UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES HO CHI MINH CITY THE HAGUE VIETNAM THE NETHERLANDS VIETNAM NETHERLANDS PROGRAMME FOR M A IN DEVELOPMENT ECONOMICS VIETNAM ECONOMIC GROWTH AND SAVING[.]
UNIVERSITY OF ECONOMICS HO CHI MINH CITY VIETNAM INSTITUTE OF SOCIAL STUDIES THE HAGUE THE NETHERLANDS VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS VIETNAM ECONOMIC GROWTH AND SAVING ANALYSIS IN THE PERIOD 1989 - 2012 BY LE DUC ANH MASTER OF ARTS IN DEVELOPMENT ECONOMICS HO CHI MINH CITY, DECEMBER 2014 UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES HO CHI MINH CITY VIETNAM THE HAGUE THE NETHERLANDS VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS VIETNAM ECONOMIC GROWTH AND SAVING ANALYSIS IN THE PERIOD 1989 - 2012 A thesis submitted in partial fulfilment of the requirements for the degree of MASTER OF ARTS IN DEVELOPMENT ECONOMICS By Le Duc Anh Academic Supervisor: Dr Dinh Cong Khai HO CHI MINH CITY, DECEMBER 2014 Acknowledgement Foremost, I would like to express my sincere gratitude to my supervisor Dr Dinh Cong Khai for the continuous support of my M.A study and research, for his patience, motivation, enthusiasm and immense knowledge His guidance helped me in all the time of research and writing of this thesis Besides my supervisor, I would like to thank VNP teaching staff for their encouragement, insightful comments, and hard questions Last but not the least, I would like to thank my family: my parents, my wife and my brothers for supporting me spiritually throughout my life OUTLINE Abstract Chapter 1: Introduction 1.1 Problem statement 1.2 Research objectives 1.3 Research questions 1.4 Methodology 1.5 Research scope 1.6 Structure of the study Chapter 2: Literature Review 2.1 Theoretical literature 2.2 Empirical literature 10 2.3 Conceptual framework 18 Chapter 3: Research Methodology 20 3.1 Data collection 20 3.2 Unit root test 21 3.3 Cointegration analysis 23 3.4 Granger causality analysis 24 Chapter 4: Empirical Analysis 26 4.1 Empirical evidence 26 4.1.1 Unit root test results 26 4.1.2 Cointegration test results 28 4.1.3 Granger causality test results 30 4.2 Discussion of findings 33 Chapter 5: Concluding Remarks 40 References Appendix LIST OF TABLES AND FIGURES Figures Figure 1: Relationship between saving and economic growth stated by Solow (1956) Figure 2: The grapth of LGDP, LGDS and LFDI series 26 Tables Table 1: The summary of the main empirical studies 13 Table 2: Unit root test of variables at level value 27 Table 3: Unit root test of variables at the first difference 28 Table 4: Results of ARDL bound test 29 Table 5: Estimated long run coefficients using the conditional ARDL(2, 0, 0) 30 Table 6: Results of VECM based on error terms taken from the conditional ARDL 31 Table 7: Results of diagnostic tests for equation 11 32 Table 8: Results of short-run Granger causality 33 Table 9: Average values of FDI, GDP and GDS in three stages 34 Table 10: Commercial bank network density 36 Table 11: Average percentage of rural population over total population in three stages 36 Table 12: Vietnam stock market in the period 2006 – 2012 37 LIST OF ABBREVIATIONS Abbreviation Meaning ADF Augmented Dickey-Fuller test AIC Akaike Information Criterion ARDL Autoregressive Distributed Lag ASEAN Association of Southeast Asian Nations ECT Error Correction Term FDI Foreign Direct Investment GDP Gross Domestic Product GDS Gross Domestic Saving LGDP Logarithm of Gross Domestic Product LGDS Logarithm of Gross Domestic Saving LFDI Logarithm of Foreign Direct Investment OLS Ordinary Least Squares PP Phillips-Perron test VECM Vector Error Correction Method WTO World Trade Organization VIETNAM ECONOMIC GROWTH AND SAVING ANALYSIS IN THE PERIOD 1989 – 2012 Abstract This paper employs Autoregressive-Distributed Lag model to detect the cointegrating vectors amongst three variables that are gross domestic saving per capita, gross domestic product per capita and foreign direct investment being stationary at the mixture of I(0) and I(1) in the period 1989 – 2012 The results not support for the hypothesis as Solow (1956) states in which domestic saving and foreign direct investment are sources for domestic investments, then push economic growth The long run relationship amongst these indicators is consistent with some recent empirical studies for the case of developing countries That is, the long run causal direction running from economic growth to domestic saving Additionally, foreign direct investment does not cause economic growth in both short and long run, but tends to reduce domestic saving in long run Furthermore, combining the estimated results with some statistical measurements of demographic change in total population and financial market development gives to us evidences in which financial market in Vietnam is still weak, thus does not strengthen the channel of domestic private saving accumulation, especially from huge amount of rural population FDI has increased annually, but it only takes a small percentage in GDP As a result, there is no evidence to support for the nexus from saving to growth GDS increasing annually is explained by a rise of GDP, and perhaps by some positive demographic change in total population Last but not the least, in long run FDI tends to reduce domestic saving due to ineffective channel of domestic capital accumulation, especially for Vietnam stock market Then, it casts doubts on an assumption of Solow about the positive nexus running from saving to growth, especially for developing countries Chapter 1: Introduction In this chapter, the study presents the reason of choosing this kind of research for the specific case of Vietnam to analyze the interaction amongst domestic saving, foreign direct investment and economic growth in the period 1989 – 2012 1.1 Problem statement The central assumption of the Solow’s (1956) growth model is a positive nexus between saving rate and economic growth in which saving rate plays a role of conditional factor to push growth This model implies some policy implications for a country to concentrate on increasing saving, thus economic growth will increase in response However, inversely if there is a possibility of negative impact, a country shall focus on removing barriers to growth instead of accumulating saving There are some robust empirical findings about the positive association between saving and growth Nevertheless, this correlation does not imply a causal direction amongst them, and then this controversy is still unsolved Moreover, the debate concerning with the priority of policy implication for these indicators is an important issue raised at the current macroeconomics, as stated by Schmidt-Hebbel et al (1996) Hence, determining the causal direction of saving-growth linkage is crucial, and has many implications for policy makers in developing countries At the notion of growth to saving, it has been supported by some empirical evidences of several recent studies analyzing in developing countries For instance, in the research by Ramesh (2006), he finds that in the group of low-middle income countries there is mostly the causal direction from growth to saving Furthermore, the study by Pradeep, Pravakar, and Ranjan (2008) also shows the empirical evidences from countries – Bangladesh, India, Pakistan, Srilanka and Nepal – there is a causal direction from growth to saving, however, only Bangladesh with bi-directional causality In an open economy, two capital sources contributing to domestic investment are domestic and foreign saving With foreign saving, it is represented by some forms of international capital inflows Theoretically, these inflows are assumed to be additive, or supplement domestic saving to increase an amount of investment, hence to push economic growth This hypothesis is seemingly relevant in some certain capital inflows, especially to foreign direct investment (noted as FDI) because this inflow is more stable than the other components of foreign saving in term of predictability and volatility (Taylor, 1997 and Lipsey, 1999 cited in Maite, Ana and Vicente, 2004) In the research composed by Matie, Ana and Vicente (2004), they analyze the role of FDI to the nexus between economic growth and domestic saving in Mexico and find that FDI causes both GDP and GDS positively, hence to reinforce the relationship between economic growth and domestic saving However, this hypothesis does not hold in the study by Ahmad, Marwan and Salim (2002) for the case of Malaysia, Thailand and Philippines They find foreign savings affecting domestic saving negatively in both short and long run Moreover, in long run the causal direction runs from foreign to domestic saving, but the inverse direction does not exist In the history, Latin America crisis happened in Mexico after the peso was devaluated by the authorities in 1994, then quickly spreading to several other countries in this region Consequently, a huge amount of capital outflow left these countries into shambles However, on the contrary to that of Latin America countries, East Asian economies were not seriously affected by the crisis Then, based on this fact, some economists conclude that Latin America countries were more sensitive to the shift of investor sentiments and the fluctuation of international capital outflows than Asian countries This difference relates to domestic saving rate in which the higher the domestic saving in a country is, therefore can reduce the international capital outflows’ influence through a crisis Furthermore, in this period, an increase of foreign investment in East Asian countries correlated with a huge decrease of domestic saving rate in Latin America countries In the mid of year 1997, the Asian crisis started from Thailand, then influenced to several other countries Based on this fact, it casts doubts on the economic notion held earlier through Latin America crisis by some economists The Asian economies’ currencies were devaluated sharply, then leading to a massive outflow of foreign capital at the end of year 1997 On the other hand, the experience was similar with the case of Latin America crisis The retreat of foreign capital emphasizes the necessity of more domestic financing seen as a complement to that decline in capital As dramatized by Latin America and Asian crisis, therefore, it is important to analyze the nexus between domestic and foreign saving because foreign capital could be withdrawn as easily as it enters, consequently a domestic economy could be left into shambles through a crisis In recent years, Vietnam’s economy has succeeded in the new stage of development after opening the economy in 1988 Especially, Vietnam has been a member of ASEAN since 1995, and WTO since 2007 Then, its economic growth has increased rapidly in the period 1989 – 2012 Tariffs and trade barriers have been reduced step by step amongst members in ASEAN and WTO, therefore, Vietnam’s trade balance has been improved as well From a poor country with per capita income only equals to 290 U.S dollars a year in 1989, Vietnam has reached to nearly 1000 U.S dollars a year in 2012 Moreover, domestic saving rate and foreign direct investment in Vietnam have also increased annually through this period In detail, its gross domestic saving has changed from 0.837 to 26.9 billion U.S dollars, and FDI has also moved from 0.01 to 4.7 billion U.S dollars, respectively in 1989 and 2012 (World Bank Indicators, 2012) from growth to saving, except Singapore Pradeep, Granger GDS, Real A period of (1) Mixed results about Pravakar, causality test, per capita 1960-2004 the effect of a and VECM, GDP, Bangladesh, determinant on saving Ranjan Dynamic inflation rate, India, function (2008) OLS dependency Pakistan, Sri (2) Causal direction ratio, foreign Lanka and from growth to saving, saving, Nepal except Bangladesh with interest rate, bi-directional causality financial sector development, and bank branch density Nurudeen Johansen co- GNS and Annual data (1) Existing long run (2010) GDP in Nigeria equilibrium between test followed for years gross national saving by VECM, 1970 to and gross domestic Granger 2007 product integration causality (2) Causal direction runs from growth to saving 17 2.3 Conceptual framework In the Solow (1956) model, it assumes that higher saving rate will definitely lead to higher domestic investment financed by two capital resources known as domestic and foreign saving Furthermore, in this study, foreign saving is assumed to complement domestic saving, then will push economic growth through the channel of domestic investment These components are expected to positively affect economic growth, hence the conceptual framework is constructed as below Gross Domestic Product (GDP) Gross Domestic Foreign Saving Saving (GDS) represented by FDI As stated by Solow (1956) we should measure domestic saving and economic growth in a form of per capita unit to capture the population growth effect Moreover, FDI seems to be a permanent component of foreign saving when comparing with the other inflows in terms of volatility and predictability Therefore, an addition of foreign direct investment seen as a component of foreign saving, this, then contributes to domestic investment, in turn pushes economic growth Last but not the least, because the nature of cointegration analysis, then with each pair amongst three variables of interest, the analytical model will measure two possible causal directions In the study, with three variables as mentioned above – GDS, FDI and GDP – we have three pairs of variables, thus conceptual framework is built as 18 shown above Once more, we should note that even the econometrical method step by step will estimate all possible cases of causality amongst variables, however, the main objectives of the study are to focus on interpreting the causal direction from GDS and FDI to GDP in long run described by dashed lines in the conceptual framework, and their effects are expected to positively cause GDP Finally, it is important to note that in the context of three variable cointegration analysis, the study will only measure the independent effect of each explanatory variable on economic growth, according to the research objectives raised above Furthermore, the econometric technique employed in this study can’t deal with the mixed effects of independent variables to the dependent variable 19 Chapter 3: Research Methodology In this chapter, the study will present the relevant time-series econometric techniques employed to analyze the specific data set of Vietnam for years 1989 to 2012 The following sections will discuss in detail each method necessarily to measure Granger causality amongst three variables of interest (GDS, GDP and FDI) 3.1 Data collection In the study, the variables are gathered from World Bank Indicators publicly released in 2012 Three variables used in the econometric model are collected from year 1989 to 2012 because of the limited data and historical milestone in Vietnam Per capita GDP, per capita GDS (simply denoted as GDP and GDS respectively) and FDI are measured in real term at year 2005 (U.S dollar unit) Then transforming the original series into logarithmic values for each variable, hence the first difference represents for the growth rate Moreover, because LGDP and LGDS are stationary at level 1, but LFDI is stationary at level (results of unit root test shown in Chapter 4) As a result, ARDL cointegration bound test will be employed for the mixture of I(0) and I(1) Furthermore, in the data set of World Bank 2012, the indicators of interest are calculated respectively as followings: (1) Foreign direct investment is measured by the net capital inflows to invest into a domestic enterprise for acquiring a lasting management interest with an amount of at least 10 per cent of voting stock In detail, FDI is the sum of equity capital, reinvestment of earnings, and other short-term and long-term capital as shown in the balance of payments (2) GDP is calculated as the sum of gross value added by all resident producers plus any product taxes, and minus any subsidies not included in the value of the products However, this calculation does not include deductions for depreciation of capital assets, or for depletion and 20 degradation of natural resources (3) Gross domestic savings are calculated as GDP less final consumption expenditure or total consumption Finally, as the calculating methods mentioned above, foreign direct investment and gross domestic saving are measured as a percentage of GDP, therefore, we multiply these series’ values with GDP measured in billions of U.S dollars constant at year 2005 Furthermore, two series, GDS and GDP, should be measured in term of per capita as suggested by Solow (1956), then we need to divide these two series’ values with total population per year in the period 1989 – 2012, to receive the measure unit of interest In sum, GDS and GDP are measured at U.S dollar per capita a year, but FDI is measured at U.S billion dollars a year 3.2 Unit root test One of the most important issues in time-series analysis that is a spurious regression, then affects the reliability of estimated coefficient meanings Moreover, most of macroeconomic time-series variables have trend, hence they usually are not stationary Then, the OLS regression on these variables is not applied There are many methods could be employed to solve this problem, and usually are to transform original series into values of logarithm, or to detrend the series Additionally, taking the difference of series is mostly used to obtain a stationary series According to Gujarati (2003), “a time series is said to be stationary if its mean and variance are constant over time and the value of the covariance between the two periods depends only on the distance or gap or lag between the two time periods and not the actual time at which the covariance is computed” Furthermore, Asteriou (2007) states that logarithm transformation and taking the difference of original series are relevant, usually finds these kind of series are stationary at the first difference denoted as I(1) There are two most-frequentlyemployed techniques to test a problem of stationary known as Augmented DickeyFuller (ADF) and Phillips-Perron (PP) test 21 In Augmented Dickey-Fuller method, the test is less restricted than Simple DickeyFuller because the error term is unlikely white noise, therefore, ADF test should be employed rather than Simple Dickey-Fuller test Three possible forms of ADF test are described by the following equations: 𝑝 𝛥𝑌𝑡 = 𝛿𝑌𝑡−1 + ∑𝑖=1 𝛽𝑖 𝛥𝑌𝑡−1 + 𝑢𝑡 𝑝 𝛥𝑌𝑡 = 𝛼 + 𝛿𝑌𝑡−1 + ∑𝑖=1 𝛽𝑖 𝛥𝑌𝑡−1 + 𝑢𝑡 𝑝 𝛥𝑌𝑡 = 𝛼 + 𝛶𝑇 + 𝛿𝑌𝑡−1 + ∑𝑖=1 𝛽𝑖 𝛥𝑌𝑡−1 + 𝑢𝑡 (eq.1) (eq.2) (ep.3) Where equations 1, and present for three cases that are without intercept and trend, with intercept and no trend, at last with drift and trend, respectively However, which equation should be applied that depends on a specific case of data It is important to note that equation and are more important than equation because the intercept coefficient just only plays a role of rescaling On the other hand, we should check whether or not Y series has a trend In Phillips-Perron test, it allows for a fairly mild assumption about the error term distribution compared with that of ADF – iid(0, σ2 ) And the PP test equation is written as below: 𝛥𝑌𝑡 = 𝛼 + 𝛿𝑌𝑡−1 + 𝑢𝑡 (eq.4) Last but not the least, in empirical analysis, researchers usually employ both of ADF and PP test to conclude about a problem of non-stationary It is often more reliable when two tests show a same result If |δ| < in both ADF and PP test, it is concluded there is no unit root existence or series is said to be stationary 22 3.3 Cointegration analysis In order to empirically analyze the long run nexus and short run dynamic amongst the variables, we apply autoregressive distributed lag (ARDL) cointegration technique as a general vector autoregressive (VAR) model of order p in the vector of these variables (with p is the maximum lag length of dependent variable) The ARDL bound testing methodology composed by Pesaran et al (2001) has some advantages compared to Johansen cointegration test that include: The first is that ARDL does not need all variables of interest must be integrated of the same order, on the other hand it can be applied when the variables of interest are integrated at the mixture of I(0) and I(1) The second advantage is that ARDL test is still relatively effective in the case of finite small sample size The last advantage is that ARDL regression gives unbiased estimations of long run model (Harris and Sollis, 2003 cited in Anshul, 2013) The ARDL models used in this study are expressed as followings: 𝑝 𝛥𝐿𝐺𝐷𝑆𝑡 = 𝑎01 + 𝑏1𝑡 𝐿𝐺𝐷𝑆−1 + 𝑏2𝑡 𝐿𝐺𝐷𝑃−1 + 𝑏3𝑡 𝐿𝐹𝐷𝐼−1 + ∑𝑖=1 𝑎1𝑖 𝛥𝐿𝐺𝐷𝑆𝑡−𝑖 + 𝑞2 ∑𝑞𝑖=1 𝑎2𝑖 𝛥𝐿𝐺𝐷𝑃𝑡−𝑖 + ∑𝑖=1 𝑎3𝑖 𝛥𝐿𝐹𝐷𝐼𝑡−𝑖 + 𝑢1𝑡 (eq.5) 𝑝 𝛥𝐿𝐺𝐷𝑃𝑡 = 𝑎01 + 𝑏1𝑡 𝐿𝐺𝐷𝑃−1 + 𝑏2𝑡 𝐿𝐺𝐷𝑆−1 + 𝑏3𝑡 𝐿𝐹𝐷𝐼−1 + ∑𝑖=1 𝑎1𝑖 𝛥𝐿𝐺𝐷𝑃𝑡−𝑖 + 𝑞2 ∑𝑞𝑖=1 𝑎2𝑖 𝛥𝐿𝐺𝐷𝑆𝑡−𝑖 + ∑𝑖=1 𝑎3𝑖 𝛥𝐿𝐹𝐷𝐼𝑡−𝑖 + 𝑢2𝑡 (eq.6) 𝑝 𝛥𝐿𝐹𝐷𝐼𝑡 = 𝑎01 + 𝑏1𝑡 𝐿𝐹𝐷𝐼−1 + 𝑏2𝑡 𝐿𝐺𝐷𝑆−1 + 𝑏3𝑡 𝐿𝐺𝐷𝑃−1 + ∑𝑖=1 𝑎1𝑖 𝛥𝐿𝐹𝐷𝐼𝑡−𝑖 + 𝑞2 ∑𝑞𝑖=1 𝑎2𝑖 𝛥𝐿𝐺𝐷𝑆𝑡−𝑖 + ∑𝑖=1 𝑎3𝑖 𝛥𝐿𝐺𝐷𝑃𝑡−𝑖 + 𝑢3𝑡 (eq.7) Where LGDS is logarithmic per capita gross domestic saving, LGDP is logarithmic per capita gross domestic product, and FDI is logarithmic foreign direct investment Therefore, the first difference measures growth rate of each variable, and u1t , u2t , u3t are the error terms 23 The bound test is based on the joint F-statistic coefficient test which its asymptotic distribution is non-standard under the null hypothesis of no cointegration The first step in the ARDL bound test is to estimate separately three equations (5, and 7) by ordinary least squares regression (OLS) The estimation of these equations detects for an existence of the long run relationship by conducting an F-test for the joint significance of the lagged levels of the variables, on the other hand it means 𝐻0 : 𝑏1𝑡 = 𝑏2𝑡 = 𝑏3𝑡 = For a given siginificance level, lower-bound and upper-bound critical values, can be determined and given in the research by Pesaran (2001) In the three equations (5, and 7) above, the lagged level terms are calculated on the assumption that all variables included in the ARDL model are integrated of order zero, while the latter lagged first-difference terms are calculated on the assumption that the variables are integrated of order one The null hypothesis of no cointegration is rejected when the statistical value of the test is higher the upper critical bound value, inversely it is accepted if the F-statistic is lower than the lower bound value Other cases, the cointegration test is inconclusive The employment of this approach is guided by the short time span of the data set, and the case of the mixture of I(0) and I(1) amongst variables Furthermore, the determination of optimal lag order is based on Akaike information criterion (AIC) while estimating ARDL bound test equation 3.4 Granger causality analysis If cointegration exists amongst variables, the conditional ARDL (p, q1, q2) long run model for LGDS or LGDP or LFDI, could be estimated as forms below: 𝑝 𝑞 𝑞 𝑝 𝑞 𝑞 𝐿𝐺𝐷𝑆𝑡 = 𝑎01 + ∑𝑖=1 𝑎1𝑖 𝐿𝐺𝐷𝑆𝑡−𝑖 + ∑𝑖=0 𝑎2𝑖 𝐿𝐺𝐷𝑃𝑡−𝑖 + ∑𝑖=0 𝑎3𝑖 𝐿𝐹𝐷𝐼𝑡−𝑖 + 𝜀1𝑡 𝐿𝐺𝐷𝑃𝑡 = 𝑎01 + ∑𝑖=1 𝑎1𝑖 𝐿𝐺𝐷𝑃𝑡−𝑖 + ∑𝑖=0 𝑎2𝑖 𝐿𝐺𝐷𝑆𝑡−𝑖 + ∑𝑖=0 𝑎3𝑖 𝐿𝐹𝐷𝐼𝑡−𝑖 + 𝜀2𝑡 𝑝 𝑞 𝑞 (eq.8) (eq.9) 𝐿𝐹𝐷𝐼𝑡 = 𝑎01 + ∑𝑖=1 𝑎1𝑖 𝐿𝐹𝐷𝐼𝑡−𝑖 + ∑𝑖=0 𝑎2𝑖 𝐿𝐺𝐷𝑆𝑡−𝑖 + ∑𝑖=0 𝑎3𝑖 𝐿𝐺𝐷𝑃𝑡−𝑖 + 𝜀3𝑡 (eq.10) 24 As the suggestion by Anshul (2013), we should estimate the short-run dynamic coefficients by adding error correction terms taken from the long run equation The long run cointegration between the variables implies that Granger causality exists in at least one direction which is determined by the F-statistic of the lagged first-difference terms and the error-correction term as shown in equations 11 to 13 In detail, the shortrun causal effect is determined by F-statistic on the coefficients of lagged firstdifference terms of explanatory variables while the F-statistic on the coefficient of the lagged error-correction term (denoted as ECT) represents for the long run causal relationship However, there are only equations where the null hypothesis of no cointegration rejected would be estimated with an error-correction term 𝑝 𝑝 𝑝 𝑝 𝑝 𝑝 𝛥𝐿𝐺𝐷𝑆𝑡 = 𝑎0 + ∑𝑖=1 𝑎1𝑖 𝛥𝐿𝐺𝐷𝑃𝑡−𝑖 + ∑𝑖=1 𝑎2𝑖 𝛥𝐿𝐹𝐷𝐼𝑡−𝑖 + 𝛼𝐸𝐶𝑇1,𝑡−1 + 𝑢1𝑡 𝛥𝐿𝐺𝐷𝑃𝑡 = 𝑎0 + ∑𝑖=0 𝑎1𝑖 𝛥𝐿𝐺𝐷𝑆𝑡−𝑖 + ∑𝑖=0 𝑎2𝑖 𝛥𝐿𝐹𝐷𝐼𝑡−𝑖 + 𝛼𝐸𝐶𝑇2,𝑡−1 + 𝑢2𝑡 𝛥𝐿𝐹𝐷𝐼𝑡 = 𝑎0 + ∑𝑖=0 𝑎1𝑖 𝛥𝐿𝐺𝐷𝑆𝑡−𝑖 + ∑𝑖=0 𝑎2𝑖 𝛥𝐿𝐺𝐷𝑃𝑡−𝑖 + 𝛼𝐸𝐶𝑇3,𝑡−1 + 𝑢3𝑡 (eq.11) (eq.12) (eq.13) Where a1i and a2i are the short run coefficients of the model being convergence to equilibrium and α is the coefficient of the speed adjustment Furthermore, ECT1 , ECT2 and ECT3 are error-term series taken from long run equations 8, and 10, respectively The equations (11) to (13) are estimated by OLS regression separately to obtain short-run and long run effects amongst these variables Once again it is important to note that only an equation with the null hypothesis of no cointegration rejected could be estimated with error term 25 Chapter 4: Empirical Analysis In Chapter 4, the study presents estimated results for the interaction between economic growth and saving with the confirmation of FDI in Vietnam Step by step while analyzing the 1989-2012 data, specific problems are solved and discussed as the followings 4.1 Empirical evidence In this part, the study will employ the econometric techniques mentioned briefly at Chapter in which results of unit root test, ARDL cointegration bound test, long-run coefficient estimation and VECM conditional on ARDL are presented respectively 4.1.1 Unit root test results The three variables are plotted on the graph as shown on Figure below There exists a deterministic trend inside each series Therefore, ADF and PP tests employed to detect a problem of unit root existence should have a trend factor in unit-root test equations Figure 2: The graph of LGDP, LGDS and LFDI series 24 22 20 18 16 90 92 94 96 98 LFDI 26 00 02 LGDP 04 06 LGDS 08 10 12 In detail, we could see that LFDI is said to be stationary at level by both unit root test methods as shown on Table Inversely, LGDP and LGDS are not stationary at level, even the significance level is 10 per cent Detecting orders of stationary in time series analysis is the most important step because it will influence which method should be employed to estimate short-run and long run effects amongst variables of interest Then, the next step is to check whether LGDP and LGDS are stationary at level or not Table 2: Unit root test of variables at level value ADF PP τ τ log(GDP) -3.01 -1.62 log(GDS) -2.49 -3.13 log (FDI) -9.31*** -7.43*** Variable Notes: the subscript τ in the model allows for a drift and deterministic trend These (*), (**) and (***) indicate the rejection of null hypothesis at 10 per cent, per cent and per cent respectively Additionally, critical value obtains from MacKinnon (1995) After taking the first difference of LGDP and LGDS, applying ADF and PP test, the results in Table show information as expected that two series are stationary at the first difference As Asteriou (2007) states most of economic indicators are trended, therefore, rarely they are stationary at level data and usually stationary after taking the first difference, and our research data is the case as he mentioned For the case of LGDS, the null hypothesis of non-stationary is rejected by two methods, thus this series is said to be stationary at the first difference 27 Table 3: Unit root test of variables at the first difference Variable ADF PP τ Μ τ Μ ΔLGDP -4.03** -2.64* -2.96 -2.74* ΔLGDS -3.92** -5.13*** -6.39*** -5.09*** Notes: the subscript τ in the model allows for a drift and deterministic trend while μ allowing for a drift These (*), (**) and (***) indicate the rejection of null hypothesis at 10 per cent, per cent and per cent respectively Additionally, critical value obtains from MacKinnon (1991) Tải FULL (76 trang): https://bit.ly/3h9DmHf Dự phòng: fb.com/TaiHo123doc.net Last but not the least, for the case of unit root tests of LGDP, PP test seems to fail to reject the null hypothesis of non-stationary at the significance level per cent Then, we must test on an assumption of ADF in which error term series of this test whether follows the normal distribution or not As shown at Appendix A.1, this error term series follows a normal distribution, therefore, the results of ADF test for LGDP are robust Then, we could conclude that LGDS and LGDP are stationary at level 1, and LFDI is stationary at level Finally, ARDL method should be applied to test a cointegration existence amongst these variables 4.1.2 Cointegration test results Based on the results of unit-root test, Johansen cointegration test for multivariate causality could not be applied because this method only works if all relevant variables are stationary at the same level Therefore, ARDL bound test should be employed instead of Johansen method As mentioned in Chapter 3, ARDL is also effective in the case of finite short time span This is really important when analyzing Vietnam’s data because the time span could not be longer the period 1989 – 2012 for three variables of interest 28 Table 4: Results of ARDL bound test Dependent variable Independent Lag variables Equation F- Result statistic LGDS LGDP, LFDI 8.61 Cointegration LGDP LGDS, LFDI 0.38 No cointegration LFDI LGDS, LGDP 3.82 No cointegration Equation Equation Tải FULL (76 trang): https://bit.ly/3h9DmHf Dự phòng: fb.com/TaiHo123doc.net Lower-bound critical value for “without intercept and trend” at 1% = 3.88 Upper-bound critical value for “without intercept and trend” at 1% = 5.30 Applying the ARDL cointegration tests, we estimate three cointegration equations – as shown in Chapter The maximum of lag length is because the time span of Vietnam data set is short, only 24 years, or on the other hand we have only 24 observations Reducing the lag length of the first-difference terms corresponding to each variable in each equation step by step, then we receive the optimum lag length is based on Akaike information criterion for equations to 6, but equation with the optimum lag length is (see at Appendix A.2) the F statistical values are calculated for the joint hypothesis of coefficients of lagged level terms in each equation respectively given in Table Comparing these statistical values with critical ones given in the research by Pesaran (2001), we could only reject the null hypothesis of no cointegration in equation That means there is an existence of cointegration direction from LGDP and LFDI to LGDS Nevertheless, in equation to 7, statistical values are less than the lower-bound critical value, thus we accept the null hypothesis of no cointegration 29 4.1.3 Granger causality test results The next step of ARDL bound test is to estimate a long run coefficient equation in the form of equation Detecting the optimum lag length based on Akaike information criterion as the same way when estimating ARDL cointegration equations – (see at Appendix A.3), finally ARDL (2, 0, 0) is employed to estimate long run coefficients between LGDS, LGDP and LFDI The results of long run coefficient equation are shown on Table As we could see, in long run LGDP really causes and pushes LGDS represented by the coefficient value equals to 1.06 and is significant at per cent Inversely LFDI causes, but tends to reduce LGDS represented by the coefficient value equals to -0.12 and is significant at percent Table 5: Estimated long run coefficients using the conditional ARDL(2, 0, 0) Equation 8: LGDS is dependent variable Variable Coefficient t-Statistic Probability C -0.95 -0.74 0.47 LGDS(-1) 0.33 3.24 0.00 LGDS(-2) 0.04 0.43 0.67 LGDP 1.06 5.52 0.00 LFDI -0.12 -2.09 0.05 R2 = 0.97 Durbin-Watson Stat =2.06 F-statistic p-value = 0.00 In the causality analysis, estimating the long run adjustment coefficient 𝛼 is the most important because it shows how strong the nexus amongst variables is Based on the long run coefficient equation estimated in form of equation above, we receive the error-term series (denoted as ECT – Error Correction Term) used to estimate the adjustment coefficient As mentioned above, we should estimate the short run 30 coefficients combined with error terms Then the VECM is applied in form of equation 11 because only this equation has the cointegration vector The optimum lag length is detected through estimating Unrestricted VAR amongst ΔLGDP, ΔLGDS and ΔLFDI series Then the optimum lag length is selected by AIC (see more at Appendix A.3) With the VECM results shown on Table 6, the long run adjustment coefficient (α) equals to -0.706 and is significant at per cent It means there is an existence of long run causal direction running from LGDP and LFDI to LGDS as early proved by the ARDL bound test results above The α value is -0.706 means the speed of adjustment to equilibrium after an economic shock is high Approximately 70.6% of disequilibrium from the previous year’s shock could converge back to the long run equilibrium in the current year (Asteriou, 2007) Table 6: Results of VECM based on error terms taken from the conditional ARDL VECM in a form of Equation 11: ΔLGDS is dependent variable Variable Coefficient t-statistic Probability C 0.001 0.008 0.993 ΔLGDP 2.068 0.678 0.507 ΔLFDI -0.105 -0.916 0.372 ECM(-1) -0.706 -2.184 0.043 R-squared = 0.277 0.277 Durbin-Watson 1.499 stat Furthermore, this equation also passes all diagnostic tests including Ramsey RESET test, White Heteroskedasticity test, Breusch-Godfrey Serial Correlation test and JarqueBerra Statistic for normality at the significant level per cent Furthermore, the results of CUMSUM and CUMSUM-Squared tests (see more at Appendix A.3) show the estimated coefficients and variance of VECM are stable in the period 1989 – 2012, on 31 6670216 ... analyzing in developing countries For instance, in the research by Ramesh (2006), he finds that in the group of low-middle income countries there is mostly the causal direction from growth to saving. .. of saving- growth linkage is crucial, and has many implications for policy makers in developing countries At the notion of growth to saving, it has been supported by some empirical evidences of. .. data of GDS and GDP from 25 countries for years 1960 to 2001, and also divides the countries of interest into groups including low income, low middle, upper middle and high income In low income countries,