IMF-自動(dòng)化與福利：任務(wù)與教育的作用（英）-2024.1

Automation

and

Welfare:The

Role

of

Bequests

andEducationPreparedby

ManukGhazanchyan,AlexeiGoumilevski,andAlexMourmourasWP/24/11IMFWorkingPapersdescriberesearchinprogressbytheauthor(s)andarepublishedtoelicitcommentsandtoencouragedebate.TheviewsexpressedinIMFWorkingPapersarethoseoftheauthor(s)anddonotnecessarilyrepresenttheviewsoftheIMF,itsExecutiveBoard,orIMFmanagement.2024JANWP/24/11?2023InternationalMonetaryFundAutomation

and

Welfare:

The

Role

of

BequestsandEducation*PreparedbyManukGhazanchyan,AlexeiGoumilevski,andAlexMourmourasAuthorizedfordistributionby

AlexMourmourasJanuary2024IMF

Working

Papers

describe

research

in

progress

by

the

author(s)

and

are

published

to

elicitcomments

and

to

encourage

debate.

The

views

expressed

in

IMF

Working

Papers

are

those

of

theauthor(s)anddonotnecessarilyrepresenttheviewsoftheIMF,itsExecutiveBoard,or

IMFmanagement.ABSTRACT:

This

paper

examines

the

welfare

effects

of

automation

in

neoclassical

growth

modelswith

and

without

intergenerational

transfers.

In

a

standard

overlapping

generations

model

without

suchtransfers,

improvements

in

automation

technologies

that

would

lower

welfare

can

be

mitigated

by

shiftsin

labor

supply

related

to

demographics

or

pandemics.

With

perfect

intergenerational

transfers

basedon

altruism,

automation

could

raise

the

well-being

of

all

generations.

With

imperfect

altruism,

fiscaltransfers(universalbasicincome)andpublicpolicies

toexpandaccesstoeducationopportunitiescanalleviatemuchofthenegativeeffectofautomation.JELClassificationNumbers:Keywords:E13,E62,D64Automation;

Aging;

Altruism;

Fiscal

Policy;

Education;OverlappingGenerationsmghazanchyan@;

agoumilevski@;amourmouras@Authors’E-MailAddresses:_________________________________________________________________________________________________?Manuk

Ghazanchyan

is

an

Economist

in

the

Western

Hemisphere

Department,

Alexei

GoumilevskiisaSeniorScientificComputingEngineerintheInformationandTechnology

Department,

andAlexMourmourasisDivisionChiefintheAsiaPacificDepartment.*Presented

at

the

28th

International

Conference

on

Computing

in

Economics

and

Finance

in

Dallas,Texas,

June

17-19th,

2022.

We

thank

conference

participants

for

their

useful

comments.

We

arealso

grateful

to

Peter

Rangazas,

Marina

Mendes

Tavares,

and

Jeremy

Clift

for

the

insightfulcommentsthatshapedourpaper.-2-ContentsI.Introduction

...............................................................................................................................................1II.RelevantLiterature...................................................................................................................................3III.AutomationandWelfareinOverlappingGenerationModels

.............................................................5B.IntroducingaOne-Timetax.................................................................................................................9C.K&SModelwithaBequest

...............................................................................................................11D.K&SModelwithBequestandEducation..........................................................................................141.2.PrivateEducation.......................................................................................................................15PublicEducation

........................................................................................................................17IV.ArtificialIntelligence.............................................................................................................................22A.K&SModel

........................................................................................................................................22IV.Conclusions

..........................................................................................................................................26Appendices.................................................................................................................................................27AppendixA.

K&SModelwithBequest.................................................................................................27AppendixB.

ReplicatingKotlikoffandSachs(2012)

...........................................................................31AppendixC.

RobustnessChecks........................................................................................................31A.References

....................................................................................................................................34-3-I.

IntroductionAutomation

has

accelerated

in

recent

decades,

driven

by

ongoing

improvements

in

computing

andinformation

technologies

and

associated

cost

reductions.

Machines

in

a

widening

range

of

industriesperform

increasingly

complex

tasks,

powered

by

sophisticated,

networked

software.

The

accelerationin

automation

and

its

economy-wide

diffusion

in

blue-

and

white-collar

occupations

alike

is

creating

newemployment

categories

but

is

also

contributing

to

widening

inequality

and

fueling

demand

forgovernment

policies

to

reverse

long-term

income

losses

of

labor.

This

long-standing

promise

andconcerns

are

vividly

illustrated

by

the

latest

breakthrough

in

Artificial

Intelligence

involving

generative,pretrainedtransformers.Looking

ahead,

while

the

pace

of

automation

is

likely

to

continue,

its

effects

may

be

mitigated

byoffsetting

forces.

Populations

are

aging

almost

everywhere.

In

the

advanced

economies,

the

working-agepopulationhasstartedshrinkingfor

thefirsttimesinceWorldWar

II(Spence,

2022).

Globally,thepopulation

ofworking

age

is

expected

tocontinue

togrow

until

about2040,butthe

ratio

ofthe

workingage

population

to

the

total

is

already

declining

globally

(Chart).

In

the

case

of

China,

for

example,

theworking-age

population

is

expected

to

shrink

by

a

fifth

over

the

next

30

years.

As

Goodhart

and

Pradhanstress,

our

age

is

one

of

demographic

reversal

in

which

the

“l(fā)ong

glut

of

inexpensive

labor

that

had

keptprices

and

wages

down

for

decades,

is

giving

way

to

an

era

of

worker

shortages,

and

hence

higherprices”.Recurring

global

pandemics

also

adversely

affect

labor

supply,

either

by

depressing

growth

in

the

laborforce

directly

(AIDS

pandemic),

or

indirectly

by

reducing

the

participation

of

older

workers

and

othersin

contact-intensive

occupations

(pandemic

related

to

Covid-19).

In

the

absence

of

mass

south-northmigration,

robots

may

turn

out

to

be

essential

in

meeting

more

of

the

needs

of

the

elderly

and

reversedeclinesinaggregateoutput

andwelfare1.This

paper

examines

the

combined

welfare

effects

of

automation

and

lower

labor

supply

usingneoclassical

growth

models

with

and

without

intergenerational

transfers.

It

begins

by

replicating

aversion

of

the

well-known

result

of

Kotlikoff

and

Sachs

(K&S,

2012)

that

a

one-time

improvement

in

thetechnical

efficiency

of

machines

ends

up

immiserating

all

future

generations.

This

striking

result

relieson

two

crucial

assumptions.

First,

machines

are

very

good

substitutes

for

unskilled

labor

throughoutthe

economy,

so

that

improved

automation

ends

up

displacing

workers

and

lowering

wages

across

theboard.

Second,

there

are

no

operative

intergenerational

transfers

of

any

kind,

so

that

the

owners

ofcapitalend

upconsumingthe

entirewindfallfrom

theimprovementsinautomationduringtheirlifetime(a

generation,

roughly

thirty

years).

The

positive

shock

to

the

efficiency

of

machines

does

not

raisesaving,

depresses

investment

in

physical

and

human

capital,

and

sets

in

motion

a

never-ending

cycleof

declining

welfare.

Government

policy

is

therefore

needed

to

spread

this

windfall

more

equitablyacross

future

generations.

K&S

consider

wealth

taxes,

in

particular

socializing

a

portion

oftheeconomy’scapitalstockthatallowsthegovernmenttofinanceasustainableincomestream(universalbasic

income)

for

all

future

generations.

Resorting

to

compulsion

is

essential

when

generations

areselfish,precludinganysortof

voluntaryintergenerationaltransfers.Infact,

privateintergenerationaltransfers

aresubstantial,

with

about

half

ofallhouseholdsplanning

toleave

estates

(Laitner

and

Juster,

2017).

The

first

objective

of

the

paper

is

to

reassess

the

welfareeffects

of

automation

in

the

presence

of

intergenerational

transfers,

both

bequests

and

privately

andpublicly

funded

schooling.

In

a

version

of

the

K&S

model

with

operative

bequests,

we

find

thatintergenerational

transfers

are

positive

in

equilibrium

if

the

strength

of

altruism

exceeds

a

certainthreshold,

mitigating

the

negative

effects

of

automation.

But

while

it

is

comfortable

to

know

that

thegains

from

automation

may

be

passed

to

future

generations

without

the

need

to

nationalize

capital,

thismodel

of

perfect

altruism

is

also

extreme:

many

families

in

each

generation

cannot

make

efficienttransferstotheirchildren.What

is

needed

is

a

more

balanced

model,

one

that

features

heterogeneity

both

within

and

acrossgenerations,

with

some

households

making

efficient

bequests

and

others

stuck

in

a

corner

solution.We

assess

whether

automation

is

immiserating

in

a

version

of

K@S

model

that

incorporates

purealtruism,

and

in

Glomm

and

Ravikumar’s

model

(G&R,

1992)

in

which

parents

make

investments

in

theschooling

of

their

children.

We

study

how

fiscal

and

educational

policies

can

best

raise

welfare,

byalleviatingfinancingconstraintsinthefinancingofhumancapitalinvestmentsandrestoringequalityofopportunity.Similar

results

obtain

in

a

version

of

the

model

used

by

Ivanyna,

Mourmouras

andRangazas

(2018)

in

with

two

types

of

households

(the

poor,

who

are

bequest-constrained)

and

the

rich(whoareunconstrained).The

paper

then

turns

to

an

analysis

of

a

combined

shock

involving

a

jump

in

automation

and

asimultaneous

reduction

in

labor

supply

driven

by

demographics.

As

expected,

strengthening

altruisticbondsraisebequestsandhumancapital

investmentsoftheyoung,providinganadditionalstimulustoeconomic

growth.

In

addition,

government

transfers

of

tax

revenue

levied

on

the

rich

can

improve

the1Businessleadershavealsomadeaconnectionbetweenautomationandagingrecently.

OneexamplewastherecentarticleinFortunebyIBMCEO,ArvindKrishnaherehepointstodecliningpopulationstocalmfearsaboutA.I.takingjobs.Hefurtheraddedthatultimately“thereisgoingtobejobcreation”withA.I.,asjobswillalsobeaddedinareaswithmorevaluecreation.-2-welfare

of

the

poor

and

reduce

inequality

within

and

across

generations

when

altruistic

links

betweengenerationsareweak.II.

RelevantLiteratureThe

literature

on

automation

and

its

economic

impact

is

evolving,

with

some

earlier

studies

from

Gordon(2012),

Cowen(2011),Acemoglu

andRestrepo(2017,2018),

SachsandKotlikoff

(2012,2015),

Ford,(2015);

Freeman,

(2015)

amongst

pessimists,

and

Brynjolfsson

and

McAfee

(2014),

Autor

(2014,

2015)among

the

optimists.

The

key

issue

is

whether

automation

replaces

labor

share

and

employmentthrough

replacement

of

routine

tasks

of

ever-increasing

scope

and

complexity

or

whether,

on

net,

itincreaseslaborparticipationbycreatinghigh-payingjobsinemergingnewoccupations.Somegloomyscenarios

for

labor

resulting

from

artificial

intelligence

and

simultaneous

automation

breakthroughs

aredescribed

in

Bostrom

(2014).

Graetz

et.

al.

(2018)

examined

the

economic

contribution

of

modernindustrialrobotsin17countriesfortheperiod1993-2007.Contrary

to

the

pessimistic

view,

these

authors

found

that

the

increasing

use

of

robots

raised

the

annualgrowth

of

GDP

and

labor

productivity

by

0.37

and

0.36

percentage

points,

respectively.

Authorsconclude

that

those

robots

did

not

significantly

reduce

total

employment,

although

they

did

reduce

low-skilled

workers’

employment

share.

Gaaitzen

et

al.

(2020)

studied

the

effects

of

adaptation

of

industrialrobots

and

occupational

shifts

by

task

content

in

the

thirty-sevencountries

for

the

period

from

2005

to2015.

The

authors

found

that

increased

use

of

robots

is

associated

with

positive

changes

in

theemployment

share

of

non-routine

analytic

jobs

and

negative

changes

in

the

share

of

routine

manualjobs.

Of

course,

enhancing

policy

including

R&D

and

the

regulatory

platforms

in

both

private

and

publicsectors

to

support

digital

technologies

is

key

to

improve

productivity.

While

the

2020-22

pandemichelped

to

accelerate

the

digital

transformation,

many

sectors

–

including

the

public

sector

–

are

lagging,andhenceconcernsabouttheeffectsofautomationonemploymentwillpersist(Spence,2022).Only

a

few

studies

examined

the

effect

of

automation

and

population

aging

on

the

labor

market

asidefrom

the

classical

work

by

Frey

and

Osborne

(2017)

focusing

on

the

probability

of

automation

affectingvarious

jobs

and

occupations.

One

of

the

earliest

studies

on

automation

and

population

aging

was

byAcemoglu

and

Restrepo

(2017),

where

the

authors

examined

the

relationship

between

economicgrowth,

population

aging,

and

automation

at

the

country

level.

Phiromswad

et

al

(2022)

is

amongst

themost

recent

studies

to

focus

on

those

effects

but

also

on

the

interaction

effects

of

automation

andpopulation

aging

on

the

labor

market.

Consistent

with

previous

literature

including

with

Graetz

andMichaels

(2018)

the

authors

found

strong

evidence

that

automation

negatively

affects

employmentgrowth

holding

other

factors

constant.

They

also

found

strong

evidence

that

the

disaggregatedmeasures

of

age-related

abilities

affect

employment

growth

but

not

the

aggregate

measure.

Asexpected

and

consistent

with

findings

that

automation

is

still

evolving

in

affecting

high

value

jobs,

theauthors

find

that

with

occupations

with

low

score

on

both

the

age-appreciated

cognitive

ability

as

wellage-depreciated

physical

ability

(such

as

production

occupations

and

food

preparation

and

servingrelated

occupations),

the

negative

effect

of

automation

on

employment

tends

to

be

strongest.

However,for

occupations

with

a

high

score

in

both

age-appreciated

cognitive

ability

as

well

as

age-depreciatedphysical

ability

(such

as

protective

service

occupations

and

healthcare

practitioners

and

technicaloccupations),thenegativeeffectofautomationonemploymenttendstobeweakest.Aghion-Jones-Jones

(2017)

study

the

implications

of

artificial

intelligence

for

economic

growth

in

lightof

reconciling

evolving

automation

with

the

observed

stability

in

the

capital

share

and

per

capita

GDPgrowth

over

the

past

century.

The

authors

create

sufficient

conditions

to

generate

overall

balanced-3-growth

with

a

constant

capital

share

that

stays

well

below

100

percent,

even

with

nearly

completeautomation.

In

other

words,

while

Baumol’s

cost

disease

leads

to

a

decline

in

the

share

of

GDPassociated

with

manufacturing

or

agriculture

(once

they

are

automated),

this

is

balanced

by

theincreasing

fraction

of

the

economy

that

is

automated

over

time.

The

authors

also

study

the

effects

ofintroducing

A.I.

in

the

production

technology

for

new

ideas

and

the

possibility

that

A.I.

could

generatesome

form

of

a

singularity,

where

the

authors

nevertheless

claim

that

the

Baumol

theme

here

alsoremains

relevant:

even

if

many

tasks

are

automated,

growth

may

remain

limited

due

to

areas

thatremainessentialyetarehardtoimprove.Pizzinelliandothers(forthcoming)examinetheimpactofArtificial

Intelligence

(AI)

onlabormarketsinboth

Advanced

Economies

(AEs)

and

Emerging

Markets

(EMs).

The

authors

propose

an

extension

toa

standard

measure

of

AI

exposure,

accounting

for

AI's

potential

as

either

a

complement

or

a

substitutefor

labor,

where

complementarity

reflects

lower

risks

of

job

displacement.

Then

they

analyze

worker-level

microdata

from

two

AEs

(US

and

UK)

and

four

EMs

(Brazil,

Colombia,

India,

and

South

Africa),revealing

substantial

variations

in

unadjusted

AI

exposure

across

countries.

The

authors

found

thatwhileAI

posesrisksoflabordisplacement

duetotaskautomation,

italsoholdspromiseinitscapacityto

enhance

productivity

and

complement

human

labor,

especially

in

occupations

that

require

a

highlevelofcognitiveengagementandadvancedskills.

TheauthorsalsofindthatAEsmayexpectamorepolarizedimpactofAIonthelabormarketandarethuspoisedtofacegreaterriskoflaborsubstitutionbutalsogreaterbenefitsforproductivity.The

extent

and

form

of

voluntary

intergenerational

transfers

is

dictated

by

the

strength

ofintergenerationalaltruismandisanimportantconsiderationinmacroeconomicsthatisrelevantforourpaper.

Kotlikoff

(2001)

provides

an

excellent

survey

of

key

works

on

the

role

of

intergenerationalaltruism,includingempiricalfindings—forexample,the

resultsofAltonjiandHayashi(1994)whichareconsistent

with

the

pure

altruism

theory.

A

closely

related

area

of

research

concerns

the

form

of

humancapital

investments,

specifically

the

rationale

behind

education

or

other

bequests

in

kind.

Razin

andRosenthal(1990)showthatfamilytaxationasaresponsetoinformationasymmetrybetweenaparentand

a

child

could

reduce

the

need

for

government

intervention

and

taxation.

Hood

and

Joyce

(2017)provideanexcellentupdatetotheempiricalrelevanceofaltruism.OurpaperismostcloselyrelatedtoMichel,Thibault,andVidal(2004),whostudytheeffectsofaltruismandfiscalpoliciesongrowthinanoverlapping

generations

model

in

the

tradition

of

Diamond,

and

to

Glomm

and

Ravikumar

(1992)

whostudy

bequests

in

the

form

of

human

capital

investments

in

children.

We

study

privately

fundedschoolingforfamilieswithoperativebequestsandpubliclyfundededucation.We

find

that

government

spending

on

education

promote

economic

growth.

These

conclusions

aresupported

by

a

vast

volume

of

research

that

link

individuals’

education

attainment

to

economy-wideprosperity.

Fabrizio

Carmignani

(2016)

studied

effects

of

government

expenditures

on

education

toeconomy.

Author

used

the

World

Bank's

World

Development

Indicator

database

data

on

151

countriesfor

2000

-

2010

years.

He

concluded

that

“increase

in

education

expenditure

by

1

point

of

GDPincreases

GDP

growth

by

0.9

percentage

points”.

Gheraia,

Zouheyr

et

al.

(2021)

investigatedrelationship

between

the

cost

of

education

and

economic

growth

in

the

Kingdom

of

Saudi

Arabia

forthe

period

1990-2017.

Authors

found

that

in

the

long

run

the

rise

in

educational

expenditure

by

1%would

lead

to

an

increase

in

economic

growth

by

0.89%.

Similar

results

were

obtained

by

Yahya,

Mohdet

al.

(2012).

Authors

analyzed

the

long-run

relationship

and

causality

between

the

governmentexpenditure

in

education

and

economic

growth

in

Malaysia

for

the

period

1970

to

2010.

Theyconcluded

that

economic

growth

is

positively

correlated

with

fixed

capital

formation,

labor

forceparticipation

and

expenditure

in

education.

Regarding

Granger

causality,

the

educational

expenditure-4-is

the

short-term

Granger

cause

of

economic

growth

and

vice

versa.

Mehmet

Mercana

et

al.

(2014)performed

cointegration

analysis

between

the

real

gross

domestic

product

and

total

expenses

to

theeducation

for

the

case

of

Turkey

for

the

period

1970-2012.

Authors

used

Autoregressive

DistributedLag

model

with

bounds

testing.

Authors

found

that

1%

increase

in

education

expenses

increaseseconomicgrowthby0.3%.III.

AutomationandWelfareinOverlappingGenerationModelsA.

AnAnalyticalTool:TheK&SModel(2012)We

begin

with

a

simple

model

featuring

two

period-lived

overlapping

generations

(OLG).

Each

period?

=

1,2,

…

thepopulationconsistsofyoungandoldhouseholds.Theyoungareendowedwith

oneunitof

inelastically

supplied

unskilled

labor.

They

consume

part

of

this

income

and

invest

the

rest

in

physicalcapital

(machines,

?)

and

in

their

own

human

capital

(skilled

labor,

?).

Machines

and

human

capitalareperfect

substitutesin

savingportfolios.In

the

second

period

oflife,householdsrenttheirmachinesand

skills

in

perfectly

competitive

markets,

consuming

all

interest

and

principal.

Gross

output

?

is

aconstant

elasticity

of

substitution

(CES)

production

function

of

the

economy-wide

stocks

of

?,

?,

and?:?

=

?(?(??,

?),

?)(A.1)?

and

?

combine

in

a

CES

production

function

with

elasticity

???

to

produce

an

intermediate

product?,and?

and?

combineinaCESproductionfunctionwithelasticity???

toproducethefinaloutput?.The

parameter

?

is

a

parameter

measuring

the

technical

efficiency

of

machinery.

A

rise

in

?

is

a

puretechnical

advance.

Kotlikoff

and

Sachs

(KS,2012)

examine

whether

a

rise

in

?

can

reduce

economicwellbeing,anoutcometheyrefertoas“im-mesmerizingproductivity.”Competitive

firms

hire

?,

?,

and

?

to

the

point

where

their

marginal

products

(denoted

as

?

for

?

=??,

?,

and

?)

equal

their

market

wages:

?

=

?

.

Following

K&S

the

partial

derivative

of

wage

with??respecttoproductivityu

is,???(?

)?

?=

[???

?????]????(A.2)???(?)where

θ

is

the

share

of

skilled

labor

in

the

economy,

equal

to

(?

??)??.

We

see

that

a

rise

in?machine’s

productivity

reduces

the

unskilled

wage

if

???

>

?????.

Thus,

“im-mesmerizing”productivityismorelikelyif:?

Substitutabilityofmachinesandunskilledlaborishigh(???

large)?

Substitutabilityofintermediategoodsandskilledlaborislow(???

small)?

Theshareofskilledlaborinfinaloutputishigh(?

high)Below

we

present

theoretical

underpinnings

for

the

K&S

conclusions.

Note

that

the

income

of

theyoung,?

,iscomprisedofwages?

,partofwhichisinvestedinmachinesandhumancapital.Income??-5-when

old,

??+1,

comes

from

the

ownership

of

machines

and

acquired

skills.

The

lifetime

budgetconstraintsofgenerationbornat?,whoareyounginperiodtandoldat?

+

1

are:?

=

?

?

=

?

+

??(A.3)(A.4)???????????+1

=

(

)

?

+

(

)

?=

??+1?+1?+1??????Here

?

is

the

consumption

when

young,

?is

consumption

when

old,

and

??

is

saving,

which

is??+1?investedinmachinesandskills:??

=

?

+

??+1(A.5)??+1This

allocation

of

savings

is

made

under

perfect

foresight

to

maximize

utility,

so

that

investment

inmachines

and

skills

are

chosen

to

equalize

marginal

products

of

M

and

S

to

the

gross

interest

rate

intheeconomy:?????

=

(

)

=

(

)

=

1+

?(A.6)????????