The effects of public expenditure on economic growth in Asia countries: A Bayesian Model Averaging approach

This ambiguity a ects the reliability of the results. Using Bayesian Model Averaging (BMA) method with the data obtained from 43 Asian countries in the period 2004-2016, we estimate the impact of public expenditure on economic growth with a large number of explanatory variables included in the model.

Asian Journal of Economics and Banking

ISSN 2588-1396http://ajeb.buh.edu.vn/Home

The Eects of Public Expenditure on Economic Growth in Asia Countries: A Bayesian Model Averaging Approach

Nguyen Ngoc Thach1,† *, Le Hoang Anh2, and Pham Thi Ha An3

1Banking University HCMC, Ho Chi Minh City, Vietnam

2HCMC University of Food Industry, Ho Chi Minh City, Vietnam

3Van Lang University, Ho Chi Minh City, Vietnam

Bayesian model averaging,

Eco-nomic growth, Public

a negative impact on the economic growth in Asian countries.

On the other hand, the components of public expenditure have

a weak impact on economic growth The empirical results

con-rm that since the majority of Asian countries are developing countries with a large proportion of state-owned sectors and low governance quality, large scale of public expenditure does not have positive eects on the economic growth Based on the research results, this study provides policy implications to im- prove governance quality and eciency of public expenditures

in Asian countries.

*† Corresponding author: Nguyen Ngoc Thach, Banking University HCMC, Ho Chi Minh City Email address: thachnn@buh.edu.vn

1 INTRODUCTION

Economic growth has been the topic

of great concern in economic theory

from classical schools to modern

ideolo-gies It is also one of the most

topi-cal issues under research in several

na-tions Over the past decades, the

im-pact of public expenditure on economic

growth has been studied by many

re-searchers such as [1], [10], [14], [23], [33]

However, the studies on the impact of

public expenditure on economic growth

show mixed ndings In general, three

main conclusions were drawn in the

re-lated literature: rst, public

expendi-ture has no eect on economic growth

[10]; second, public expenditure has a

positive impact on economic growth [1],

[33]; and third, the relationship

be-tween public expenditure and economic

growth is nonlinear, it means that

in-creasing public expenditure spurs

eco-nomic growth, but when public

expen-diture surpasses a certain threshold, it

will reduce economic growth [23]

Most of the empirical studies above

indicate that the number of

explana-tory variables was predened in such

regression models The research

objec-tive is then focused on the estimation of

the parameters of the models

Employ-ing such approach, there is an

uncer-tainty in the model because the

num-ber of explanatory variables that need

to be included is unspecied The

un-certainty of the model and its

parame-ters, as well as the potentially variable

deviation, may be omitted due to the

inadequate set of explanatory variables

that would make the econometric

anal-ysis inaccurate BMA analanal-ysis is the

best tool for estimating transnational

variations and nding strong growth terminants To explain the uncertainty

de-of the model, Hoeting et al [16] showthat the BMA results were superior tothose obtained by traditional estima-tion methods Especially, the BMA es-timates a large number of models andtakes the following average value of thecoecient (total model space weights),

so the result will be more accurate forpredictions The unconditional parame-ters obtained from the BMA do not de-pend on a specic model because theyare the average of the conditional pa-rameters of all models in the modelspace This helps to avoid a bias fromchoosing a particular model

The study is structured into six tions Following the introduction, pre-sented in Section 1, Section 2 presentsthe theoretical framework and empiricalstudies on the impact of public expen-diture on economic growth Section 3explains the research methods and dataanalysis The results and discussion ofempirical analysis are then presented inSection 4 Based on the research results,Section 5 draws conclusions and makespolicy recommendations

STUDIES2.1 Theoretical Framework

In order to study the relationshipbetween public expenditure and eco-nomic growth, the relationship betweenthe state and the market is analyzed be-cause public expenditure is one of themain tools of the scal policy which

aimed at the proactive impact of the

state on an economy There are two

well-known concepts regarding this

rela-tionship: market asco and government

asco

Market asco In order to have

strong arguments for the objective

ne-cessity for state intervention in an

econ-omy, Keynesian economics uses the

concept of market failures

Recog-nizing the certain advantages of the

free-market economic system over the

centrally planned economy, Keynesians

point out that the market economy is

not able to solve many problems or solve

them with low eciency The ability of

the free market to ensure general

equi-librium and high eciency is very low

This situation is called market failure

The typical types of market asco are:

(i) the full eect of producing or

con-suming many goods and services is not

fully reected in their market prices

be-cause of the existence of externalities,

so the state is responsible for

neutraliz-ing those eects; (ii) the market cannot

produce enough public products (e.g

education, health or defence services,

etc.), so the state must participate in

the production of all or part of

pub-lic goods; (iii) market relations, based

on increasing competition, lead to the

emergence of monopolies, so one of the

state's tasks is to protect the

compet-itive environment; (iv) business cycles

are an objective phenomenon

appear-ing periodically in a market economy,

which goes through ourishing periods

that are replaced by crisis periods, so

the state has a duty to stabilize business

cycles, its economic policies should aim

at two objectives: full employment and

price stability; (v) the market nizes the only income distribution that

recog-is considered fair when the winner gets

a high income, while the loser does notreceive any income or receive a low in-come But such income distribution isnot fair in terms of humanity, so thestate needs to redistribute national in-come to minimize the dierentiation ofwelfare; and (vi) market mechanismsthat allow eective use of social forces,but cannot create a huge breakthrough

in fundamental research as well as deepshifts in economic structures In someeconomic sectors where investment has

a long payback period, high levels ofrisk are not attractive to private com-panies Therefore, another importanttask of the state is to stimulate tech-nical progress and implement economicrestructuring

Market failures require state vention in an economy However, exces-sive state intervention can distort themarket mechanism, causing an ine-cient distribution of resources Neoclas-sical economists, especially new institu-tionalists, pay attention to governmentfailures

inter-Government asco The cal and neo-classical schools support theview of government failures The fail-ures of the state are summarized as fol-lows: (i) access to information is oftenlimited, so the government reduces itsresponsibility in cases when it does nothave all necessary information to accu-rately forecast the consequences of eco-nomic policies; (ii) the state cannot con-trol the reaction of agents to its policies,

classi-so their nal impact does not dependentirely on itself; (iii) incomplete politi-

cal processes such as electors' irrational

behavior, arbitrary decisions, and the

inuence of interest groups; (iv) control

over the government apparatus is

lim-ited A system or part of public

ser-vants uses state power to realize their

interests; (v) seeking to indiscriminately

increase production of public goods and

services that lead to waste of existing

re-sources; (vi) creating privileges for

cer-tain groups of people; and (vii) inating

the costs of maintaining the excessive

bureaucracy

The world has witnessed many

eco-nomic growth models applied among

which there are ve popular ones These

models are based on one or the other

market failures or government failures

Demand-side model (see [18],

[15], [9]) Based on Keynesian theory,

this model was applied in the United

States after the Second World War

un-til the 1970s Its main task is to keep

aggregate demand higher than

aggre-gate supply to boost economic growth

in which public expenditure was

consid-ered the most important element of

ag-gregate demand

Supply-side model (based on the

perspective of the classical and

neo-classical schools from the neo-classical

economists, such as Adam Smith, David

Ricardo, John Stuart Mill to the new

classical school with [12], [27], [20] This

model aims to enhance the development

of productive forces and to restructure

an economy, even allowing some

imbal-ance in it These measurements which

help encourage economic growth were

used in the least developed countries

and some developed countries in

1970s-1980s after the Keynesian theory was in

crisis

Export promotion model It isapplied in the countries where exportsaccount for a large share (such as Swe-den, France, Germany and emerging in-dustrial countries of Asia) The maintask of this model is to improve ef-

ciency in export industries and thus

to increase the country's internationalcompetitiveness in the world market.Import substitution model.This model is used in the countrieswhich have been greatly dependent onimport in the context of weak compet-itiveness of domestic products Themodel was applied in the former So-viet Union and some Latin Americancountries with an aim to implementthe catch-up industrialization strategywith the states active role, but theireconomies were low-ecient

Balanced growth model Thismodel aims to maintain steady eco-nomic growth with the possible high-est level of balance So, maximizinggrowth is not a goal in itself This model

is widely used in developed countriesnowadays

Next, the relationship between lic expenditure and economic growth isanalyzed

pub-For a proper understanding of theprobable impact of public expenditure

on economic growth, it is necessary

to classify public expenditure in somemeaningful ways Since there are dier-ent classication system, choice of suit-able system depends on the objectivesthat an analyst would like to achieve.Aschauer [3] further makes classica-tions of public expenditures in the con-text of productive and protective expen-

ditures Productive expenditure

com-prises economic services and social and

community services while protective

ex-penditure includes administration and

transfers Devarajan et al [8] also note

the productive and unproductive public

expenditures when they opine that

pro-ductive expenditures when used in

ex-cess, could become unproductive The

results of their study imply that

devel-oping countries' governments have been

misallocating public expenditures in

fa-vor of capital expenditures at the

ex-pense of current expenditures

Productive and unproductive

expen-ditures show that while some

expendi-tures are like consumption, others are

like investments which helps an

econ-omy to improve its productive

capac-ity Bhatia [6] submits that under the

laissez-faire philosophy, the only

pro-ductive public expenditures are those

which are incurred to create and

main-tain social overheads Expenditures

on administration, defense, justice, law

and order, and maintenance of the state

are unproductive (i.e., protective) It

must be noted, however, that these

pro-tective expenditures would be essential

for the productive eciency of the

econ-omy

Rele and Westerhout [26] view

the classication of public expenditure

clearly in an analytical manner They

classify public expenditure into two

main categories The rst category

in-cludes consumption expenditure which

are the expenditure items generating

benets in the period in which the

ex-penditure occurs The second one is an

investment, including all items of

pub-lic expenditure that generate benets in

the future

In most countries, public ture is used as a tool of scal policy,but its impact on economic growth is

expendi-a controversiexpendi-al issue The two tant conceptions of the relationship be-tween public expenditure and economicgrowth are the Wagner Law and Key-nesian theory Wagner [32] points out

impor-a cimpor-ausimpor-al relimpor-ationship between public penditure and national income How-ever, the researcher states that publicexpenditure is not the cause but an en-dogenous variable of economic growth.The rise of economic growth results in

ex-an increase in public expenditure trary to Wagner's point of view, Key-nesian theory [18] suggests that an in-crease in public expenditure has a posi-tive impact on economic growth Thus,public expenditure is an exogenous forcethat promotes economic growth [19].According to Keynesian theory, dis-crete scal policy is an important toolavailable to the governments to stim-ulate economic growth [28] UnlikeKeynes, Solow [30] proposes a neoclas-sical growth model in which there is

Con-no long-term eect of public ture on economic growth Neoclassi-cal growth models prove that scal pol-icy cannot lead to long-term changes inoutput growth Neoclassical economistsclaim that long-term growth is due tochanges in population, labor force andtechnical progress which are identied

expendi-as exogenous In contrexpendi-ast to Keynes'theory, with the new classicals' endoge-nous growth model, Barro [5] arguesthat public expenditure has a negativeimpact on economic growth He main-tains that public expenditure may over-

whelm private investment but does not

provide a stimulus to compensate for

in-vestment and growth

2.2 Empirical Studies

There has been a large body of

re-search on the impact of public

expen-diture on economic growth However,

their results are not consistent with the

essence of this impact Some

stud-ies show the positive linear impacts of

public expenditure on economic growth,

e.g., [1], [7], [33] Meanwhile, others

in-dicate that public expenditure does not

aect economic growth (see in [10]) or

that the eect is nonlinear (see in [23])

Yasin [33] examines the eects of

government expenditure on economic

growth using panel data from

Sub-Saharan, Africa This model is derived

from an aggregate production function

that includes input variables of

govern-ment expenditure, private investgovern-ment,

ODA (Ocial Development Aid) and

trade openness Fixed eect (FE) and

random eect (RE) models were

ap-plied The results from both estimation

techniques indicate that government

ex-penditure, trade openness and private

investment have a signicant positive

impact on economic growth

Similar to Yasin [33], Alexiou [1]

provides further evidence of the

rela-tionship between government

expendi-ture and economic growth Like Yasin

[33], the researcher employed two

esti-mation methods, xed eect and

ran-dom eect, with panel data These

models were used for seven transition

economies in South East Europe (SEE)

between 1995 and 2005 The ndings

show that government expenditure can

improve the economic eciency of thecountries in this region More speci-cally, the evidence has shown that four

of the ve variables used in the modelgovernment expenditure, ODA, privateinvestment, and trade openness have asignicant positive impact on economicgrowth

Cooray [7], one of the well-knownstudies, analyzes the impact of govern-ment expenditure on economic growth.This study aims to assess the role

of government in enhancing economicgrowth based on the extension of theclassical production function by consid-ering two aspects as the size of pub-lic expenditure and government qualityknown as public institutions The studywas conducted in 71 countries The em-pirical results show that both public ex-penditure and public institutions have apositive impact on economic growth

In contrast to the studies of Yasin[33], Alexiou [1], Cooray [7], and East-erly & Rebelo [10] uses data obtainedfrom 125 countries during the 1870-1988period By using the multiple regressionmethod with many explanatory vari-ables, including government expendi-ture, real government consumption/realGDP, public investment, taxation, theresearch results show that taxation andpublic expenditure have no impact oneconomic growth

Also, Nurudeen and Usman [25] alyze government expenditure and eco-nomic growth in Nigeria By us-ing the co-integration and error correc-tion methods and employing time-seriesdata for the 1979-2007 period, they de-velop a model based on Keynesian andendogenous growth one Their study

an-concludes that capital expenditure,

to-tal recurrent expenditures, and

govern-ment expenditure on education have a

negative eect on economic growth

3 RESEARCH METHODS AND

DATA

3.1 Research Methods

3.1.1 Research Models

To assess the impact of public

ex-penditure on economic growth, we

con-struct a model based on the studies of

Alexiou [1] and Cooray [7], starting with

the Cobb-Douglas production function

as follows:

Y (t) = (K (t))α(H (t))β(G (t))γ

×(A (t) × L (t))1−α−β−γ (1)

Where Y, H, K, G, L, and A stand

for national income, human capital,

pri-vate capital, government capital,

la-bor, and technical progress,

respec-tively And α + β + γ < 1

Suppose L and A have growth rates

n and g, respectively, then: L (t) =

L (0) × ent; A (t) = A (0) × egt

With the assumption of savings is

the constant s rate of national income

and savings equal to investment Thus,

savings will be allocated to invest in

hu-man capital, private capital and

govern-ment capital, then: s = sk + sh + sg,

where sk is the savings ratio used to

in-vest in private capital, sh is the savings

ratio used to invest in human capital,

sg is the savings ratio used to invest in

government capital

Divide both sides of (1) by A(t)L(t)

and carry out some transformations, we

Putting h∗ = sh k ∗

s k and s g k ∗

s k into the rstequation of the system, will calculate:

In the same way, it will also calculate:

h∗ = (s

1−α−γ

h sα

ksγ g

n + g + δ )

1 1−α−β−γ

Placing k*, h*, g* into (2), we get:

×(s

1−α−γ

h sα

ksγ g

n + g + δ )

β 1−α−β−γ

By taking the logarithm of the two

sides of the above equation and the

basic transformations, the equilibrium

state of economic growth is expressed as

a linear logarithmic function as follows:

On the other hand, the growth rate

of GDP per capita in equilibrium is pressed in the form of:

ex-lny (t) − ex-lny (t − 1)

= 1 − e−λ
[lny∗− lny (t − 1)] (4)Where y (t-1) is GDP per capita of theprevious year and y∗is the average GDPper capita in the equilibrium dened byequation (3) Placing (3) into (4) andperforming basic transformations, ob-tain:

lny (t) −lny (t − 1)

= ϕ0+ ϕ1lny (t − 1) + ϕ2ln sk+ ϕ3lnsh+ ϕ4lnsg

Therefore, the growth rate of GDP percapita will depend on the growth rate ofGDP per capita in the previous period,private capital, human capital, and gov-ernment capital

Also, we also analyze the impact

of public expenditure components oneconomic growth The components ofpublic expenditure are included in themodel based on Anh [2] and Efendic &Trkic-Izmirlija [11] These componentsare public expenditure on health (g1),government consumption (g2); publicexpenditure on education (g3); andpublic expenditure on defence (g4).The role of institutions for economicgrowth was rst claimed by North andThomas [24] In addition to the vari-ables shown above, empirical research

by [29] shows that governance aectseconomic growth Nguyen Ngoc Thach

et al [31] also conrm that tutional quality, democracy freedom,

insti-and economic freedom play an

impor-tant role in economic growth

There-fore, governance variables need to be

included in the model In this study,

we use observed variables as the

indica-tors of governance quality in two data

sets: Worldwide Governance Indicators

(WGI) and International Country Risk

Guide (ICRG) Each set of data

in-cludes six major indicators: Voice and

Accountability, Political Stability and

Absence of Violence, Government

Ef-fectiveness, Regulatory Quality, Rule of

Law, Control of Corruption Siddiqui

and Ahmed [29] also include trade

open-ness and ination in the model to

ana-lyze its impact on economic growth

Thus, in this study, up to 23

vari-ables can be included in the model to

explain economic growth as the

depen-dent variable In general, according to

Zeugner [35], the probability of each

model is the same and there will be a

maximum of 223models to be estimated

3.1.2 Method of Estimation

We estimate the model by Bayesian

Model Averaging (BMA) This method

is chosen because of its advantages over

traditional probabilistic methods In

the context that many explanatory

vari-ables can be included in the model,

there will be more than one model with

equal explanatory capacity If only one

model is selected, it may lead to risks

from unstable model The main

pur-pose of model averaging is to consider

and estimate all possible models (the

model space) and to focus on

summa-rized statistics based on weighted

aver-ages of the models in the model space

Madigan and Raftery [22] and Kass and

Raftery [17] provide a sound statistical

derivation for a model combination cedure, called BMA, where the modelweights are derived as additional statis-tical parameters in a Bayesian estima-tion set up

pro-Consider a linear regression modelwith a constant term, β0, and k poten-tial explanatory variables x1, x2, , xk

as follows:

y = β0+ β1x1 + β2x2+ + βkxk+ εWith k potential explanatory variables,

we will have 2k combinations of planatory variables on the right side.Each of these combinations will cre-ate a new model denoted by Mj with

ex-j = 1, 2, , 2k At this time, a modelspace has been constructed The poste-rior distribution for any coecient, say

βh, given the data D, is

P (D|Mj)

=

ˆ

P D|βj, Mj
P (βj|Mj)dβjand βj is the estimated parameter vec-tor of the model Mj P (βj|Mj) is the

a prior probability distribution assigned

to the parameters of the model Mj

P (Mj) is the prior probability that Mj

is the true model

The estimated posterior means andstandard deviations of βh is then con-

Thus, to apply the BMA method, it

is necessary to determine the prior

prob-ability of the model Mj(P (Mj))and the

prior probability assigned to parameters

of model Mj(P (βj|Mj))

According to Zeugner [35], a

popu-lar choice for a prior probability P (Mj)

is a uniform probability distribution

be-cause each model is equally likely to

oc-cur, so:

P (Mj) = 1

2k

In contrast to the choice of a prior

probability P (Mj), the prior

probabil-ity (P (βj|Mj) depends signicantly on

the information that the researcher has

about the probability distribution of

βj In the Bayesian linear regression

model, a priori probability distribution

of the parameters commonly used by

researchers is Zellner's g-prior

Zell-ner [34] proposed g-prior as a common

benchmark prior The g-prior depends

on the data and thus does not violate

the conditional probability rule

Fer-nandez et al [13] show that the most

ef-cient g-prior is benchmark prior which

is g = [max {N, K2}]−1

3.2 Data Analysis

According to the Asian

Develop-ment Bank (ADB), there are 50

tries in Asia However, since some tries have data omission, the study useddata obtained from 43 countries, ac-counting for 86% of Asian countries.Therefore, the research sample is stillrepresentative

coun-In terms of observation time, we alyze the data obtained from 43 coun-tries during the 2004-2016 period Thisresearch period is selected for many rea-sons First, this period allows 43 coun-tries to have sucient data available forthis research Second, this period cov-ers the time before, during and after theglobal economic crisis (2004-2007, 2008-

an-2009, 2010-2016 respectively) fore, we can comprehensively analyzethe impact of public expenditure on eco-nomic growth during an economic cycle

There-in Asian countries

Research database concludes ondary data collected from reliablesources The measurement data of GDPper capita, private investment on GDP,public expenditure on GDP, the propor-tion of public expenditure components

sec-on GDP, trade openness, inatisec-on, man capital are taken from World Eco-nomic Outlook (WEO) of the Interna-tional Monetary Fund (IMF) and WorldDevelopment Indicators (WDI) of theWorld Bank for 43 Asian countries dur-ing the period 2004-2016

hu-For governance data, WorldwideGovernance Indicators (WGI) of 43Asian countries from 2004 to 2016are obtained from the World Bankdatabase International Country RiskGuide (ICRG) of 43 Asian countriesfrom 2004 to 2016 is taken from thePolitical Risk Services Group (PRSGroup)

Table 1 Summary of variables

Variables Notation Denitions and CalculationsEconomic

growth growthit

lnyit− lnyi(t−1) with yit and yi(t-1) respectivelyGDP per capita of country i year t and year(t-1)

capita lngdpi(t-1) Logarithm of GDP per capita of country i year(t-1)Investment invit Investment capital per GDP of country i year

tHuman capi-

tal lit Labor force ratio of country i year t

git Total public expenditure (fraction of GDP)g1it Public expenditure on health (fraction of

GDP)Public expen-

ditures g2it Public expenditure on consumption (fractionof GDP)

g3it Public expenditure on education (fraction of

GDP)g4it Public expenditure on defence (fraction of

GDP)Trade open-

ness Open Import-export ratio per GDP of country i yeartInation inf Ination rate of country i year t

CCICRG Control of Corruption belongs to the ICRG

indexRLICRG Rule of Law belongs to the ICRG indexGovernance RQICRG Regulatory Quality belongs to the ICRG in-

dexGEICRG Government Eectiveness belongs to the

ICRG indexPVICRG Political stability and Absence of Violence be-

longs to the ICRG indexVAICRG Voice and Accountability belongs to the ICRG

indexCCWGI Control of Corruption belongs to the WGI in-

dex

RLWGI Rule of Law belongs to the WGI indexRQWGI Regulatory Quality belongs to the WGI indexGEWGI Government Eectiveness belongs to the WGI

indexPVWGI Political stability and Absence of Violence be-

longs to the WGI indexVAWGI Voice and Accountability belongs to the WGI

index

4 RESULTS AND DISCUSSION

Descriptive statistical results of

measuring specic quantities for study

variables are shown in Table 2.Table 2 Descriptive statistics