Bilateral effects between health expenditures, health outcomes and economic growth : evidence from time series and panel granger non-causality tests

(1)

uef.fi

PUBLICATIONS OF

THE UNIVERSITY OF EASTERN FINLAND Dissertations in Social Sciences and Business Studies

ISBN 978-952-61-2649-4 ISSN 1798-5749

The current thesis has attempted to explain the potential bilateral impacts between investment in human capital resources in the formation of health and national income.

From theoretical point of view this relationship can be explained in both directions, but the most previous empi

rical studies have focused only on one side of causal rela

tionships, while ignoring the adverse plausible effects. This dissertation used the most appropriate econometrics tech

niques of Granger causality to prepare precise information about the possible bilateral causal relationship between the major health variables – included health care expenditure

per capita, HIV/AIDs, child health and life expectancy at older ages – and economic growth. The results showed that bilateral relationship is the predominant and the results of

previous studies are missspecified.

ARSHIA AMIRI

DISSERTATIONS | ARSHIA AMIRI | BILATERAL EFFECTS BETWEEN HEALTH EXPENDITURES... | No 161

Dissertations in Social Sciences and Business Studies

PUBLICATIONS OF

THE UNIVERSITY OF EASTERN FINLAND

ARSHIA AMIRI

BILATERAL EFFECTS BETWEEN HEALTH EXPENDITURES, HEALTH OUTCOMES AND ECONOMIC GROWTH:

EVIDENCE FROM TIME SERIES AND PANEL

GRANGER NON-CAUSALITY TESTS

(2)

(3)

BILATERAL EFFECTS BETWEEN HEALTH EXPENDITURES, HEALTH OUTCOMES AND

ECONOMIC GROWTH: EVIDENCE FROM TIME SERIES AND PANEL GRANGER NON-

CAUSALITY TESTS

(4)

(5)

Arshia Amiri

BILATERAL EFFECTS BETWEEN HEALTH EXPENDITURES, HEALTH OUTCOMES AND

ECONOMIC GROWTH: EVIDENCE FROM TIME SERIES AND PANEL GRANGER NON-

CAUSALITY TESTS

Publications of the University of Eastern Finland Dissertations in Social Sciences and Business Studies

No 161

University of Eastern Finland Kuopio

2017

(6)

Grano Oy Jyväskylä, 2017

Sarjan vastaava toimittaja: Kimmo Katajala Sarjan toimittaja: Eija Fabritius Myynti: Itä-Suomen yliopiston kirjasto

ISBN: 978-952-61-2649-4 (nid.) ISBN: 978-952-61-2650-0 (PDF)

ISSNL: 1798-5749 ISSN: 1798-5749 ISSN: 1798-5757 (PDF)

(7)

5 Amiri, Arshia

Bilateral Effects between Health Expenditures, Health Outcomes and Economic Growth: Evidence from Time Series and Panel Granger non-causality Tests

Joensuu: Itä-Suomen yliopisto, 2017

Publications of the University of Eastern Finland Dissertation in Social Sciences and Business Studies ISBN: 978-952-61-2649-4 (print)

ISSNL: 1798-5749 ISSN: 1798-5749

ISBN: 978-952-61-2650-0 (PDF) ISSN: 1798-5757 (PDF)

ABSTRACT

The major research questions of dissertation are related to the potential bilateral positive impacts between investment in human capital resources in the formation of health and national income. From theoretical point of view these relationships can be either in both directions. However the empirical results are inconclusive. To prepare precise information about the predictable relationships between health variables and economic activity, the method of Granger non-causality tests is used in this context. Studies 1 and 2 examine time series and panel data Granger causality between health care expenditure and Gross Domestic Product per capita (GDPc) in OECD countries in years 1970 – 2012. The analysis is conducted with two modified versions of Granger non-causality tests. The test results indicate that bidirectional causal relationships are predominant. Study 3 tests for panel data Granger non-causality between HIV/AIDS mortality and GDPc in 44 African countries in years 1970 – 2012. The results highlight the predictable relationship from mortality to GDPc. Study 4 investigates Granger causality – both in panel data and at country level – between child health and economic growth in a sample of 175 countries in years 1990 – 2014. The results indicate that relationships run in both directions. Interestingly, the impact of economic growth on child health growth is more frequent in lower income countries relative to high income countries. Study 5 analyses relationship between life expectancy at older ages and GDPc in OECD countries in years 1970 - 2012. Results demonstrate that these variables are co-integrated, and bilateral causal relationship is present in 65% of total countries. Overall, obtained results alert economists about the risk of endogeneity bias and specification errors in empirical analyses which aim to define the relationships between health variables and GDPc.

Keywords: health care expenditures, GDPc HIV/AIDS, child health, life expectancy at older ages, Granger non-causality tests

(8)

6

Amiri, Arshia

Kaksisuuntaiset vaikutukset terveysmenojen, terveyden ja taloudellisen kasvun välillä. Tulok-sia Grangerin ei-kausaalisuustesteillä aikasarja- ja paneeliaineistoilla.

Joensuu: Itä-Suomen yliopisto, 2017

Publications of the University of Eastern Finland Dissertation in Social Sciences and Business Studies ISBN: 978-952-61-2649-4 (print)

ISSNL: 1798-5749 ISSN: 1798-5749

ISBN: 978-952-61-2650-0 (PDF) ISSN: 1798-5757 (PDF)

TIIVISTELMÄ

Väitöskirjan keskeiset tutkimuskysymykset koskevat positiivisia ja mahdollisesti kaksisuuntaisia yhteyksiä terveysresursseja muodostavien investointien ja kansan- tulon välillä. Teoreettisesti nämä yhteydet voivat olla molempiin suuntiin. Empii- riset tulokset eivät ole kuitenkin vakuuttavia. Tämän seurauksena työssä käytetään Grangerin ei-kausaatiotestimetodia tarkemman informaation aikaansaamiseksi terveysmuuttujien ja taloudellisen aktiviteetin välisestä ennustettavuudesta. Tutki- mukset 1 ja 2 keskittyvät aikasarja- ja paneeliaineistokohtaiseen Granger –kausaati- oon per capita terveysmenojen ja bruttokansatuotteen (BKTc) välillä OECD maissa vuosina 1970 – 2009. Analyysi suoritettiin kahden edelleenkehitellyn Granger ei- kausaatiotestin avulla, jotka osoittavat, että kaksisuuntainen ennustesuhde on val- litseva. Tutkimus 3 testaa paneeliaineisto Granger –kausaatiota HIV/AIDS kuollei- suustapausmäärien ja BKTc:n väliltä 44 Afrikan maassa vuosina 1970 – 2012 koros- taen tulosta, että yhteys kulkee kuolleisuudesta ja BKTc:hen. Tutkimus 4 tarkastelee sekä aikasarja että paneeliaineisto Granger –kausaatiota lasten terveyden ja taloudellisen kasvun väliltä 175 maan kohdalta vuosina 1990 – 2014. Tulokset osoittavat, että yhteys on molemminsuuntainen. On mielenkiintoista, että taloudellisen kasvun terveysvaikutukset ovat useimmin havaittavissa matalan kuin korkean tulotason maissa. Tutkimus 5 analysoi ikääntyneiden henkilöiden elinajan odotteen ja BKTc:

een välistä suhdetta OECD maissa 1970 – 2012 osoittaen, että nämä muuttujat ovat yhteisintegroituneita ja kaksisuuntainen kausaatiosuhde vallitsee 65 %:lle maista.

Kokonaisuutena väitöstyön tulokset varoittavat ekonomisteja endogeenisuushar- han ja täsmennysvirheiden riskeistä empiirisissa analyyseissa, jotka pyrkivät mää- rittämään terveysmuuttujien ja BKTc:n välistä relaatiota.

Avainsanat: terveysmenot, BKTc, HIV/AIDS, lasten terveys, elinajan odote, Grangerin ei- kausaalisuustestit

(9)

7

ACKNOWLEDGEMENTS

I would like to express my gratitude to all who helped and encouraged me to take my doctoral degree. Firstly, I want to thank Prof. Bruno Ventelou from Aix- Marseille School of Economics (AMSE), France for his support and advices during these years. For me he is like a father that took my hand during some huge obsta- cles in my life. I extremely appreciate Bruno and I tender my grateful thanks to him.

I am extremely thankful to Prof. Ulf-G. Gerthdam from The Swedish School of Health Economics and Lund University for granting and collaborating with me from 2011 to 2013 in a World Health Organization (WHO) project. He is one of the best health economists in the world and I really appreciate him for giving the chance of collaborating with him.

I want to thank gratefully my supervisors Prof. Mikael Linden, Dr. Matti Estola, and Dr. Eila kankaanpää for their great support and advices. By far, Mika is the most professional economist that I have ever worked with in my life. He is so hard working and a good instructor in econometrics model analysis. I learnt many prac- tical econometric skills of him and he is the best academic leader for me during these years. I really appreciate that he trusted me and collaborated with me. I want to give a special thanks to Eila for her time, attempts to financing my project, and solving my problems in university. I want to give a special thanks to Prof. Hannu Valtonen, the previous head of health economics in University of Eastern Finland, for giving me the chance of studying in Finland, helping and supporting me during my PhD program. I want to thank Prof. Johanna Lammintakanen, the head of de- partment, for financing my study and her kind supports.

I would like to give special thanks to my pre-examiners Prof. Markus Jäntti from University of Stockholm and Prof. Petri Böckerman from University of Turku for their constructive comments.

Finally, I would like to thank my lovely wife, Emma, my parents and my friends here and there. It is hard to say how amazing you are in words. Thanks for the chance of interesting life with you all.

Kuopio, December 2017 Arshia Amiri

(10)

8

(11)

9

LIST OF TABLES

Table 1. Basic health economic studies with a focus on GDP-lead-health theory using cross-sectional data analysis ... 22 Table 2. Basic health economic studies with a focus on GDP-lead-health theory using time series and panel data analyses ... 24 Table 3. Basic health economic studies with a focus on health-lead-GDP theory

... 30 Table 4. Basic health economic studies with a focus on feedback theory ... 32 Table 5. Basic health economic studies with a focus on health-lead-GDP theory using HIV/AIDs as proxy for health ... 34 Table 6. Basic health economic studies with a focus on GDP-lead-health theory

using infant mortality (IMR) and under-five mortality rate (MR5) as proxies of child health ... 37 Table 7. Basic health economic studies with a focus on health-lead-GDP theory with aging proxies ... 40 Table 8. Description of health and GDP measures and data sources in each

study ... 44 Table 9. Names of countries included and dimension of data used ... 45 Table 10. Study design and econometric models used for research questions . 46 Table 11. Summary results of Toda-Yamamoto version of Granger non-causality tests between HCEc and GDPc in 20 OECD countries 1970-2009 ... 55 Table 12. Summary results of Hsiao’s version of Granger non-causality tests

between de-trended values of HCEc and GDPc in 34 OECD countries 1970-2012 ... 58 Table 13. Summary results of Granger non-causality tests between HIV and

GDPc in 44 African countries 1990-2009 ... 59 Table 14. Summary results Granger non-causality tests between MR5 and GDPc growth in 175 countries 1990-2014 ... 62 Table 15. Summary results of Granger non-causality tests between LE65+ and

GDPc in 26 OECD countries 1970-2012 ... 65

LIST OF FIGURES

Figure 1. Income-health relationship ... 16 Figure 2. Bilateral links between health outcomes and GDPc growth ... 18 Figure 3. Cross plot of HCEc (lnHCEc) and GDPc (lnGDPc) in 20 OECD coun-

tries 1970-2009 ... 54 Figure 4. Strategy of Toda-Yamamoto test and summary of results ... 55

(14)

12

Figure 5. Cross plot of lnHCEc and lnGDPc, their growth rates (lnHCEc and lnGDPc) and de-trended values (lnHCEc_de-trended and lnGDPc_de- trended) in 34 OECD countries 1970-2012 ...57 Figure 6. Strategy of Pairwise, Pairwise Dumitrescu-Hurlin and Hsiao’s version

of Granger non-causality tests and summary of result ...58 Figure 7. Cross plot of HIV and GDPc in 44 African countries 1990-2009 ...59 Figure 8. Cross plot of child health growth (DlnMR5) and GDP growth (DlnG-

DPc) in 4 different income-level groups, 175 countries 1990-2014 ....61 Figure 9. Strategy of Pairwise, Pairwise Dumitrescu-Hurlin and panel fixed effect Granger non-causality tests and summary of results...61 Figure 10. Cross plot of life expectancy in older ages (lnLE65+) and GDP (lnG-

DPc) in 26 OECD countries 1970-2012 ...63 Figure 11. Strategy of panel analysis, Granger non-causality tests and summary

of results ...64

ABBREVIATIONS

CMH = Commission on Macroeconomics and Health FPE = Final Prediction Error

GDP = Gross Domestic Product

GDPc = Gross Domestic Product per capita HCE = Health Care Expenditure

HCEc = Health Care Expenditure per capita HIC = High-Income Countries

IMR = Infant mortality rate

LE65+ = Life Expectancy at 65 years’ old LIC = Low-Income Countries

LMIC = Lower-Middle-Income Countries Ln = logarithm

MDGs = Millennium Development Goals MR5 = Under-five Mortality Rate

MWALD = Modified Wald ODA = Official Development

OECD = Organisation for Economic Co-operation and Development OVB = Omitted-Variable Bias

(15)

13

1 INTRODUCTION

The idea of thesis, which focuses on potential bilateral impacts between health outcomes, health expenditures, and GDP per capita variations, is important in health policy and empirical macroeconomic analysis. Although it contributes to health economics, my approach is partly different from the present day health economics, which is based on microdata and strives for experimental evidence. I study the wider societal and economic consequences of population health. My idea is to study the interactions between health of the population, health expenditures, and the macroeconomic performances (GDP per capita) in different countries.

All the major research questions in dissertation are related to the plausible positive bilateral impacts between investment in human capital resources in the formation of health and national income level and growth. From the theoretical point of view, there are three main health economic theories to explain the relationship between health outcome indexes and economic development performances:

1) GDP-lead-health: More income leads to increases in health care spending and health level of population. This is the concept of “Wealthier is Healthier”.

2) Health-lead-GDP: Healthier people can educate themselves, work harder, longer and better, and the productivity of society increases.

3) Feedback theory: Health and GDP link together simultaneously to each other.

To our knowledge, most of empirical studies highlight either the existence of unidirectional (causal) relationship from income to heath care expenditures and health indexes (GDP → health) or vice versa (health → GDP), but they dramatically disregard the simultaneous impacts between these variables. By ignoring the possible bilateral link between health variables and GDP, the results of one-way regression either with health on GDP or vice-verse are misleading and biased. Thus if GDP and health variables are included as the major determinative variables in an empirical model and they significantly determine each other simultaneously in both directions, then the standard single equation model estimation procedures face the endogeneity problem and model specification errors. In other words, if health variables and GDP are robust variables in defining empirical models and they are significantly related to each other, the significance of the bilateral relationship must be considered and tested. Therefore, as GDP and health are the key elements of empirical analysis in health economics, there is a need of empirical analysis to test whether GDP and health expenditures and health outcomes are determined simultaneously or not.

(16)

14

However, the potential (causal) link between income and health may not occur instantaneously but with some lags. Therefore, the Granger non-causality test proposed by Granger (1969) is an efficient method to estimate and test the directions of temporal causality between health outcomes, health expenditures and GDP.

To our knowledge, there are a very few studies that mentions the pairwise relationships to be bi-directional, and more dramatically, the results of previous empirical literature in this context are inconclusive and contra-dictionary. In response to this the dissertation revisits the basic question of whether health as index of human capital accumulation stimulates economy, and whether GDP per capita level and growth sustains health outcomes and health care expenditures.

However our analyses take the simultaneous feedback relationship between health outcomes, health care spending and incomes as the maintained hypothesis. To give as wide as possible support to testing this hypothesis, countries from different country groups and different health variables from different databases are included in the analysis. Studies 1, 2 and 5, use OECD country observations in which the health systems and health outcomes are quite uniform like the high ageing populations in the recent years. Studies 3 and 4 analyse African countries, and a sample of the world countries in different income levels, respectively, where the health economic data are available.

To test the potential bilateral relationship between health indexes and macroeconomic performances, we investigate different Granger non-causality tests between most popular health outcome variables, e.g. health care expenditure (HCE), HIV/AIDS prevalence, under-five years’ child mortality (MR5), and life expectancy at older ages (LE65+), and GDP per capita for the largest possible set of countries.

The content of the dissertation is following. First, Chapter 2 presents a background and definitions of economic theories on relationship between health outcomes and national income. This chapter also reviews the history of the topic, relevant literature, and debates. Chapter presents first the conceptual framework to describe the relationship between the health outcome variables and economic activity, and second we describe the endogeneity problem and specification error caused by ignoring the matter of simultaneous relationship between health variables and GDP. Based on Chapter 2, Chapter 3 provides basic hypothesis of the dissertation, presents the main research questions, and aims of the study. Definition of the data, study design, and methodology of Granger causality and its different versions are presented in Chapter 4 in details. Granger’s non-causality test idea and its modified versions will be set out here. The obtained results will be introduced in the following Chapter 5. Chapter 6 will summarize the results and discuss the main questions of the dissertation. Finally, the conclusion of dissertation will be presented in Chapter 7.

(17)

15

2 THEORETICAL FRAMEWORK, BACK- GROUND AND LITERATURE REVIEW

The chapter first reviews the theoretical framework of explaining the relationship between macroeconomic health performance variables and economic growth included: 1) GDP-Lead-Health theory, 2) Health-Lead-GDP theory, and 3) the feedback theory. Secondly, the results of endogeneity bias and specification error are explained next. Finally, the background of health care expenditures, used in studies 1 and 2, and three main health outcome indexes used in studies 3, 4 and 5 (i.e. child health, HIV/AIDS and life expectancy at older ages), are presented separately.

2.1 THEORY OF HEALTH PERFORMANCE AND ECONOMIC GROWTH

The effects of health outcomes and expenditures on economic growth have given consideration by the health economists as health expenditures (HCE) trend up and subsequently improvements in health level of societies take place. As the need for improvement in health services in developed countries is still high, the health spending growth path is considerably higher than national income growth. This is a challenge for industrialized countries. On the other hand, under-developed and developing countries face the challenge of maintenance and increasing health services as an essential economic stimulus for economic development (Murray et al.

1994). Therefore, it is important to have a theoretical framework for analysing the plausible channels where health expenditures, health outcomes and macroeconomic performance – i.e. GDP per capita (GDPc) and economic growth – are linked together.

In health economics modelling, the definition of presence and the direction of relationship between health human capital and macroeconomic performance like national income growth have become as one of the basic questions since the pioneering surveys proposed by Kleiman (1974) and Newhouse (1977). Kleiman and Newhouse showed that HCE as the main index of health performances with direct consequences on health level and health outcomes are strongly correlated with variation of GDPc in OECD countries. Since then, a huge attempt has made to analyse this relationship using different methods and databases. Although in some cases the results are mixed, the overall conclusions of numerous empirical studies propose the existence of three major health economic theoretical frameworks to explain the relationship between health performances and GDPc.

(18)

16

2.1.1 GDP-Lead-Health Theory

The first theory, we call it as GDP-lead-health theory, argues that health expenditure with direct consequences on health measures and outcomes is a function of national income – i.e. HCE = f(GDPc,…). Without exception, aggregate health care spending with significant direct consequences on health performances is a function of GDPc and economic growth (Hansen and King, 1996). This theory highlights the effects of GDPc level and growth on major health factors by selecting GDPc as a key or almost the only robust explanatory variable for health outcomes and health finance variations (see Hartwig, 2008).

More precisely, this idea is explained in some basic theories like “income-health gradient” –hypothesis which assumes that better health as a human need or an essential good - with positive effect on utility and welfare - is associated with higher income both in macro and micro levels. In macro level, with more income and GDPc growth both government and private sectors allocate more expenditures on health care services and health related technologies to satisfy human needs of being and living healthy - as mentioned in the first step of Maslow's hierarchy of human needs. This subsequently improves the health level and health indicators like mortality, morbidity, general health and health outcomes of a society. Fig. 1 demonstrates the basic shape of plausible relationship between health as a good for individuals as a function of income. More income leads to a wide health improvement at low levels of income and slight improvements take place at high income levels. More income can increase health level with two different mechanisms: 1) more expenditures on health services which cause to movements along the income-health curve, and 2) more expenditures on discovering and using better health care technologies which shifts the income-health curve up.

Figure 1: Income-health relationship

(19)

17 In this sense the “Wealthier is Healthier” theory argues that economic growth prepares more acquiring resources which helps societies’ health care system to improve and becomes more efficient. In another word, income is a vital and an inventible source of health enhancement in both public and private health sectors.

Wealthier people or societies pay more money to fight against the disease which they suffer from and this causes to improve the productive health level and life expectancy.

2.1.2 Health-Lead-GDP Theory

The theory of economic growth proposed by Solow (1956, 1957), as one of the basic frameworks in economics, highlighted the effect of health investment on long run growth output and macroeconomic performances. Solow posits a production function in which that GDP is a function of two major input factors: physical capital stock and population. Therefore, enhancement in human and physical resources may lead to raising the efficiency of economic system and improvement in GDP growth in long run. Solow anticipated long run stable steady-state equilibrium of GDP growth and concluded that improvements also in human capital resources - included higher level of education, and better health (see Saha and Gerdtham, 2012) – are a key of raising economic growth when the physical capital stocks are almost fixed or at least hard to increase. In other words, the most important stimulus of economic growth is the improvement of human capital resources, and health is the key factor of upgrading human capital assessment. Hence, health has been known for long to be as the most drastic element of the human capital stock especially in industrialized countries (see Schultz, 1961 and Mushkin, 1962).

As Bloom and Canning (2000) state, health indicators impact GDPc and economic growth in several direct and indirect pathways. Firstly, healthier people are more productive, can work harder and longer, and think better. Therefore, improvements in health level increase the efficiency and productivity of workforce.

This decreases the plausible health costs of labour force. Secondly, health has a considerable effect on individuals’ educational performance – healthy people can educate themselves better and learn more the essential skills useful for working time giving an indirect effect on increasing productivity of human capital resources.

Thirdly, there is a positive effect of improvement in health level on saving and investment with clear positive effects on aggregate national income. Finally, population size and age structure are direct consequences of investment on health and life expectancy. Similarly, better and more efficient health care services prepare better conditions for workforce to work effective and to be more productive, and consequently this increases income, welfare and safety (see Weil 2005, 2007).

(20)

18

Overall, form theoretical point of view there exist indirect and direct channels from enhancement in health to national income.

2.1.3 The Feedback Theory

The feedback theory between health performances, GDPc level and growth highlights that the relationship between HCE or health outcomes and GDPc can run in both directions simultaneously. As GDP-lead-health approach is clearly approved theoretically – i.e. in line with “wealthier is healthier” theory – in the feedback theory the simultaneous reverse effects – national income as a function of health performances or HCE is given more consideration. On the other hand, rising national income allocates more resources to the health sector and this leads to improvements in health care services and health level indexes in the society. Thus the improvements in health level increase the income formation capabilities and spending of larger population with extending life expectancy, increasing ultimately the national income growth. Hence, simultaneous effects are evident between health and economic performance. Fig. 2 provides links between GDP growth, HCE, health outcomes and indicators indicated by the feedback theory.

Figure 2: Bilateral links between health outcomes and GDP growth

(21)

19 2.1.4 Endogeneity and Specification Error

Theoretically, if income and health indexes are the major determinants to each other simultaneously – i.e. there is a significant two-way relationship between them – then the standard one-way equation model, which assumes that GDPc or health variable is a definition variable to the other, will face to endogeneity problem, i.e.

the OLS parameters of estimated equation is inconsistent.

Hence, if endogeneity of major defining variable in the model is overlooked and OLS-estimation is conducted only with unidirectional single equation model the estimated parameters are biased and inconsistent. In addition, if important temporal factors of bilateral effects are left out, this leads also to omitted-variable bias as a source of specification error (see Barreto and Howland, 2006). For more de- tailed explanation, consider following simple linear model:

= + (−1) + (−1) +

(1)

where yⁱ is dependent variable that correlates with xⁱ. The two other variables x(-1)ⁱ and y(-1)ⁱ are the factors of bilateral temporal relationship between yⁱ and xⁱ and uⁱ is the WN error term. If the bilateral effects x(-1)ⁱ and y(-1)ⁱ are omitted form the regression, then OLS method estimates the following value for the response parame- ter of xⁱ:

= ( ^′ )^{−1 ′}

(2) If we substitute for Y from the correct model, we have

= ( ^′ )^{−1 ′}( + (−1) + (−1) + )

= ( ^′ )^{−1 ′} + ( ^′ )^{−1 ′} (−1) + ( ^′ )^{−1 ′} (−1) + ( ^′ )^{−1 ′}

= + ( ^′ )^{−1 ′} (−1) + ( ^′ )^{−1 ′} (−1) + ( ^′ )^{−1 ′} (3)

If we evoke the assumption of exogeneity of X sustaining the above single equation model, the expectation of final term (X' X)^-1 X' U is zero, and the remaining terms gives

[ | ] = + ( ^′ )^{−1 ′} (−1) + ( ^′ )^{−1 ′} (−1) [ | ] = +

(4)

(22)

20

In this case, the second term after the equal sign is the bias caused by neglecting the temporal terms from the model. This is the omitted-variable bias. The bias here is non-zero for the reason of significant correlations that yi has with lagged variables of xⁱ and yⁱ. The bias is enlarged when we rule out the assumption of exogeneity.

Thus, if E XU[ ] 0≠ allowed, we add term E X X[( ' )⁻¹XU] 0≠ to the bias. In sum, both the omitted variables and endogeneity lead to biased inference of effects between variables Y and X. The Granger non-causality testing for bidirectional effects between Y and X is a solution to these problems. The test procedure is defined in de- tail in Section 4.2. Note that Granger testing allows for endogeneity and tests for possible Granger uni- or non-directionality exogeneity.

2.2 BACKGROUND AND LITERATURE REVIEW

2.2.1 Health Care Expenditures (HCE), Health Outcomes and Economic Growth

Since the pioneering studies proposed by Kleiman (1974) and Newhouse (1977)¹, the GDP-lead-health theory has been the basic framework of the most theoretical and empirical studies since the 1970s in health economics. Kleiman (1974) and Newhouse (1977) used cross-sectional data and showed that HCE and GDPc are significantly correlated and almost 90% of variation of HCE can be explained by the variation of GDPc. By estimating the coefficient of GDP→HCE, measuring the size of income elasticity of health care expenditures, and analysing the fiscal policy to allocate and distribute financial health resources, were the main topics for research.

During past decades, several empirical studies based on one-way regression analysis using international databases have tried to investigate the exact coefficient of GDP-related-health relationship. The most common findings of these studies are that: 1) total national income is the most important factor of explaining the variation of expenditures on health care services, and 2) the calculated coefficient of this relationship – the size of income elasticity of health care – is around or slightly more than one for industrialized countries. Therefore health is argued to be a luxury good in OECD countries based on cross-sectional analysis.

Previous literature used several different modelling approaches related to the type of data and methods. Although a few studies used household data, the rest used macroeconomic aggregate observations. A significant positive effect of GDPc on health expenditures in cross section analysis has been consistent in following

1 Although there are some papers like Anderson (1972), Farser (1973), and White (1975), which suggested the GDP-lead-health theory, literature cites Kleiman (1974) and Newhouse (1977) as the pioneer of this approach.

(23)

21 papers using data on developed countries: Farser (1973), Newhouse (1977, 1987), Cullis and West (1979), Maxwell (1981), OECD (1985), Leu (1986), Glaser (1987), Parkin et al. (1987), Gerdtham et al. (1988), Jonsson (1989), Poullier (1989), Vogel (1989), Gerdtham and Jonsson (1991a, b), Milne and Molana (1991), Gerdtham et al.

(1992a, b), Schieber and Maeda (1999), Musgrove et al. (2002), Jonsson and Eckerlund (2003), van der Gaag and Stimac (2008), and Murthy and Okunade (2009). Table 1 gives the sum-up of these studies.

More recently, some cross-sectional studies showed significant GDP-lead-health in global. Schieber and Maeda (1999) calculated the global health income elasticity in public and private sectors to be 1.13 in the average and verified a higher elasticity for public expenditures than for private sector in 1994. Musgrove et al. (2002) used the observation of 191 countries in 1997 and showed that depended on the countries and data, the income elasticity for HCE was between 1.133 and 1.275.

Jonsson and Eckerlund (2003) calculated the income elasticity of public and government health care spending per capita instead of aggregate real HCE for 29 OECD countries using the observation from year 1998. Their results argued that public HCE is a luxury good in OECD countries and the income elasticity of government HCE was 1.21. Moreover, Gaag and Stimac (2008) analysed the observation of 175 countries in 2004, and HCE income elasticity was 1.09, 0.83 and 1.19 in global, Middle East and OECD countries, respectively. Murthy and Okunade (2009) calculated health income elasticity between 1.08 and 1.12 for 44 African countries in 2001. Although, the strongest factor in all previous studies was GDPc, some literature analysed multivariate models with adding other non-income macroeconomic determinants for health care spending, e.g. population age proportions, relative price of HCE, and inpatient/outpatient mix, to find more correct income elasticity for HCE.

Overall, there exist three main themes in the previous literature based on cross- sectional data analysis where the basic framework is the GDP-lead-health theory:

I. Most of studies used aggregate HCE per capita or its share of GDP as the index of health services and outcomes.

II. Most of results are based on cross-section data. OECD and industrialized country observations were main database of most empirical analyses, while under-developed and developing countries – because of the lack of data availability – were ignored from the analysis².

III. Results of previous literatures argued that most variations of HCE or financial health variables can be explained by the variations of GDPc, and these two variables are strongly correlated.

2 There exist few studies with developed and underdeveloped countries like Musgrove (1983), Schieber and Maeda (1999), Gaag and Stimac (2008), and Murthy & Okunade (2009).

(24)

22

IV. In line with the conclusion of Kleiman (1974) and Newhouse (1977), almost all researches based on cross-section database found an income elasticity more than one³ for expenditures on health care services – except Milne and Molana (1991), and Gaag and Stimac (2008) – indicating that health is a luxury good in OECD and developed countries.

Table 1: Basic health economic studies with a focus on GDP-lead-health theory using cross- sectional data analysis

Study GDP measure Health measure Countries and time

Significant GDP-health coefficient Kleiman (1974) National prod-

uct per capita Total national

HCE 16 countries,

1968 1.2

Newhouse

(1977) GDP spent on

medical care Medical expendi-

ture per capita 13 developed

countries, 1972 1.15-1.31 Maxwell (1981) GDP per capita HCE per capita 10 countries,

1975 1.4

Leu (1986) Real GDP per

capita HCE per capita 19 OECD, 1974 1.18-1.36 Parkin et al.

(1987) Aggregate GDP Aggregate HCE 23 OECD, 1980 1.12-1.18 Jonsson (1989) Real GDP per

capita HCE per capita 19 OECD, 1987 1.36 Gertler & van

der Gaag (1990) GDP per capita HCE per capita 25 countries,

1975 1.3

Milne & Molana

(1991) Real GDP per

capita Real HCE per capita

11 European countries, 1980,

1985 0.73

Gerdtham et al.

(1992a, b) Aggregate GDP Aggregate HCE 19 OECD, 1974, 1980 and

1987 1.33

Schieber &

Maeda (1999) GDP per capita Public and private HCE per

capita

6 developing

regions, 1994 1.13 Musgrove et al.

(2002) GDP per capita HCE per capita 191 countries,

1997 1.13-1.27

Jonsson &

Eckerlund

(2003) GDP per capita Public HCE per

capita 29 OECD, 1998 1.21

Gaag & Stimac

(2008) GDP per capita HCE per capita 175 countries,

2004 0.83-1.19

Murthy & Okun-

ade (2009) GDP per capita HCE per capita 44 African coun-

tries, 2001 1.08-1.12 Lv & Xu (2016) GDP per capita HCE per capita 172 countries,

2009-2013 0.93-1.26

3 Studies like Kleiman (1974), Newhouse (1977), Leu (1986) and Getzen (1990) found significant income elasticity between about 1.20 and 1.50 for HCE in OECD countries.

(25)

23 Cross-sectional one-way regression method has been criticized by e.g. Parkin et al.

(1987), Culyer (1990), Levine and Reneit (1991), for the shortage of data set used in the analysis, homogeneity of effects on HCE which is assumed across the various countries with different health care systems, and the difficulty of interpreting exogenous variables coefficients, i.e. whether health is luxury good or not. From the beginning, Culyer (1990), Getzen (1990), Schieber (1990), Getzen and Poullier (1992), and Hitiris and Posnett (1992) have tried to highlight these econometric problems in previous cross-sectional analyses e.g. by using time series and panel data model estimators.

Time series analyses has become possible since OECD (1990) provided annual data of HCE in OECD countries. Some studies including Getzen (1990), Schieber (1990), Getzen and Poullier (1992), Hansen and King (1996), Blomqvist and Carter (1997), Gerdtham and Löthgren (2000, 2002), Okunade and Karakus (2001), Dreger and Reimers (2005), Herwarts and Theilen (2003), Hui-Kuang et al. (2011), Woodward and Wang (2012), Acemoglu et al. (2013), Yavuz et al. (2013), and Murthy and Okunade (2016) alerted health economists about possible non- stationarity in HCE and GDPc data, and presented the unit-root and co-integration tests results between HCE and GDPc in OECD countries, including a few different controlling variables⁴. By contrast to the cross-sectional and other time series literatures, Blomqvist and Carter (1997), Herwarts and Theilen (2003), Acemoglu et al. (2013), and Yavuz et al. (2013) concluded that HCE income elasticity is not greater than one. For the sum-up of these results, see Table 2.

In line with previous studies, several studies have used panel data and highlighted the existence of significant relationship from GDPc to health outcomes and HCE. However, the estimated income elasticity for health care spending in GDP-lead-health is sensitive to the additional assumptions of the panel model used (Xu et al. 2011). Panel data analyses and modelling have been considering e.g. by Culyer (1989), Gertler and van der Gaag (1990), Getzen (1990), Pfaff (1990), Gerdtham et al. (1992a, b), Hitiris and Posnett (1992), Vogel (1992), Murthy and Ukpolo (1994), Viscusi (1994), Hitiris (1997), Barros (1998), Gerdtham et al. (1998), McCoskey and Selden (1998), Roberts (1998), Freeman (2003), Herwartz et al. (2003), Jewell et al. (2003), Matteo (2003), Carrion-i-Silvestre (2005), Dreger and Reimers (2005), Sen (2005), Hartwig (2008), Farag et al. (2009), Baltagi and Moscone (2010), Chakroun (2010), Lu et al. (2010), Mehrara et al. (2010), Liu et al. (2011), Xu et al.

(2011), Farag et al. (2012), Lago-penas et al. (2013), Blazquez-Fernandez et al. (2014), and de Mello-Sampayo and de Sousa-Vale (2014). The findings of panel data approach to calculate income elasticity of health care spending are inconclusive.

Some researches like Barros (1998), Gerdtham et al. (1998), Freeman (2003), Dreger

& Reimers (2005), Hartwig (2008), Farag et al. (2009), Baltagi & Moscone (2010), Xu

4 However, because of lack of data availability during the time, the results of these studies in some cases compel uncertainty and efficiency problems (Hartwig 2008).

(26)

24

et al. (2011), and Farag et al. (2012) argued that income elasticity for HCE is less than one in OECD countries, while others maintained that health is a luxury good in industrialized countries ( see Table 2).

Table 2: Basic health economic studies with a focus on GDP-lead-health theory using time series and panel data analyses

Study GDP measure Health measure Countries, time

and method Significant GDP- health coefficient Getzen

(1990) GDP per capita National HCE U.S., 1966-87

Time series 1.6

Schieber

(1990) Real GDP Share of GDP devoted to HCE

7 countries, 1960-87 Time series

methods

1.2 Getzen &

Poullier

(1992) GDP growth National HCE

19 countries, 1965-86 Time series

methods

1.4 Hitiris &

Posnett

(1992) GDP per capita HCE per capita

20 OECD, 1960- 87 Panel data

analysis

1.0-1.2

Viscusi

analysis

1.1

Hansen &

King (1996) GDP per capita HCE per capita

20 OECD, 1960- 87 Time series

methods

No significant

Blomqvist &

Carter (1997) GDP HCE

22 OECD, 1970- 91 Time series

methods

0.56-0.69

Hitiris (1997) GDP per capita HCE per capita

analysis

1.14-1.17

Barros

(1998) GDP growth

per capita HCE growth per capita

analysis

0.62-0.92

Gerdtham et

al. (1998) GDP per capita HCE per capita

analysis

0.74 McKoskey &

Selden

(1998) GDP per capita HCE per capita 20 OECD, 1960- 87

Panel unit root No significant Roberts

analysis

1.00

Freeman (2003)

Disposable personal in-

come HCE per capita U.S. states, 1966-98

Panel data 0.82-0.84

(27)

25 analysis

Herwarts &

Theilen (2003)

Real GDP per

capita HCE per capita

19 OECD, 1960- Time series ECM 97

model

Short-run 0.43 Long-run 0.13

Dreger &

Reimers (2005)

Real GDP per

21 OECD, 1975- 2001 Panel data (cointegration

techniques)

0.68-0.84

Sen (2005) Real GDP per

analysis

0.21-0.51

Hartwig

(2008) Real GDP per

analysis

0.33-0.37

Farag et al.

(2009) Real GDP per

capita Domestic

government HCE

144 countries, 1995-2006 Panel data analysis

0.66-0.96 Baltagi &

Moscone

analysis

0.87

Chakroun

(2010) Real GDP per

17 OECD, 1975- 2003 Multivariate regression model

0.70-0.90

Mehrara et

al. (2010) GDP per capita HCE per capita

analysis

2.59

Hui-Kuang et

al. (2011) Real GDP per

25 developed countries, 1998

Time series methods

1.20-1.47

Liu et al.

22 OECD, 1960- 2002 Semiparametric

panel varying coefficient model

1.60

Xu et al.

143 countries, 1995-2008 Panel data (fixed-effect method)

0.75-0.95

Farag et al.

173 countries, 1995-2006 Panel data analysis

0.82-0.90

Woodward &

Wang (2012) Real GDP per

capita National HCE per capita

U.S., 1960-2008 Time series

methods 1.39

Acemoglu et

al. (2013) Real GDP HCE as share of GDP

U.S., 1950-2005 Time series

methods 0.70-1.10

Lago-penas Real GDP per Total HCE per 31 OECD, 1970- Short-run 0.3

(28)

26

et al. (2013) capita capita 2009

Panel data analysis

Long-run 1.1

Yavuz et al.

(2013) Real GDP per

Turkey, 1960- 2012 Time series

methods

Short-run 0.75

Blazquez- Fernandez et

al. (2014) GDP per capita HCE per capita &

Public HCE as share of HCE

analysis

Short-run 0.18-0.20 Long-run 1.07-1.12 de Mello-

Sampayo &

de Sousa- Vale (2014)

GDP per capita Government HCE per capita

analysis

No significant

Meladenovic

et al. (2016) Real GDP per

28 European countries, 1974-

2015 Time series data ANFIS approach

HCE and GDP series are highly linked Murthy &

Okunade (2016)

Real GDP per

U.S., 1960-2012 Time series cointegration

(ARDL)

0.92

Several non-income factors added to analysis as the determinants and covariates of health care spending and other financial health indexes are typically as follows.

I. Epidemiological proxy

When there exist the possibility of effect of an especial epidemical disease like diabetes, HIV and malaria across the observation areas, epidemiological need plays a plausible role as determinant of expenditures on health – see Deaton (2002)⁵. As GDPc plays a key role on variations of HCE, there is no evidence that epidemiological need plays a significant role on health care spending. Murthy and Okunade (2009) find no significant coefficient for maternal new born and child mortality as a HCE determinant in Africa. Furthermore, Lu et al. (2010) added HIV/AIDs as an exogenous variable and proxy of epidemiological need in aggregate government HCE as share of GDP and their result verified no evidence of significant effect of epidemiological need on health care spending in developing countries.

II. Financial characteristics of health care system

Institutional factors have been pointed as determinant proxies for financial health variables, i.e. the share of government financing on health care sector, to identify which amount of HCE has been financed by government or by private sector, e.g.

pharmaceutical spending (Clemente et al., 2008; van Elk et al., 2009). A very few

5 Deaton (2002) refers also to some health-related behaviour determinants like use of tobacco, alcohol, and drugs, obesity and sex.

(29)

27 studies – e.g. Leu (1986), Culyer (1988), Hitiris and Posnett (1992), and van der Gaag and Stimac (2008) – have found a link from financed health care spending by government to increasing HCE level. Okunade et al. (2004) and Moscone and Tosetti (2010) found negative coefficient for public share of HCE. Hence, government in line with political factors plays an important role on the formation of HCE (Potrafke, 2010). Similarly, Cameron et al. (1998) and Scherer and Devaux (2010) pointed many forms of private HCE as a strong positive determinant factor for HCE and life expectancy. Guindon and Contoyannis (2008) and Cremieux et al.

(2005a, b) find a significant positive effect of pharmaceutical expenditures on HCE.

Moreover, Wagstaff and Bank (2009) and Wagstaff and Moreno-Serra (2009a, b) compared tax-based system with insurance-based system of OECD, East European and central Asian countries. Their result suggested that amount of per capita health care spending is higher in social health insurance based systems in OECD countries. Also, the government health care spending per capita in East European and central Asian countries based on health care insurance system is higher than tax-based health system. A few studies used non-OECD panel data like Farag et al.

(2009) and Leu et al. (2010) and investigated the possible effect of Official Development (ODA) on HCE. Surprisingly, Leu et al. (2010) found that GDPc has no significant effect on government HCE as a share of GDP and private sector is positively correlated with HCE caused by ODA, while government sector has a negative effect on it.

III. Population age structure

Age proportion of population structure has been used in HCE regressions as one plausible determinant for variation of financial health variables among countries.

Traditionally, the share of young and old proportion (e.g. proportion of below 15 years, above 65 or 75 years) were added as an exogenous variable in models. Most of previous literature paid more attention to the effect of aging indicators and found significant but weak coefficient for the proportion of older population (Leu, 1986; Culyer, 1989 and 1990; Gerdtham et al., 1992; Hitiris and Posnett, 1992; Di Matteo and Di Matteo, 1998; Zweifel et al., 1999; Gerdtham and Jonsson, 2000;

Jonsson and Eckerlund, 2003). In line with previous research, more recent surveys verified a positive significant effect of aging on HCE and predicted that this effect will be increased during the time in OECD countries (Okunade et al., 2004; Dreger and Reimers, 2005; Christiansen et al., 2006; Oliveira Martins and de la Maisonneuve, 2006).

IV. Medical care technological progress

Since the pioneering work proposed by Newhouse (1992) technological progress in health care system has been known as an important determinant of financial health care indicators. As there is a limitation in data availability of health indicators, the challenge for health economists is to find a good proxy for changes in technological

(30)

28

progress of medical care. Therefore, previous studies have tried various proxies for plausible changes in technological progress such as:

I. the number of specific medical equipment (Weil, 1995) II. the surgical procedures (Baker and Wheeler, 2000)

III. time trends (Roberts, 1999; Gerdtham and Lothgren, 2000), or time- specific intercepts (Di Matteo, 2004)

IV. R&D services special for health sector (Okunade and Murthy, 2002) V. Infant mortality and life expectancy at adults (Dregen and Reimers,

2005)

VI. the number of tomographic scanners (Christiansen et al., 2006)

Gerdtham et al. (1998) used the number of renal dialyses per million of the population as index for medical care technology and found a positive effect of this proxy on HCE. Dormont et al. (2006) added a combination of above medical care technological progress and highlighted the important of technological proxies in determining macroeconomic health outcomes variables in OECD countries. Various trend variables added to models consider the effects of technological progress in the path of HCE and GDPc in different studies, e.g. O'Connell (1996), Gerdtham and Lothgren, (2000), Ariste and Carr (2003), and Freeman (2003). Typically they find time trends to have a positive and significant effect on HCE. However, the previous literatures based on non-OECD observations have not considered technological progress in their analysis because of the lack of data availability.

V. Relative price of health care services

Traditionally, health care service real prices would have a plausible effect of demand for health care, i.e. microeconomic theory explains the positive role of relative health care prices on increasing the expenditures on health (the Baumol theory of cost disease, see Hartwig 2008). Increases in health care service prices lead to a rise in relative prices compared to other sectors as productivity in health sector is generally lower than other sectors (Baltagi and Moscone, 2010). Empirically, the result of previous surveys which used real price on health care services are quite mixed. Although Gerdtham et al. (1992a) and Murthy and Ukpolo (1994) find no evidence for relative price of health, later studies, e.g. Okunade et al. (2004), Pomp and Vujic (2008), Hartwig (2008), and van Elk et al. (2009) found a significant role of relative price of health care services on health expenditures. Hence, based on Baumol’s theory – logically plausible in OECD countries, but not in developing country group – relative prices of health care services are an important determinant of growth of HCE but not necessarily for the HCE levels.

To summarize, the GDP-lead-health theory is considerably highlighted in previous studies. Income is mentioned as the most important determinant of health

(31)

29 care expenditure variations, while there is still no strong agreement on effects of other factors on the remaining variation.

The second research framework, based on health-lead-GDP theory, links health factors to GDPc level and growth. This idea was proposed by Schultz (1961) and Mushkin (1962), before the GDP-lead-health theory, but subsequently ignored in studies in 1970’s and 1980’s. In 1990’s and 2000’s, the lth-lead-GDP theory has become the major theoretical framework in the health economic literature. Some empirical studies show a significant positive effect of various health indexes on macroeconomic performances and support the theoretical framework. Literature that has empirically confirmed the health→GDPc effect and verified the positive role of health outcomes on GDP include but are not limited to Rivera and Currais (1999a, b, 2003 and 2004), Bloom and Canning (2000), Kalemli-Ozcan et al. (2000), Bloom et al. (2001, 2004), Heshmati (2001), McDonald and Roberts (2002), Webber (2002), Bloom et al. (2004), Jamison et al. (2005), Weil (2005, 2007), Temitope and Bola (2013), Kurt (2015), and Braendle and Colombier (2016). Table 3⁶ sums the results.

Most of these studies used life expectancy and health expenditures as the health proxies and they find positive relationship from health performances to GDPc level and growth in different countries and time periods. Rivera and Currais (1999a, b, 2003 and 2004) and Heshmati (2001) used health expenditures growth and its share on GDP growth and GDP per employed person. They calculated the significant positive coefficient of health level on GDP per employed person - between 0.13 and 0.33 - in OECD countries and Spanish regions. Furthermore, Weil (2005, 2007) verified that the positive effect of health on GDPc is considerably stronger in low- income countries group (LIC) than in rich countries.

However, there are studies that reject the significant positive relationship from health outcomes to income in high-income countries, e.g. Hartwig (2010).

Interestingly, Knowles and Owen (1995, 1997) and McDonald and Roberts (2002) rejected the plausible effect of life expectancy to economic growth of industrialized countries, and Bhargava et al. (2001) and Acemoglu & Johnson (2007) calculated a negative size of effect from adult survival rate to economic growth in a few rich countries. Moreover, Odubunmi et al. (2012) found a negative effect of HCE on national income in Nigeria.

In sum, previous literature confirms the existence of health-related-income relationship, while the significant coefficient of this relationship for different financial health variables and health proxies varied depend on database, time dimension and econometric methods used in their analyses.

6 Hartwig (2010) provided a part of this table which its updated version is presented here.

Bilateral effects between health expenditures, health outcomes and economic growth : evidence from time series and panel granger non-causality tests

uef.fi

ARSHIA AMIRI

Dissertations in Social Sciences and Business Studies

ARSHIA AMIRI

BILATERAL EFFECTS BETWEEN HEALTH EXPENDITURES, HEALTH OUTCOMES AND ECONOMIC GROWTH:

EVIDENCE FROM TIME SERIES AND PANEL

GRANGER NON-CAUSALITY TESTS

BILATERAL EFFECTS BETWEEN HEALTH EXPENDITURES, HEALTH OUTCOMES AND

ECONOMIC GROWTH: EVIDENCE FROM TIME SERIES AND PANEL GRANGER NON-

CAUSALITY TESTS

Arshia Amiri

BILATERAL EFFECTS BETWEEN HEALTH EXPENDITURES, HEALTH OUTCOMES AND

ECONOMIC GROWTH: EVIDENCE FROM TIME SERIES AND PANEL GRANGER NON-

CAUSALITY TESTS

ABSTRACT

TIIVISTELMÄ

ACKNOWLEDGEMENTS

CONTENTS

LIST OF TABLES

LIST OF FIGURES

ABBREVIATIONS

1 INTRODUCTION

2 THEORETICAL FRAMEWORK, BACK- GROUND AND LITERATURE REVIEW

2.1 THEORY OF HEALTH PERFORMANCE AND ECONOMIC GROWTH

2.2 BACKGROUND AND LITERATURE REVIEW