

wic

k.

____

____

an

d __

____

__

[199

0a],

"Cur

renc

y B

ubbl

es

Whi

ch A

ffec

t Fu

ndam

enta

ls:

A

Qua

litat

ive

Tre

atm

ent"

, The

Eco

nom

ic J

ourn

al, V

ol.

100,

No.

400

, Su

pple

men

t 19

90,

pp.

170—

179.

____

____

an

d __

____

__

[199

0b],

"Cur

renc

y B

ands

, T

arge

t Z

ones

an

d C

ash

Lim

its:

Thr

esho

lds f

or M

onet

ary

and

Fisc

al P

olic

y", C

EPR

Dis

cuss

ion

Pape

r N

o. 3

82, M

arch

.

____

____

an

d __

____

__

[199

0c],

Exc

hang

e R

ate

Ban

ds

with

Pri

ce I

nert

ia.

War

wic

k U

nive

rsity

, m

imeo

, Fe

brua

ry.

Mill

er,

M.H

. an

d A

. Su

ther

land

[199

0],

"Bri

tain

's R

etur

n to

Gol

d an

d Im

pend

ing

Ent

ry

into

the

EM

S: E

xpec

tatio

ns, J

oini

ng C

ondi

tions

and

Cre

dibi

lity"

, m

imeo

, Ju

ly.

____

____

__

____

__ a

nd J

ohn

Will

iam

son

[198

9],

"The

Sta

biliz

ing

Prop

ertie

s of

Tar

get

Zon

es",

in

Ral

ph C

. B

ryan

t, D

avid

A.

Cur

rie,

Ja

cob

A.

Fren

kel,

Paul

R.

Mas

son

and

Ric

hard

Po

rtes

ed

s.,

Mac

roec

onom

ic

Polic

ies

in

an I

nter

depe

nden

t W

orld

, 11

fF,

Was

hing

ton,

D.C

., pp

. 248

—27

1.

Obs

tfel

d, M

auri

ce a

nd K

enne

th R

ogof

f [1

983]

, "Sp

ecul

ativ

e H

yper

infl

atio

ns In

Max

imiz

ing

Mod

els:

Can

We

Rul

e T

hem

Out

?" J

ourn

al o

f Pol

itica

l Eco

nom

y, 9

1, p

p. 6

75—

687.

Pese

nti,

Paol

o A

. [1

990]

, "E

xcha

nge

Rat

e St

ocha

stic

D

ynam

ics

and

Tar

get

Zon

es:

An

Intr

oduc

tory

Sur

vey"

, m

imeo

, Ja

nuar

y.

Sarg

ent,

Tho

mas

J.

and

Nei

l W

alla

ce ]1

981]

, "S

ome

Unp

leas

ant

Mon

etar

ist A

rith

met

ic",

Fe

dera

l Res

erve

Ban

k of

Min

neap

olis

Qua

rter

ly R

evie

w,

5(3)

; pp

. 1—

17.

Smith

, Gre

gor

W.

and

Mic

hael

G.

Spen

cer

[199

0],

"Est

imat

ion

and

Tes

ting

in M

odel

s of

E

xcha

nge

Rat

e T

arge

t Zon

es a

nd P

roce

ss S

witc

hing

", m

imeo

, Ju

ly.

Sven

sson

, L

ars

E. 0

. [1

989]

, "T

arge

t Zon

es a

nd In

tere

st R

ate

Var

iabi

lity"

, NB

ER

Wor

king

Pa

per

No.

321

8, D

ecem

ber.

____

____

_ [1

990a

],

"The

Ter

m S

truc

ture

Of

Inte

rest

Rat

e D

iffe

rent

ials

In

A T

arge

t Zon

e:

The

ory

and S

wed

ish

Dat

a", m

imeo

, Feb

ruar

y.

____

____

[199

0b],

"The

Sim

ples

t Tes

ts o

f T

arge

t Zon

e C

redi

bilit

y", N

BE

R W

orki

ng P

aper

, N

o. 3

394,

Jun

e.

Woo

dfor

d,

M.

[198

7],

"Lea

rnin

g T

o B

elie

ve I

n Su

nspo

ts",

m

imeo

, U

nive

rsity

of C

hica

go

Gra

duat

e Sc

hool

of

Bus

ines

s.