THAMMASAT REVIEW OF ECONOMIC AND SOCIAL POLICY › paper › vol4-2 › 01_TRESP Vol.4... ·...

Faculty of Economics, Thammasat University

THAMMASAT REVIEW OF

ECONOMIC AND SOCIAL POLICY

Credit Guarantee Optimization of State-owned Enterprises

Chayanisa Chaisuekul

Investigating relationship between Government Spending

and Economic Growth: Public Spending and long-run GDP

level

Kritchasorn Jarupasin

Linkage of Technological Innovation toward ICT base and

Economic Output in CLMV Region

Theara Chhorn

Volume 4, Number 2, July - December 2018

ISSN 2465-390X (Print)

ISSN 2465-4167 (Online)

THAMMASAT REVIEW OF

ECONOMIC AND SOCIAL POLICY Volume 4, Number 2, July – December 2018

ISSN 2465-390X (Print)

ISSN 2465-4167 (Online)

Thammasat Review of Economic and Social Policy

Thammasat Review of Economic and Social Policy (TRESP) is a

double-blind peer reviewed biannual international journal published in

June and December. The journal is managed by the Research Committee

under the supervision of the Academic Affairs Division of the Faculty of

Economics, Thammasat University. Our editorial board and review panel

comprise of academicians and practitioners across various areas of

economic and social policies. The goal of the journal is to provide up-to-

date practical and policy-oriented analysis and assessment of economic and

social issues, with particular focus on Asia and the Pacific region.

However, research findings from other parts of the world that are relevant

to the theme of the journal may be considered.

Aims & Scopes

Our journal is dedicated to serve as a platform for debate and

critical discussion pertaining to the current issues of public policy. The

outcome of such research is expected to yield concrete policy implications.

Some of the targeted issues include urban and regional socio-economic

disparities, ageing society, healthcare, education and welfare policies,

environmental and natural resources, local communities, labor migration,

productivity, economic and political integration, political economy,

macroeconomic instability, trade and investment, fiscal imbalances,

decentralization, gender issues, behavioral economics and regulations; and

law and economics. The journal makes its best effort to cater a wide range

of audience, including policymakers, practitioners in the public and

business sectors, researchers as well as graduate students.

Articles should identify any particular issue concisely, address the

problems of the research explicitly and supply sufficient empirical data or

strong evidence and substantial argument to support the discussion of

policy initiatives asserted by the author(s). Theoretical and applied papers

are equally welcome provided their contributions are policy-relevant.

Authors are responsible for the published articles. The views and

opinions expressed in the articles do not necessarily reflect those of the

Editors and the Editorial Board.

Advisory Board

Dean, Faculty of Economics, Thammasat University, Thailand

Ian Coxhead, University of Wisconsin-Madison, United State

Tran Van Hoa, Centre for Strategic Economic Studies, Victoria University, Australia

Suthipun Jitpimolmard, Khon Kaen University and Thailand Research Fund, Thailand

Medhi Krongkaew, National Institute of Development Administration, Thailand

Duangmanee Laovakul, Thammasat University, Thailand

Arayah Preechametta, Thammasat University, Thailand

Sakon Varanyuwatana, King Prajadhipok's Institute, Thailand

Editor-in-Chief

Euamporn Phijaisanit, Thammasat University, Thailand

Associate Editor

Pornthep Benyaapikul, Thammasat University, Thailand

Editorial Board

Kirida Bhaophichitr, Thailand Development Research Institute, Thailand

Brahma Chellaney, Center for Policy Research, New Delhi, India

Aekapol Chongvilaivan, Asian Development Bank, Manila, Philippines

Emma Jackson, Bank of England, UK

Armin Kammel, Lauder Business School, Vienna, Austria

Somprawin Manprasert, Bank of Ayudhya, Thailand

Gareth D. Myles, School of Economics, University of Adelaide, Australia

Voraprapa Nakavachara, Chulalongkorn University, Thailand

Songtham Pinto, Bank of Thailand, Thailand

Pathomdanai Ponjan, Fiscal Policy Office, Ministry of Finance, Thailand

Watcharapong Ratisukpimol, Chulalongkorn University, Thailand

Sasatra Sudsawasd, National Institute of Development Administration, Thailand

Maria-Angeles Tobarra-Gomez, University of Castilla-La Mancha, Spain

Soraphol Tulayasathien, Fiscal Policy Office, Ministry of Finance, Thailand

Editorial Assistants

Darawan Raksuntikul

Sorravich Kingsuwankul

Panit Buranawijarn

Editorial and Managerial Contact

c/o Mrs. Darawan Raksuntikul

Thammasat Review of Economic and Social Policy (TRESP)

Faculty of Economics, Thammasat University

2 Prachan Road, Bangkok 10200, Thailand

Tel. +66 2 696 5979

Fax. +66 2 696 5987

E-mail: [email protected]

Url: http://www.tresp.econ.tu.ac.th


Volume 4, Number 2, July-December 2018

Editorial Introduction 1

ARTICLES

Credit Guarantee Optimization of State-owned Enterprises 6


Investigating relationship between Government Spending and

Economic Growth: Public Spending and long-run GDP level

Kritchasorn Jarupasin 28

Linkage of Technological Innovation toward ICT base and

Economic Output in CLMV Region 76

Theara Chhorn


Volume 4, Number 2, July – December 2018

1

Editorial Introduction

In this issue, Chayanisa Chaisuekul from Fiscal Policy

Research Institute, Thailand, conducted an investigative

research on government credit guarantee, which can be

considered one of the fiscal instruments for public support on

infrastructure and public goods investment. Her article,

“Credit Guarantee Optimization of State-owned Entreprises”,

studies the risk profile of government credit guarantees in

Thailand which are overseen by the Public Debt Management

Office (PDMO) of the Ministry of Finance. Government credit

guarantees are particularly effective in cases where the

government is the best stakeholder to anticipate, control, and

minimize risk. They are best viewed as endogeneous risks

since the risks can potentially be influenced by government

policy.

The state-owned enterprises (SOEs) examined in this

study are categorized into financial institutions and non-

financial institutions. The paper studies the relationship

between default events and credit rating of these two types of

SOEs over the period 2009-2014. In the case of financial

institutions, a separate definition for a ‘default event’ was

utilized (focusing instead on the rollover and restructuring of

debt) as no financial institution SOEs failed to pay scheduled

debt service in that time period. Intuitively, this may be due to

the fact that a financial institution defaulting on a payment has

the potential to be much more disruptive than a non-financial

institution from the government perspective. The probability of default for financial and non-financial SOEs was estimated

using the hybrid forward model.

Government credit guarantees can entail both benefits and risks for the government. In Thailand, the role of the

PDMO is to manage the fiscal risk stemming from government

guarantees. The optimal credit guarantee values were



2

calculated using linear programming models of two types:

maximizing net benefits, and minimizing net expected loss.

The paper found that the two methods yield similar results.

The probability of default for both types of SOEs closely

follows rank-order in terms of credit rating. However, in

contrast to the literature, it was found that in the time period

studied, the probability of default for financial institution

SOEs was higher than that of non-financial institutions, for a

given credit rating. This is perhaps due to the turbulent time

period which was studied where the government of Thailand

embarked on many new initiatives which were not all

successful. Further studies may consider longer time periods

to further clarify the picture.

The second article of this issue is “Investigating

relationship between Government Spending and Economic

Growth: Public Spending and long-run GDP level” by

Kritchasorn Jarupasin from Office of the National Economic

and Social Development Board (NESDB). This article

analyses the impacts of fiscal policy using a Solow-type model

with transitional dynamics, allowing for persisting effects of

fiscal policy à la Gemmell, Kneller and Sanz (2016). Previous

studies have looked at, through various methods and with

various dependent and fiscal variables, the effects of fiscal

policy changes on economic growth. While findings reveal

evidence for both significant and non-significant effects on

short-run and long-run growth, the literature indicates that

careful design is required to isolate effects and derive results.

The article displays the effects of fiscal policy changes

in 38 countries, with 17 developing countries and 21 high-

income OECD countries. The data indicates that, on average,

high-income countries spend more (as a percentage of Gross

Domestic Product, GDP) on healthcare and social welfare,

while developing countries spend more on general public

services. Unweighted averages suggested that, as a percentage



3

of GDP, developing and high-income countries spend similar

amounts on education. Pooled mean group (PMG) estimator

with an autoregressive distributed lag (ARDL(1,1)) structure

was employed to analyse short-run dynamics and long-run

equilibrium relationships between fiscal change and growth.

There is evidence that an increase in the share of a particular

type of spending could improve the level of per capita GDP in

the long run for developing countries in the sample, for

example, in healthcare and in general public services. For

high-income countries, the model suggests additional

spending on education could boost long-term GDP levels.

However, how this additional spending is financed also has

implications for growth; the data suggests that in the high-

income country group, excise taxes play a large role in

determining the effectiveness of spending increases.

Overall, one important key point from the study

implies that increasing revenue through distortionary taxes

should be avoided, since it reduces economic growth rate.

Moreover, the growth impacts of fiscal changes vary by

different implicit financing elements. In developing countries,

the focus should be on increasing the share of spending on

general public services and healthcare.

The third article, “Linkage of Technological

Innovation toward ICT base and Economic Output in CLMV

Region,” by Theara Chhorn, Chiang Mai University, considers

the relationship between information and communication

technology (ICT) development in Cambodia, Lao PDR,

Myanmar and Vietnam, and economic growth and

development, with a view towards providing policy

recommendations.

Technological innovation and progress through

improvements in ICT has driven economic growth and

reduced poverty across the world. In the vein of similar studies

in the literature on technological progress and growth, ICT is



4

assumed to augment labour and capital productivity.

Estimation methods used include the fixed-effects and

random-effects model, the maximum-likelihood random

effects model, and feasible generalised least squares

estimation. The different approaches are used to obtain more

efficient and consistent estimates, as well as minimize the

effects of heteroskedasticity and serial autocorrelation.

Estimation of the impact of technological progress is

proxied in two ways in the study: (i) number of internet server

connections, and; (ii) imports of computers and other ICT-

related goods. The literature suggests that estimation of both

factors should be positively correlated with economic growth.

The study finds that, for the period 1995-2016, internet server

connections have a positive and significant relationship with

per capita GDP growth. However, there is opposite

relationship between imports of computers and ICT products,

i.e. increased imports lead to lower per capita GDP growth.

The study supports the notion that greater access to

technology has the potential to improve economic growth and

development. Policymakers should thus harness this

relationship to bolster domestic development, as well as take

steps to attract increased investment in technology. Further

research should, however, be conducted to identify why

greater imports of computer and ICT products are negatively

related to economic development in the CLMV region.


(TRESP) is a young biannual double-blind peer reviewed

international journal published in June and December. Its first

publication was in December 2015. The Faculty of

Economics, Thammasat University and the Editorial Team of

TRESP seek to provide an effective platform for reflecting

practical and policy-oriented perspectives that links the

academic and policymaking community. Having devoted to

our ‘knowledge-for-all’ philosophy so as to drive our society



5

forward, the Faculty decided that TRESP published in an open

access model. Authors are responsible for the published

articles. The views and opinions expressed in the articles do

not necessarily reflect those of the Editors and the Editorial

Board. For further information and updates on this journal, or to submit an article, please visit our website at

www.tresp.econ.tu.ac.th.

Euamporn Phijaisanit

Editor-in-Chief



6

Credit Guarantee Optimization of State-owned

Enterprises


Senior Researcher

Fiscal Policy Research Institute

[email protected]



7

ABSTRACT

The objective of this paper is to estimate the credit risk of state-

owned enterprises (SOEs) in the form of probability of default

(PD) and then use it to analyze credit guarantee optimization.

Estimation of the probability of default by the Hybrid Model

found that the estimated PD for both financial and non-

financial SOEs are ranked by credit rating grade (the rank

ordering property), except for the 3rd rating grade PD of non-

financial SOEs. Analysis of the optimal credit guarantee for

each SOE by Linear Programming model found that the results

of maximizing the net benefit and the results of minimizing the

net expected loss from credit guarantee are similar. Moreover,

the value of expected loss implies that the magnitude of credit

risk must be mitigated and managed with appropriate tools by

the Ministry of Finance.

Keywords: Government credit guarantee, Credit risk, State-

owned enterprises

JEL Classification: C61, H63



8

Introduction

1.1. Government credit guarantees

Government credit guarantees are a fiscal instrument of

government financial support for infrastructure and public

goods investment in cases where the government is the best

organization to anticipate risk, control risk exposure, and

minimize the cost of risk (IMF, 2005).

In Thailand, the Public Debt Management Office

(PDMO) of the Ministry of Finance has guaranteed credit

(loan and bond) to some state-owned enterprises (SOEs) since

20051. Its goals are to give SOEs access to finance at a lower

financial cost (loan interest rate) and to get a greater amount

of credit from financial institutions with a favorable borrowing

term because they can benefit from the government's high

credit standing. In addition, the government’s support for

SOEs also support economic growth and social development,

such as infrastructure development projects2 and supporting

farmers.3

1.2. Fiscal risk from government credit guarantee

Government credit guarantees create a contingent liability

or government obligation for the Ministry of Finance. The

Ministry of Finance must repay the outstanding guaranteed

loans of state-owned enterprises if those state-owned

enterprises default on the loans. It is the uncertainty as to whether the government will have to pay, and if so the timing

1 Under the Thailand's Public Debt Management Act of 2005 (B.E. 2548) 2 E.g. High-speed train project by the State Railway of Thailand (SRT) 3 E.g. Rice pledging scheme and agricultural credit for rural development

project by Bank for Agriculture and Agricultural Cooperatives (BAAC)



9

and amount of spending that is complicated for estimation of

fiscal risk management (IMF, 2005).

The fiscal risk from government credit guarantees are

classified as endogenous risks that are generated from

government activities or where the probability of the event can

be influenced by government actions (IMF, 2016).

Table 1. Government Credit Guarantee Fee Rate

Credit

rating

grade

Guarantee Fee Rate

(Percentage of outstanding guaranteed loan)

5 years

but 10 years

of loan

period

Government

agency - 0.01 0.05 0.10 0.15

SOE

1 0.01 0.05 0.10 0.15

2 0.05 0.10 0.15 0.20

3 0.10 0.15 0.20 0.25

4 0.15 0.20 0.25 0.30

5 0.20 0.25 0.30 0.35

6 0.25 0.30 0.35 0.40

7 0.30 0.35 0.40 0.45

8 0.35 0.40 0.45 0.50

Source: The Guarantee Fee Rate and Condition Ministerial

Regulation of 2008 (B.E. 2551)

Currently, the Ministry of Finance of Thailand partially

manages this fiscal risk by charging a fee for credit guarantees

from state-owned enterprises. The fee pricing is around 0.01-

0.5 %, depending on the credit rating grade4 and the loan

4 The credit rating grade has 8 levels from grade 1 (the lowest risk) to

grade 8 (the highest risk) that is evaluated by the PDMO every fiscal year.



10

period, which reflects the level of credit risk of each state-

owned enterprise. Nevertheless, the Ministry of Finance can

use the SOEs credit rating grade to evaluate fiscal risk

exposure from credit guarantee each fiscal year to take

advantage of the value of such contingent liabilities and to

determine the optimal value of credit guarantee for each SOE

which the Ministry of Finance has the highest net benefit from

credit guarantees, and the least risk from credit guarantees for

the SOE in each fiscal year.

Credit Risk of State-owned Enterprises

2.1. Credit risk from government credit guarantee

A financial instrument that is similar to government credit

guarantees are bank loans. The credit risk of bank loans

consists of three components, Exposure at default (EAD), Loss

given default (LGD), and Probability of default (PD). Then,

the magnitude of credit risk from credit guarantees is a

multiplied result of EAD, LGD, and PD (Naksakul, 2006;

Public Debt Management Office, 2014).

In the case of government credit guarantees, EADs are the

risk of disbursement that can be estimated from the value of

guaranteed loans outstanding, while LGDs are risk in

collateral value that can be estimated from a proportion of loss

value after recovery from collateral that is equal to one since

no collateral is required in this case. So, the variable that must

be estimated is the probability of default or default risk.

2.2. Simulated default for state-owned enterprises

Default events are important data for estimating the

probability of default. Standard & Poor’s Global Rating (2017)



11

and the Bank of Thailand (2016) define "default event" in the

general case as the event in which "the debtor or the counter-

party cannot pay back the loan principal or interest on the due

date contained in the original terms of a debt issue". However,

in practice, when referring to sovereign debt, including

government, state-owned enterprises, and other government

agency debt, both in the form of loan and bond, default events

never occur when considering the definition of default in the

general case. Especially in the case of Thailand if the state-

owned enterprise fails to pay the loan principal or interest on

the due date, the Ministry of Finance will give assistance such

as finding a new source of funds to repay original debts,

according to financial market conditions and financial status

of state-owned enterprises at that time.

Because of the special characteristics of SOE debt, it is

necessary to define a new default event as applied to SOEs to

calculated their PD. Default events for state-owned enterprises

are defined in this article according to Public Debt

Management Office (2014) which defined SOEs default

events as “when their liabilities greater than their assets in a

year, when they have EBITDA5 negative for three consecutive

years, when they received a credit rating grade as 8 in a year,

or when they received a credit rating grade as 7 for three

consecutive years.” Those conditions of the Public Debt

Management Office (2014) apply only to the non-financial

state-owned enterprise 6 ; none of the financial state-owned

enterprise have performance that meet these conditions.

5 EBITDA is Earnings before Interest, Taxes, Depreciation and

Amortization that calculated in the following manner: EBITDA =

Operating Profit + Depreciation Expense + Amortization Expense. 6 The total number of state-owned enterprises that are given credit rating

by the Public Debt Management Office in each fiscal year is 20 SOEs,

including 16 non-financial institutions and 4 financial institutions.



12

Therefore, the Public Debt Management Office (2014) can

estimate default probabilities only for non-financial SOEs.

To resolve that limitation to estimate all types of SOE’s

default probabilities, consider the characteristics of sovereign

debt. The Standard & Poor’s Global Rating (2017) defines a

default for sovereign debt as “when they fail to pay scheduled

debt service on the due date or tenders an exchange offer of

new debt with less favorable terms than the original issue, or

when their notes or bonds are converted into a new currency

of less than the equivalent face value.” The default definition

of Standard & Poor’s Global Rating (2017) focuses on the

characteristic of the loan agreement, not the SOE’s

performance like the default definition of the Public Debt

Management Office (2014). Thus, focusing on the

characteristics of a loan agreement is more appropriate to

define SOE’s default event.

Therefore, to properly define default events for Thailand’s

SOEs, consider S&P Global Rating’s default condition with a

framework for public debt management of the Ministry of

Finance of Thailand, in the case of state-owned enterprises. A

framework for public debt management found that the type of

SOE’s debt management, which meets the S&P Global

Rating’s default condition, is debt management with rollover

method7. Thus, this article defined a new definition of default

for SOE (Simulated default) as "when the SOE either issues

new bonds to repay original bond with less-favorable terms

than the original bond or negotiates with the bank creditors a

rescheduling of principal or interest at less-favorable terms

than in the original loan."

However, with limited access to SOE insights data, i.e.,

we did not have more details about the payment terms of the

pair of loan, which rollover or issue new bonds to repay

7 The SOE issue new bonds to repay original bond



13

original bonds. So, this article will use default condition as

“when the SOE manages its debt with rollover” to be a proxy

for our default condition as defined above.

Debt data on SOEs during the fiscal years 2009-2014 were

used in this study; it was found that this period has data on 110

SOEs where their credit was rated by Thailand's public debt

management office. Among them, it was found that there were

4,087 contracts for loans and bonds that were outstanding, and

restructured debt as rollover totaled 355 contracts that

presented in Table 2.8

Table 2. Number of SOE’s loan and bond that restructure

debt as rollover during the years 2009-2014

Fiscal

year

The number of

state-owned

enterprises that

PDMO's credit

rated

The number of

loan and bond

outstanding

The number of loan

and bond that

restructure debt as

a rollover

- Enterprises contracts contracts

2009 19 658 47

2010 17 705 54

2011 16 672 5

2012 18 683 95

2013 20 690 85

2014 20 679 69

Total 110 4,087 355

8 In practice, the public debt management office did not collect data about

payment structure between pairs of loan or bond that restructure debt as

rollover and their original issue, so this study uses the loan and bond that

restructure debt as rollover every case as a simulated default event instead

of the new definition of default for SOE which mentioned above. As a

result, this estimation will give a conservative probability of default.



14

Note: This data, includes both non-financial institutions and

financial institutions, and their debt, including all currencies.

2.3. Probability of default for state-owned enterprises

The nature of SOE debt is low default portfolios, which

limits default events and thus general estimation models

cannot estimate PD in the case of no default events occurring

and may yield an underestimation of credit risk. So, the

estimation of probability of default for SOE in this study will

use the model that is appropriate for SOE debt characteristics,

which is the Hybrid model of Roengpitya (2012). The Hybrid

model was based on two existing estimation approaches,

including the most prudent estimation of Pluto and Tasche

(2006) and the maximum likelihood of Forrest (2005). In this

study, we use the Hybrid model with forward method to

estimate PD and assume that there is no asset correlation

between SOEs. The PD estimates are split between SOEs that

are non-financial institutions and SOEs that are financial institutions.

First of all, let the general likelihood function of N rating

grade be defined as ℒ(𝑝1,…,𝑝𝑁) = ∏ ℒ(𝑝𝑖)𝑁𝑖=1 , where 𝑝1,…,𝑝𝑁

are the PD estimates for each rating grade. Let 𝑖 =1 be the lowest risk grade and 𝑖 = 𝑁 be the highest risk grade. In the non-financial institutions case, 𝑖 =1, 2, 3, … , 7 9 , while financial institutions rank from, 𝑖 =3, 4, 5, 6.10 Then, using the concept of the most prudent to collapse the rating grade –

assuming that the N rating grade satisfy the rank order

9 Total of PDMO's credit rating grade is eight grades, but this estimation

ignores grade 8 because this grade never rated to any state-owned

enterprise. 10 During the fiscal year 2009-2014, state-owned financial institutions

received a PDMO's credit rating in grade 3-6 out of the total grade (8

grades).



15

requirement, then we must have 𝑝1 ≤ 𝑝2 ≤ 𝑝3 ≤ ⋯ ≤ 𝑝𝑁 . And to find the upper bound of 𝑝1, the most prudent implies that the condition is 𝑝1 = 𝑝2 = 𝑝3 = ⋯ = 𝑝𝑁 = 𝑝 –in the general likelihood function, and we get the new likelihood

function:

ℒ(�⃗�𝑖) = 𝑝∑ 𝑘𝑗

𝑁𝑗=𝑖 (1 − 𝑝)∑ (𝑛𝑗−𝑘𝑗)

𝑁𝑗=𝑖 (1)

Next, let ℒ̅ = 𝑒(−0.5⋅𝜒2(∝,𝑁))

∙ ℒ(𝑀𝐿𝐸) where

ℒ(𝑀𝐿𝐸) = ∏ ℒ(𝑝𝑀𝐿𝐸𝑖 )𝑁𝑖=1 is the maximum likelihood value

that is evaluated at the estimated PD from the maximum

likelihood method and set confident level is 0.95 and degree

of freedom is N.11 The hybrid forward method begins with

solving for the best grade PD first so 𝑝1 solves

(𝑝1)∑ 𝑘𝑗

𝑁𝑗=1 (1 − 𝑝1)

∑ (𝑛𝑗−𝑘𝑗)𝑁𝑗=1 = ℒ̅ (2)

The PD estimates for other rating grades will be solved

through the following iterative process. The estimated PD for

grade 𝑖 = 2, …, N is 𝑝𝑖 that solves

𝓛(�⃗�𝑖) = (𝑝𝑖)∑ 𝑘𝑗

𝑁𝑗=𝑖 (1 − 𝑝𝑖)

∑ (𝑛𝑗−𝑘𝑗)𝑁𝑗=𝑖 =

ℒ̅

∏ ℒ(𝑝𝑀𝐿𝐸𝑙 )𝑖−1𝑙=1

(3)

This model can estimate the probability of default for both

financial and non-financial institution presented in Table 3 and

Table 4. This study can resolve previous limitations in studies

that only estimated PD for non-financial institutions because

of the new definition of simulated default for state-owned

11 The degree of freedom in case of non-financial institution and financial

institution is 7 and 4 respectively, according to the results of the credit

rating grade of the Public Debt Management Office in the past fiscal year

2009-2014.



16

enterprise and new model that is more appropriate with low

default portfolio than the earlier model.

Table 3. The estimated PD for Non-financial state-owned

enterprises (Percentage)

Model Hybrid forward MLE The most

prudent

Rating grade (i) 𝒑𝒊𝑵𝑩 𝒑𝑴𝑳𝑬

𝒊 𝒑𝑴𝑷𝒊

1 1.8202 2.0548 8.3220

2 2.1529 1.2987 8.6504

3 2.0951 3.8462 9.0805

4 10.4135 7.0485 10.7972

5 14.6874 12.6984 11.9110

6 15.2261 10.6816 11.7999

7 22.8343 14.8876 14.8876

In addition, this study found that the estimated PD for both

financial and non-financial are ranked by credit rating grade

(the rank ordering property), except the 3rd rating grade PD of

non-financial. However, the Hybrid forward PD (𝑝𝑖𝑁𝐵) has

failed to rank order condition less than the maximum

likelihood PD (𝑝𝑀𝐿𝐸𝑖 ).

In case of financial institutions, estimation of PD found

that all the Hybrid forward PD are ranked by credit rating

grade and the Hybrid forward model can estimate PD in rating

grade 6. The maximum likelihood model and the most prudent

model cannot estimate this because there is no default event

occurring at this rating grade. This is to be expected as it is

characteristic of Low default portfolios, hence the Hybrid



17

forward model can solve this characteristic of Low default

portfolios.

Table 4. The estimated PD for Financial state-owned

enterprises (Percentage)

Model Hybrid

forward MLE

The most

prudent

Rating grade (i) 𝒑𝒊𝑩 𝒑𝑴𝑳𝑬

𝒊 𝒑𝑴𝑷𝒊

3 12.0480 5.3571 12.0301

4 12.5684 7.4510 12.5337

5 19.9955 15.3846 15.1951

6 54.5415 n.a. n.a.

Note: During the fiscal year 2009-2014, state-owned financial

institutions received a PDMO's credit rating in grade 3-6 out of the

total grade (8 grades).

It should be noted that when we compare the estimated

PD for financial and non-financial SOEs at the same credit

rating grade, it was found that the financial institutions were

more likely to default on debt than the non-financial

institutions (see Figure 1), in spite of the fact that financial

institutions should have a lower default. That result may be

due to the data used during 2009 and 2014 which experienced

many changes in government policies and interventions, such

as intervention in the Bank for Agriculture and Agricultural

Cooperatives and Government Housing Bank. Even these

policies have ended, the liabilities and obligations from

implementing these policies remain with these banks since the

majority of loans and bonds that rollover debts restructure are

from these banks and these debts was from government

projects.



18

Figure 1. The estimated PD of financial and non-financial

state-owned enterprise

Thus, these estimated PD reflect the magnitude of the

credit risk of each state-owned enterprise which are important

data to risk management process of the Ministry of Finance in

the next section.

Credit Guarantee Optimization

3.1. The objectives of government credit guarantees and risk

exposure

Government credit guarantees to state-owned enterprises

create both benefits and fiscal risk for the Ministry of Finance.

Government credit guarantees were used as a fiscal tool for

financially supporting SOE investment in the production of

public goods and services and infrastructure investment.

1.82 2.15 2.1010.41

14.69

15.23

22.83

12.05 12.5720.00

54.54

1 2 3 4 5 6 7

rating grade

non-bank bank



19

Meanwhile, this government transaction creates a fiscal risk

and obligation to the Ministry of Finance as the guarantor.

The Public Debt Management Office (PDMO), the main

player in public debt management, should manage the fiscal

risk from government guarantees. Risk management may be

within the design of the guarantee system, including the

optimal credit guarantees, which the PDMO stands to have a

maximum benefit from while fulfilling the goals of the

government credit guarantee, while also experiencing

minimum fiscal risk, under the relevant legal framework.

Based on portfolio selection theory, the expected utility of

the investor is a function that depends on the expected return

and risk of investing. However, the PDMO is a government

agency that is not intended for profit, but it has the primary

purpose of managing public debt of the country. So, the net

benefit from credit guarantees of PDMO depends on achieving

the goals of government credit guarantees to SOEs against the

fiscal risk of this operation, which can be characterized as

𝑛𝑒𝑡 𝑏𝑒𝑛𝑒𝑓𝑖𝑡 = 𝑎𝑐ℎ𝑖𝑒𝑣𝑖𝑛𝑔 𝑡ℎ𝑒 𝑜𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒𝑠 − 𝑓𝑖𝑠𝑐𝑎𝑙 𝑟𝑖𝑠𝑘(4)

Consider the goals of government credit guarantees in

Thailand: one is to provide SOEs with access to finance at a

lower financial cost (loan interest rate) and another one is to

support those that have high credit risk to get access to finance

because they can benefit from the government’s high credit

standing. The fiscal risk from government credit guarantees is

the credit risk of SOEs that can be estimated in the form of the

value of the net expected loss from credit guarantees. Thus, the

net benefit from credit guarantees of PDMO is as the following

equation:



20

𝐵(𝑋𝑗) = {[∑ (𝑅𝑗𝑁𝐺 − 𝑅𝑗

𝐺)𝑋𝑗20𝑗=1 ] + ∑ 𝑋𝑗

20𝑗=1 } − {∑ 𝑃𝐷𝑗𝑋𝑗

20𝑗=1 −

∑ 𝑟𝑗20𝑗=1 𝑋𝑗}(5)

Where 𝑋𝑗 is guaranteed loan of state-owned enterprise 𝑗;

𝑗 =1, 2, 3,…, 20. Achieving the goals of government credit guarantees to SOEs is (i) to provide state-owned enterprises

with access to finance at a lower financial cost that is evaluated

by the spread of interest rate between non-government

guaranteed loan and government guaranteed loan (𝑅𝑗𝑁𝐺 − 𝑅𝑗

𝐺)

and (ii) to support those that have high credit risk to get access

to finance that is evaluated by the amount of guaranteed loan

of each state-owned enterprise (∑ 𝑋𝑗20𝑗=1 ). The net expected

loss from credit guarantees is the value of expected loss after

recovery by the total value of guarantee fee. The expected loss

from credit guarantees is the sum of multiple of the probability

of default (𝑃𝐷𝑗 )12 by guaranteed loan (𝑋𝑗 ) of state-owned

enterprise 𝑗 . Income from government guarantee fees is a multiple of the fee rate (𝑟𝑗)

13 by guaranteed loan (𝑋𝑗) of the

state-owned enterprise 𝑗.

3.2. Laws and regulations

Considering the optimal guaranteed loan of each state-

owned enterprise (𝑋𝑗∗), the value of guaranteed loan at which

the Public Debt Management Office receives a maximum the

net benefit from credit guarantees is subject to the relevant

laws and regulations. The relevant laws and regulations are as

follows:

12 Use the estimated PD from the previous section. 13 The fee rate is set by the Guarantee Fee Rate and Condition Ministerial

Regulation of 2008 (B.E. 2551), that fee rate varies by the credit rating

grade and loan period for each state-owned enterprise.



21

(i) A quasi-budget constraint: in each fiscal year the Ministry of Finance can guarantee no more than 20% of

the annual budget that is in force at that time14. This can

be expressed as ∑ 𝑋𝑗 ≤ 0.2𝐴20𝑗=1 where 𝐴 is the annual

budget that is in force at fiscal year 2014.

(ii) Fiscal Sustainability Framework: in each fiscal year the government can generate public debt outstanding no

more than 60% of GDP. This can be expressed as

∑ 𝑋𝑗 + 𝐵 ≤ 0.6 𝐺𝐷𝑃20𝑗=1 where 𝐵 public debt is

outstanding at the end of fiscal year 2013 and the

forecast of other public debt in fiscal year 2014.

(iii) Exchange rate risk requirement: in each fiscal year, the proportion of foreign debt to export income must not be

more than 9%15. This can be expressed as ∑ 𝑋𝑗𝑓 +20𝑗=1

𝐶 ≤ 0.09 𝐸𝑋𝑃 where 𝑋𝑗𝑓

is the value of guaranteed

loan in foreign currency of state-owned enterprise 𝑗 and 𝐶 is another foreign currency debt outstanding of the government at the end of the fiscal year 2013.

(iv) Leverage ratio: the Ministry of Finance can provide credit guarantees to each state-owned enterprise an

amount (𝑋𝑗) that, after summing with that SOE's other

debt (𝐷𝑗), does not exceed three times the capital of the

SOE (𝐸𝑗). In the case where the SOE is a Public Limited

Company (𝑗 = 9) the amount must not exceed six times the capital of the SOE, and similarly for SOEs which are

14 Section 28 of Public Debt Management Act of 2005 (B.E. 2548) 15 The Regulation of the Ministry of Finance on Public Debt Management

of 2006 (B.E. 2549)



22

financial institutions ( 𝑗 = 17, 18, 19, 20). This can be expressed as 𝐷𝑗 + 𝑋𝑗 ≤ 3𝐸𝑗 ; 𝑗 = 9 and 𝐷𝑗 + 𝑋𝑗 ≤ 6𝐸𝑗 ;

and 𝑗 = 17, 18, 19, 20 respectively.

Furthermore, the optimal guaranteed loan also considers

the demand for loan of each state-owned enterprise in a fiscal

year that can be expressed as 𝑋𝑗 ≤ 𝐿𝑑𝑗 where 𝐿𝑑𝑗 is the

demand for loan of state-owned enterprise 𝑗 in fiscal year 2014.

3.3. Optimization for government credit guarantees

The study of the optimization of credit guarantees of state-

owned enterprise using linear programming models is divided

into two groups: maximizing the net benefit from credit

guarantee, and minimizing the net expected loss from credit

guarantee. This optimization found that (i) subject to all the

relevant laws and regulations, and demand for fund of each

state-owned enterprise in that time, the results from

maximizing the net benefit and the results from minimizing

the net expected loss from credit guarantee are similar in terms

of optimal guaranteed loan distribution, the value of the net

benefit, and the value of the net expected loss from credit

guarantees (models 1.1 and 2.1). In addition, (ii) adding the total value of guaranteed loan, which happened in fiscal year

2014 as constraints of the optimization, it was found that high

risk enterprises have been reduced to a guaranteed amount, if

considered with a focus on the net benefit of PDMO, while

some enterprises are not guaranteed, if considered with a focus

on the net expected loss value (models 1.2 and 2.2).



23

Table 5. The optimal values from credit guarantee

optimization

Model Model 1.1 Model 1.2 Model 2.1 Model 2.2

Method Maximize Maximize Minimize Minimize

Objective

function

Net

Benefit

Net

Benefit

Net

Expected

loss

Net

Expected

loss

Constraints ≤ ≤ = =

1. Budget

constraints I √ √ √ √

2. Budget

constraints

II

√ √ √ √

3. Exchange rate

risk

constraints

√ √ √ √

4. Leverage

constraints √ √ √ √

5. Demand for

loan √ √ √ √

6. Total of credit

guarantee - √ - √

Optimal values (Million Baht)

Guaranteed

loan

SOE1 =

751.45

SOE1 =

751.45

SOE1 =

751.45

SOE1 =

751.45

SOE5 =

8,810.00

SOE5 =

8,810.00

SOE5 =

8,810.00

SOE5 =

8,810.00

SOE7 =

30,033.13

SOE7 =

15,790.15

SOE7 =

30,033.13

SOE7 =

30,033.13

SOE8 =

20,539.94

SOE8 =

20,539.94

SOE8 =

20,539.94

SOE8 =

20,539.94

SOE10 =

3,000.00

SOE10 =

3,000.00

SOE10 =

3,000.00

SOE10 =

3,000.00

SOE11 =

10,532.99

SOE11 =

10,532.99

SOE11 =

10,532.99

SOE11 =

290.00



24

Table 5. (Continued) Model Model 1.1 Model 1.2 Model 2.1 Model 2.2 Optimal values (Million Baht)

Guaranteed

loan

SOE13 =

3,000.00 SOE13 =

3,000.00 SOE13 =

3,000.00 -

SOE16 =

1,000.00 SOE16 =

1,000.00 SOE16 =

1,000.00 -

SOE17 =

285,360.90 SOE17 =285,360.90

SOE17 =285,360.90

SOE17 =285,360.90

Total of credit

guarantee 363,028.41 348,785.42 363,028.41 348,785.42

Net benefit 320,228.76 309,046.31 320,228.76 306,578.97 Net expected

loss 45,765.10 42,576.90 45,765.10 45,115.41

Conclusion and Implication

Estimation of default probability with the Hybrid Model

found that the estimated PD for both financial and non-

financial SOE is ranked by credit rating grade (the rank

ordering property), except the 3rd rating grade PD of non-

financial SOE. It should be noted that financial SOEs have a

higher estimated PD than non-financial SOEs at the same

rating grade despite the fact that financial institutions should

perhaps have a lower default risk than non-financial

institutions. However, data limitations must be acknowledged

as the estimation of PD in this study used data over a period of

six years, even if it is under the requirements specific for PD

Estimation of the Bank of Thailand (2012) that require use of

data of at least five years. In the future, a longer time series

data of SOE debts may be used, and thus the estimated PD may

be more consistent with financial theory, including satisfying

the rank order condition and the financial institution PD less

than the non-financial institution PD.

Moreover, estimated PD in this article represents the size

of the credit risk of each SOE's credit rating grade that



25

increases by each rating grade exponentially, while the

guarantee fee rate increases by each rating grade linearly. This

is so despite the fact that, in financial theory, the guarantor

should charge a guarantee fee equal to the default probability

of each borrower to mitigate their risk exposure. This implies

that the Public Debt Management Office just manages partial

risk from a guarantee. However, in the case of SOEs, the

Public Debt Management Office cannot charge a guarantee fee

more than as prescribed in the Guarantee Fee Rate and

Condition Ministerial Regulation of 2008 (B.E. 2551) and

increasing the fee contradicts the goal of credit guarantee to

SOE, which helps SOE to reach the financial source with a

lower cost.

Nevertheless, the Public Debt Management Office can use

this estimated default probability to recognize size of credit

risk from their credit guarantees to better manage risk, for

example counter-guarantees funds. Moreover, estimated

default probability can be used to enhance the efficiency of

guarantee allocations by using PD as a criterion in the decision

about the optimal value of guaranteed loans for each state-

owned enterprise.



26

References

Bank of Thailand. (2012). Notification of the Bank of Thailand

Re: Rule of Credit Risk Internal Ratings-Based

Approach by Internal Ratings-Based Approach.

Bangkok: Bank of Thailand.

Bank of Thailand. (2016). Notification of the Bank of Thailand

Re: Rules for Credit Ratings and Reserves Requirement

of Financial Institutions. Bangkok: Bank of Thailand.

Forrest, A. (2005). Likelihood Approaches to Low Default

Portfolios. Discussion paper, Dunfermline Building

Society, September 2005.

International Monetary Fund. (2005). Government Guarantees

and Fiscal Risk. Prepared by the Fiscal Affairs

Department (In consultation with other departments).

Approved by Ter-Minassian, Teresa.

International Monetary Fund. (2016). Analyzing and

managing fiscal risks-Best Practices.

Naksakul, P. (2006). Credit Risk Models in Relation to Capital

Reserve Framework of Financial Institutions. Bangkok:

Bank of Thailand.

Pluto, K., & Tasche, D. (2006). Estimating Probabilities of

Default for Low Default Portfolios. In B. Engelmann &

R. Rauhmeier (Eds.), The Basel II Risk Parameters:

Estimation, Validation, and Stress Testing (pp. 79–103).

Berlin, Heidelberg: Springer Berlin Heidelberg.

https://doi.org/10.1007/3-540-33087-9_5

Public Debt Management Office. (2014). Improvement of Credit Rating model for State-owned Enterprises.

Bangkok. Ministry of Finance, Thailand.



27

Roengpitya, R. (2012). Proposal of New Hybrid PD

Estimation Models for The Low Default Portfolio

(LDPs), Empirical Comparisons and Policy

Implications. Discussion paper. Bangkok: Bank of

Thailand.

Standard & Poor's Global Rating. (2017). Default, Transition,

and Recovery: 2016 Annual Sovereign Default Study

and Rating Transitions



28

Investigating relationship between Government

Spending and Economic Growth: Public

Spending and long-run GDP level

Kritchasorn Jarupasin*

Plan and Policy Analyst

Office of the National Economic and Social Development

Board

Office of the Prime Minister

Thailand

[email protected]

* The contents of this article do not reflect the author’s affiliation. This

article is a part of the author’s doctoral thesis submitted to the University

of Exeter. The author thanks Professor Gareth D. Myles, Professor Ben

Zissimos, Professor Norman Gemmell and Dr. Xiaohui Zhang for their

kind advice and support during his doctoral study. Sincere gratitude must

also be extended to the Royal Thai Government for the research grant. All

remaining errors rest with the author.



29

ABSTRACT

This study investigates the relationship between public

spending and long-run GDP per capita. While most fiscal-

growth studies put emphasis on the relationship in the public-

policy endogenous growth model, this analysis allows for

Solow-type transitional dynamics where the effects of fiscal

policy can be persistent. Moreover, the long-run and short-run

effects of fiscal changes are identified separately in this

analysis using the groups of countries comprising 38 countries

(17 developing countries and 21 high-income OECD

countries). Our results show that an increase in total spending

which is financed by non-distortionary taxes only enhances the

level of GDP per capita in high-income OECD countries. With

a given level of total spending, increases in the shares of

healthcare and general public services spending can improve

the levels of GDP per capita in developing countries. On the

other hand, increasing the share of education spending in a

high-income OECD country is conducive to increasing the

level of GDP per capita

Keywords: Fiscal policy, Economic growth, Public

expenditure, Government

JEL Classification: E62, H50, O23, O47



30

1. Introduction

The fiscal-growth studies in a number of studies focus on

the public-policy endogenous growth model. In those studies,

permanent growth effects of fiscal changes are analysed

without transitional dynamics. The analysis of fiscal policy

impacts in this study allows for Solow-type transitional

dynamics, but the effects of fiscal policy may be persistent

according to the framework proposed by Gemmell, Kneller

and Sanz (2016).

Gemmell et al.’s (2016) study was motivated by the recent

fiscal stimulus enacted after 2009 in order to counteract the

global financial crisis. Governments’ spending choices in

these short-term packages are partially influenced by their

ambitions to comply with long-term growth objectives. With

this policy design, there are two different questions to be

addressed: how forceful is the evidence that long-run income

levels or growth rates react to changes in public spending, and

if they do, which expenditure types produce most considerable

impacts? In the following section, we attempt to respond to

these questions similar to Gemmell et al. (2016) by looking at

both developing countries and high-income OECD countries.

2. Literature review

In terms of the period of study, recent studies on fiscal

policy and long-run size of economy (either level of GDP or

rate of growth) include recent data, especially the Acosta-

Ormaechea and Morozumi’s (2013) study which uses data

from 1970 to 2010. Other studies (Afonso & Jalles, 2014;

Arnold et al., 2011; Gemmell, Kneller & Sanz, 2011; Gemmell

et al., 2016; Xing, 2012) also cover periods from the 1970s

until 2010. Ojede and Yamarik (2012) focus on an earlier

period; 1967 to 2008.



31

In terms of the sets of fiscal variables used, these studies

either put emphasis on tax policy (Arnold et al., 2011; Ojede

& Yamarik, 2012; Xing, 2012), public spending (Gemmell et

al., 2016), or both types of variables at once (Afonso & Jalles,

2014; Gemmell et al., 2011). In addition, Afonso and Jalles

(2014) look at both functional and economic classes of fiscal

variables according to the Government Finance Statistics

(GFS) definitions provided by the International Monetary

Fund (IMF).

As well as being categorised by their focus (on revenue

and/or expenditure), the effects of fiscal changes can be

classified by their impact on the size of economy (short-run or

long-run impact). Some of these studies focus only on

permanent growth effects (Acosta-Ormaechea & Morozumi,

2013; Afonso & Jalles, 2014), while others distinguish

between the long-run and short-run impacts of changes in

fiscal variables (Arnold et al., 2011; Gemmell et al., 2011;

Gemmell et al., 2016; Ojede & Yamarik, 2012; Xing, 2012).

Our study pays specific attention to the latter set of studies.

While many studies that differentiate between the long-

run and short-run effects of fiscal change capture the size of an

economy by using the growth rate of GDP or the growth rate

of GDP per capita, Arnold et al. (2011), Xing (2012) and

Gemmell et al. (2016) use the level of per capita GDP.

Gemmell et al. (2016) claim that using this specification is

advantageous since it allows the degree of persistence in GDP

growth responses to be identified by the data, rather than by

using a functional form incorporating permanent effects. For

this reason, our study will focus on the impact on level of GDP

per capita.

The three studies (Arnold et al., 2011; Gemmell et al.,

2016; Xing, 2012) referred to above use cross-country data,

whereas Ojede and Yamarik (2012) evaluate the growth

effects of tax policy at state level. Instead of investigating the



32

growth effects of fiscal policy, Lamartina and Zaghini (2011)

test the validity of Wagner’s law in high-income OECD

countries.

Although there are differences in the model specifications

for investigating change in fiscal composition and their effects

on either the level of GDP or economic growth, the findings

are, to a certain degree, harmoniously aligned. We go on to

discuss previous findings, econometric methods, variables

included in the model, the role played by budget constraint,

and other econometric issues in these studies. 2.1. Previous findings

This strand of literature, like the permanent growth effects

of fiscal change studies in endogenous growth model, mainly

considers high-income countries and, more specifically, high-

income OECD countries (Arnold et al., 2011; Gemmell et al.,

2011; Gemmell et al., 2016; Xing, 2012). Other studies, e.g.

Acosta-Ormaechea and Morozumi (2013) and Afonso and

Jalles (2014), consider a wider set of countries.

Some studies find the reallocation of fiscal composition to

be robustly related to long-run growth or GDP level, while

others do not. In order to understand this incongruity clearly,

we need to take several aspects of the preceding results into

consideration. Firstly, there are two different types of fiscal

variables being considered, namely public expenditure and

public revenue. Secondly, we need to consider the way in

which an increase in public expenditure is financed. We

previously refer to this as an implicit financing element. For

example, Gemmell et al. (2016) find that an increase in total

spending enhances GDP per capita level in the long run when

financed by non-distortionary taxes. Thirdly, fiscal variable

classifications can be interpreted differently when we analyse

the impacts of changes in these variables on GDP or growth of



33

GDP. This depends on the aspect of fiscal change we need to

evaluate in order to assess its impact. The following

paragraphs summarise the key findings of the papers

mentioned earlier.

Arnold et al. (2011) find that shifting taxes on income

towards consumption and immovable property enhances long-

run GDP per capita. In particular, increasing revenue by

raising current taxes on immovable property and consumption

is least harmful to growth. Arnold et al.’s (2011) findings are

supported by Xing (2012), suggesting that shifting tax revenue

away from corporate income, personal income, and

consumption taxes, and towards property taxes is associated

with a higher level of income per capita in the long run. When

investigating state-level data, Ojede and Yamarik (2012)

obtained different results from Arnold et al. (2011) and Xing

(2012). They found that increases in sales and property taxes

reduce long-run real income growth.

Gemmell et al. (2011) observe that the growth effects of

fiscal policy in the short run appear to persist. Although some

fiscal variables only have transitory effects, others might have

persistent growth effects. However, the positive growth effects

associated with productive spending are often counteracted by

the negative effects of tax changes.

Gemmell et al. (2016) raise awareness of the significance

of financing methods for increasing any type of public

expenditure when determining long-run GDP level. By using

pooled mean group estimators (PMG) with contemporaneous

correlation, they find robust long-run positive effects on GDP

per capita levels for reallocating total spending towards

transportation and communication, and education spending.

In contrast, Afonso and Jalles (2014) find that an increase

in government revenue has no significant impact on growth.

Moreover, the coefficients of government expenditures appear

to have highly significant negative signs.



34

Acosta-Ormaechea and Morozumi (2013) find that an

increase in education spending offset by a fall in social

spending seems to be robustly related to higher growth rates.

These results also hold true at the general government level.

Their results also show that education spending promotes

growth as well as public capital does in the long run.

2.2. Improvement of econometric methods

Recent developments in data collection have improved

the availability of data, so it has become possible to investigate

the compositional change of public spending and its impact on

long-run GDP per capita level or growth.

From above reason, the updated data can be used under

the assumptions of short-run heterogeneity and long-run

homogeneity. This econometric method proposed by Pesaran

Shin, and Smith (1999) is pooled mean group estimators

(PMG). It is a compromise between the fixed effects model

and the mean group estimator (MG). While intercept, short-

run coefficients and error variances are allowed to differ across

groups, the long-run coefficients are equal. This method has

been analysed by Arnold et al. (2011), Gemmell et al. (2011),

Ojede and Yamarik (2012), Xing (2012), and Gemmell et al.

(2016).

2.3. Variables included in recent studies

Dependent variable

The choice of dependent variable is distinctly separable

between growth of GDP and level of GDP. While most studies

use the growth rate of either GDP or GDP per capita, Arnold

et al. (2011) use a change in log of GDP per capita and a

change in log of total factor productivity (TFP) of a given firm.

Similarly, Xing (2012) also uses a change in log of real GDP



35

per capita as a dependent variable. While Gemmell et al.

(2011) use a change in the growth rate of GDP in one of their

studies, the level of GDP per capita is employed by Gemmell

et al. (2016). The dependent variable in Ojede and Yamarik

(2012) is the change in growth rate of real income.

It is important to note that all of these studies (Arnold et

al., 2011; Gemmell et al., 2011; Gemmell et al., 2016; Ojede

& Yamarik, 2012; Xing, 2012) estimate the results with an

error correction model. When interpreting the results, we need

to refer back to the equations in terms of the autoregressive

distributed lag model: i.e. the long-run level of GDP per capita

impact from fiscal change is analysed in Xing (2012), rather

than the growth effect (change in log of real GDP per capita).

Afonso and Jalles (2014) use the real growth rate of GDP

per capita, and Acosta-Ormaechea and Morozumi (2013)

select the growth of output per capita.

Fiscal variables

Different classes of expenditure and revenue can be

considered. Two broad categories of each type of fiscal

variables are included, following the example of Kneller,

Bleaney and Gemmell (1999) and based on the framework

proposed by Barro (1990); namely productive expenditure,

unproductive expenditure, distortionary taxes and non-

distortionary taxes. We will now describe some of the fiscal

variables included in recent studies.

Arnold et al. (2011) focus on tax structures which can be

classified mainly into income taxes, consumption taxes and

property taxes. Ojede and Yamarik (2012) and Xing (2012)

also emphasise on the composition of tax revenues.

Gemmell et al. (2011) use broad categories of revenue and

expenditure: productive expenditure, non-productive

expenditure, distortionary taxes and non-distortionary taxes.



36

Gemmell et al. (2016) utilise broad categories of revenue

similar to those used by Gemmell et al. (2011) and functional

classifications of public expenditure, namely transportation

and communication, education, health and housing etc.

Afonso and Jalles (2014) focus on both aggregate levels

and GFS (Government Finance Statistics) classifications of

fiscal variables. This includes functional and economic

classifications of both government expenditure and revenue.

Acosta-Ormaechea and Morozumi (2013) also look at both

economic and functional classifications of public expenditure.

Economic classifications include the compensation of

employees, other expenses and the net acquisition of non-

financial assets. Functional classifications include defence,

transportation and communication, health, education, and

social protection expenditures.

Non-fiscal control variables

A number of factors can be used as non-fiscal control

variables. The criteria used to decide which variable should be

chosen are highly dependent on the type of question or

particular model being investigated. Since using pooled mean

group estimators limits the number of control variables due to

a decrease in the degree of freedom, this strand of literature

often only includes a few non-fiscal control variables in

analyses.

Arnold et al. (2011) and Xing (2012) include investment

rate, human capital and population growth. Gemmell et al.

(2011, 2016) use investment rate and employment growth.

Like Gemmell et al. (2011, 2016), Ojede and Yamarik (2012)

include growth in private employment and private investment

share in their set of non-fiscal control variables.

While Afonso and Jalles (2014) use population growth,

investment, education and trade openness, Acosta-Ormaechea



37

and Morozumi (2013) include initial GDP per capita and initial

human capital in their set of non-fiscal control variables.

2.4. The role of government budget constraint

Government budget constraint needs to be considered in

order to avoid the production of invalid results due to biases

occurring as a result of not including both revenue and

spending variables in the same equation. Recent studies take

government budget constraint into account while avoiding

perfect multicollinearity in accordance with the specification

in Kneller et al. (1999). As a result, the important role played

by the implicit financing element is highly relevant in this

analysis.

2.5. Other econometric issues

There are also other econometric issues which should be

addressed, such as endogeneity and a robustness check.

Firstly, Afonso and Jalles (2014) investigate the

robustness of their results by adding variables (labour force

participation and unemployment rates) into their baseline

regression. Similarly, Acosta-Ormaechea and Morozumi

(2013) add inflation, openness, population growth and terms

of trade growth into their original set of control variables.

Different specifications, including lagged fiscal variables and

the different developmental levels of countries in the sample,

might also be considered.

Secondly, using pooled mean group estimators to analyse

the error-correction model requires some tests as prerequisites.

Gemmell et al. (2016) tested the order of integration and

cointegration, autoregressive distributed (ARDL) lag

structure, and weak exogeneity. They found that their

variables are best treated as non-stationary. Imposing two-lags

of ARDL tends to strengthen the case for significant causal



38

effects from a number of public spending categories on the

level of long-run GDP per capita. Their estimated results offer

relatively strong support for the theory that expenditure share

variables can be considered to be weakly exogenous, allowing

interpretation of the estimated long-run expenditure

parameters as capturing causal effects on GDP. The issues

tested in Gemmell et al. (2016) will be further investigated in

Section 4 of our study. Public expenditure composition is

analysed in the next section.

3. Public expenditure composition of countries in our

sample

In this section, we analyse the data on public expenditure

composition for 38 selected countries in our sample according

to the availability of control variables, which is mainly

affected by labour growth. The set of countries in Table 1 is

divided into two main groups: 17 developing countries and 21

high-income OECD countries.

As shown in Figure 1, total public spending in the group

of countries in our sample has slightly increased in the past

four decades. This can be seen from the increase in unweighted

10-year average total spending to GDP from 23.42% in 1972-

1981 to 25.44% in 2002-2011 for developing countries. For

high-income OECD countries, the level of total public

spending to GDP increased from 32.55% in 1972-1981 to

35.19% in 2002-2011. Public spending in high-income OECD

countries increased significantly during the 1970s and 1980s

but subsided in later periods. On the other hand, the proportion

of government spending to GDP in developing countries

increased consistently during 1992-2001 and 2002-2011.



39

Table 1. List of countries in our study by group

Developing countries High-income OECD countries

Bolivia Morocco Austria* Korea, Rep.

Brazil South Africa Canada* Luxembourg*

Cameroon Thailand Chile Netherlands*

Costa Rica Tunisia Denmark* New Zealand*

Dominican Republic Turkey Finland* Norway*

Egypt, Arab Rep. Nepal France* Portugal

India Hungary Spain*

Indonesia Iceland* Sweden*

Iran, Islamic Rep. Ireland UK*

Malaysia Israel United States*

Mauritius Italy

Note: Our 14 OECD countries included in Gemmell et al.’s (2016) group of 17

OECD countries

3.1. The composition of public spending

Table 2 presents the average amount of particular types of

public spending by groups of countries as percentages of GDP.

In percentage terms, government spending in our sample of

high-income OECD countries is obviously higher than in our

sample of developing countries. The same also applies to

many other types of spending, although not to spending on

general public services. The level of spending on education as

a share of GDP is relatively similar across different groups of

countries in the sample, with an average of around 3.38% of

GDP.



40

Figure 1. Unweighted 10-year averages of total public

spending as percentages of GDP for groups of countries in

our sample (1972-2011)

Table 2. Unweighted averages of public spending by type as

percentages of GDP for groups of countries in our sample

(1972-2012)

Developing

countries

High-income

OECD countries

Total spending 23.96 35.74

General public services 3.80 2.73

Defence 2.07 2.41

Transportation and communication 1.42 1.64

Education 3.38 3.39

Healthcare 1.45 3.62

Social welfare 3.13 12.59

0

5

10

15

20

25

30

35

40

1972-1981 1982-1991 1992-2001 2002-2011

Developing OECD



41

The composition of public spending in different groups of

countries varies depending on the policies and problems that

particular governments encounter. The unweighted averages

of public spending by type as percentages of total spending

from 1972 to 2012 for developing countries and high-income

OECD countries are shown in Table 3. In developing

countries, general public services (16.2%), education (14.1%)

and social welfare (11.9%) spending are crucial elements of

government budgets. In contrast, high-income OECD

countries spend a large proportion of public expenditure on

social welfare (34.7%), healthcare (10.1%) and education

(9.7%). Social welfare spending accounts for more than a third

of total public spending in high-income OECD countries.

Table 3. Unweighted averages of public spending by type as

percentages of total spending for groups of countries in our

sample (1972-2012)

Developing

countries

High-income

OECD countries

General public services 16.20 7.65

Defence 8.69 7.30

Transportation and communication 6.51 4.71

Education 14.10 9.67

Healthcare 6.23 10.05

Social welfare 11.93 34.72

The public spending composition of developing countries

is presented in Figure 2 using unweighted 10-year averages for

the period from 1972 to 2011. It is clear that social welfare

spending as a proportion of total public spending has increased

significantly over time. In contrast, spending on defence, and



42

transportation and communication has decreased significantly

relative to other types of public spending.

The proportions of most spending types, including general

public services, education and social welfare spending, in

relation to total spending in high-income OECD countries

have not changed dramatically in the past forty years as can be

seen in Figure 3. Healthcare spending has increased more

noticeably over time than other types of expenditure, whereas

spending on defence has been decreasing.

Figure 2. Unweighted 10-year averages of spending by type

as percentages of total spending for developing countries in

our sample (1972-2011)

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1972-19811982-19911992-20012002-2011

General public services

Defence

Transportation and

communication

Education

Healthcare

Social welfare



43

Figure 3. Unweighted 10-year averages of spending by type

as percentages of total spending for high-income OECD

countries in our sample (1972-2011)

The following subsection explains the estimation method

used to analyse long-run relationship between fiscal variables

and GDP per capita.

4. Public spending and long-run GDP per capita

investigating heterogenous panel data: estimation method

In this section, we discuss the econometric methods used

to study the relationship between public spending and long-

run levels of GDP per capita in our sample’s groups of

countries. Later (in Section 5), we present the estimates

separately, according to the country groupings, i.e. developing

countries and high-income OECD countries. This section

includes the discussion of pooled mean group estimator

(PMG), and tests for cointegration and ARDL lag structure.

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

1972-19811982-19911992-20012002-2011

General public services

Defence

Transportation and

communication

Education

Healthcare

Social welfare



44

4.1. Pooled mean group (PMG) estimator

The endogenous growth model in Devarajan, Swaroop

and Zou (1996) captures the permanent growth effects from

fiscal changes without transitional dynamics (Gemmell et al.,

2016). Allowing for Solow-type transitional dynamics while

the effects of fiscal change may be persistent requires a more

flexible functional form than that of Devarajan et al. (1996).

Using an autoregressive distributed lag (ARDL) model

parameterised in error correction form in Gemmell et al.

(2016) allows both the short-run dynamic and the long-run

equilibrium relationships between GDP and fiscal variables to

be identified separately. The ARDL(p,q) specification is:

𝑦𝑖,𝑡 = ∑ 𝛼𝑖,𝑗𝑦𝑖,𝑡−𝑗 + ∑ 𝛽𝑖,𝑗𝑋𝑖,𝑡−𝑗 + 𝜇𝑖 + 𝜀𝑖,𝑡𝑞𝑗=0

𝑝𝑗=1 (1)

where Xi,t-j includes all explanatory variables. Equation (1) can

be expressed in error correction form:

𝑔𝑖,𝑡 = ∆𝑦𝑖,𝑡 = ∅𝑖(𝑦𝑖,𝑡−1 − 𝛽𝑖𝑋𝑖,𝑡) + ∑ 𝛼𝑖,𝑗∗ ∆𝑦𝑖,𝑡−𝑗

𝑝−1𝑗=1

+ ∑ 𝛽𝑖,𝑗∗𝑞−1

𝑗=0 ∆𝑋𝑖,𝑡−𝑗 + 𝜇𝑖 + 𝜀𝑖,𝑡 (2)

where ∅i captures the error correcting speed of adjustment and βi captures the long-run equilibrium relationship between y

and X with short-run effects measured by β*i,j. The estimates

of long-run coefficient βi are not affected by the choice

between Xi,t and Xi,t-1 in determining the long-run relationship.

While Arnold et al. (2011), Ojede and Yamarik (2012), and

Xing (2012) use Xi,t, Gemmell et al. (2011, 2016) prefer Xi,t-1.

We use Xi,t in our study, since it provides better computational

convenience in our statistical package than using Xi,t-1.

Blackburne and Frank (2007) suggest several approaches

which can be taken in order to estimate Equation (2). Firstly,



45

a fixed effect (FE) estimation approach could be used when

data from each group is pooled and only the intercepts are

allowed to differ across groups. Pesaran and Smith (1995)

show that these regressions are likely to be biased if the

assumption of homogeneity of the short-run parameter

estimates across countries is rejected. Secondly, the model

might be fitted separately for each group and the arithmetic

average of coefficients could be calculated by using mean

group estimators (MG). The MG estimators allow both short

and long-run parameter heterogeneity. Thirdly, Pesaran et al.

(1999) proposed a PMG estimation that combines both

methods of pooling (FE) and averaging (MG). The intercept,

short-run coefficients and error variances are allowed to differ

across groups, but the long-run coefficients are constrained to

be equal across groups. Furthermore, Pesaran et al. (1999)

have also demonstrated that allowing for short-run parameter

heterogeneity results in more reliable estimates of the long-run

responses.

We present the results of PMG estimates, as the Hausman

test prefers PMG to MG.1 The implication of the results from

the Hausman test is that the assumption of homogenous long-

run parameter estimates across countries is valid. The PMG

method selected is then comparable to Gemmell et al.’s (2016)

study.

Our study investigates the long-run relationship between

public spending and the GDP per capita level of the 38

countries (see Table 1) which are classified as developing

countries (17 countries) and high-income OECD countries (21

countries). These groups of countries were selected based on

the availability of control and fiscal variables. The groups of

developing countries and high-income OECD countries in our

1 We do not show the results of the Hausman test in this paper; however,

they could be provided by request.



46

sample have been analysed separately, in a similar way to

Gemmell et al. (2016), by looking at the effects of total public

expenditure and public expenditure composition. The study

period is 1972-2012.

Our dependent variable gi,t is the change in log of GDP

per capita. Although the growth rate of per capita GDP is the

dependent variable, as shown in Equation (2), the regression

measures the impacts of fiscal and other variables on long-run

per capita GDP level. Equation (2) is only a re-

parameterisation of Equation (1). As discussed earlier,

Gemmell et al. (2016) argue that using level specification

allows the identification of the degree of persistence in GDP

growth responses.

The non-fiscal control variables included in this study are

labour force growth (LG) and investment ratio to GDP (K).

Labour force growth before 1990 is assumed to be constant

(from the average of available data) in a number of countries

in the sample where accurate data is not readily available.

When taking government budget constraint into account, our

fiscal control variables include the ratio of total expenditure to

GDP, distortionary taxes to GDP, non-distortionary taxes to

GDP and budget surplus to GDP. In the cases where we

consider public spending composition, the expenditure share

of a particular type of public spending in relation to total public

spending is added individually. The list of variables included

in this study is shown in Table 4.



47

Table 4. List of variables for this study Variables Description of the variables

y

K

LG

tot_gdp

distax_gdp

tgs_gdp

SURBP

TOT

gps_tot

def_tot

trc_tot

edu_tot

hea_tot

soc_tot

Log of GDP per capita (2005 USD)

Gross capital formation (% of GDP)

Labour force growth (%)

Total public spending (% of GDP)

Distortionary taxation (% of GDP)

Non-distortionary taxation (% of GDP)

Budget balance (% of GDP)

Total public spending in local currency unit

Spending on general public services (% of

TOT)

Spending on defence (% of TOT)

Spending on transportation and

communication (% of TOT)

Spending on education (% of TOT)

Spending on health (% of TOT)

Spending on social welfare (% of TOT)

The first part of the analysis of each group of countries in

our sample looks at the total public expenditure effect with

four different implicit financing elements: budget deficit;

distortionary taxes; non-distortionary taxes; and a mix of both

distortionary and non-distortionary taxes. In the second part of

the analysis, we use budget deficit as an implicit financing

element, focussing on the impact of shifting expenditure

towards a particular type of public spending composition on

the long-run GDP per capita level. There are two groups of

countries consider

THAMMASAT REVIEW OF ECONOMIC AND SOCIAL POLICY › paper › vol4-2 › 01_TRESP Vol.4... ·...

Documents

Transcript of THAMMASAT REVIEW OF ECONOMIC AND SOCIAL POLICY › paper › vol4-2 › 01_TRESP Vol.4... ·...