CHAPTER 3 FOREIGN DIRECT INVESTMENT AND TRADE IN ASEAN COUNTRIES: A SPATIAL ECONOMETRIC ANALYSIS
3.2.1 The development of the spatial econometrics
Recently, spatial econometric techniques have been widely used to examine spatial effects in many research areas. Anselin (1988) is the first study introducing models and methods of spatial econometrics specifically. This book presents the foundations of spatial econometrics, including the spatial effects, the spatial econometrics models as well as how to estimate and test the models.
Anselin (2010) takes the year 1979 as “the historical starting point for spatial econometrics” and defines spatial econometrics as “the collection of techniques that deal with the peculiarities caused by space in the statistical analysis of regional science models”.
Elhorst (2014) can be considered as a very important introduction to the spatial econometric model. It develops and summaries three generations of spatial econometric models: the models based on cross-sectional data, the static models based on spatial panels and the dynamic spatial panel data models, as illustrated below.
36
Source: Elhorst (2014)
Figure 3.4 The spatial dependence models for cross-sectional data
Firstly, the relationship between the spatial dependence models based on cross-sectional data is summarized in Figure 3.4.5
Secondly, the spatial panel data models including the fixed-effect model, the random-effect model, the fixed-coefficient model, the random-coefficient model, and the multilevel model are summarized and expressed as follows:
𝑌𝑡 = 𝜌𝑊𝑌𝑡+ 𝛼𝜄𝑁+ 𝑋𝑡𝛽 + 𝑊𝑋𝑡𝜃 + 𝜇 + 𝜉𝑡𝜄𝑁+ 𝑢𝑡, 𝑢𝑡= 𝜆𝑊𝑢𝑡+ 𝜀𝑡,
where 𝜇 = (𝜇1, … , 𝜇𝑁)𝑇 .
Finally, the dynamic spatial panel model is constructed as:
𝑌𝑡 = 𝜏𝑌𝑡−1+ 𝛿𝑊𝑌𝑡+ 𝜂𝑊𝑌𝑡−1+ 𝑋𝑡𝛽1+ 𝑊𝑋𝑡𝛽2+ 𝑋𝑡−1𝛽3+ 𝑊𝑋𝑡−1𝛽4+ 𝑍𝑡𝜋 + 𝑣𝑡, 𝑣𝑡 = 𝜌𝑣𝑡−1+ 𝜆𝑊𝑣𝑡+ 𝜇 + 𝜉𝑡𝜄𝑁+ 𝜀𝑡,
𝜇 = 𝜅𝑊𝜇 + 𝜁.
5 𝑌 is an 𝑁x1 vector including one observation on the dependent variable for every unit in the sample (𝑖 = 1, … , 𝑁).
𝑋 represents an 𝑁x𝐾 matrix of exogenous explanatory variables, 𝑊 is a nonnegative 𝑁x𝑁 matrix illustrating the spatial arrangement of the units in the sample. SAC model includes both a spatially lagged dependent variable and a spatially autocorrelated error term and the SLX (spatial lag of X) model contains exogenous interaction effects model.
(Elhorst, 2014).
𝜃 = 𝛿𝛽
𝜃 = 0
𝛿 = 0
𝜃 = 0
𝜆 = 0 𝛿 = 0
𝛿 = 0 𝜃 = 0 𝜆 = 0
𝛿 = 0 𝜆 = 0 𝜃 = 0
General nesting spatial model 𝑌 = 𝛿𝑊𝑌 + 𝛼𝜄𝑁+ 𝑋𝛽 + 𝑊𝑋𝜃 + 𝑢
𝑢 = 𝜆𝑊𝑢 + 𝜀
SAC
𝑌 = 𝛿𝑊𝑌 + 𝛼𝜄𝑁+ 𝑋𝛽 + 𝑢 𝑢 = 𝜆𝑊𝑢 + 𝜀
Spatial Durbin model 𝑌 = 𝛿𝑊𝑌 + 𝛼𝜄𝑁+ 𝑋𝛽 + 𝑊𝑋𝜃
+ 𝜀
Spatial Durbin Error model 𝑌 = 𝛼𝜄𝑁+ 𝑋𝛽 + 𝑊𝑋𝜃 + 𝑢
𝑢 = 𝜆𝑊𝑢 + 𝜀
Spatial lag model 𝑌 = 𝛿𝑊𝑌 + 𝛼𝜄𝑁+ 𝑋𝛽 + 𝜀
Spatial Error model 𝑌 = 𝛼𝜄𝑁+ 𝑋𝛽 + 𝑢
𝑢 = 𝜆𝑊𝑢 + 𝜀 (If 𝜃 = 𝛿𝛽 then 𝜆 = 𝛿)
SLX
𝑌 = 𝛼𝜄𝑁+ 𝑋𝛽 + 𝑊𝑋𝜃 + 𝜀
OLS 𝑌 = 𝛼𝜄𝑁+ 𝑋𝛽 + 𝜀
37
Many researches adopt these models introduced above to deal with the causes and consequences of economic growth, FDI and trade. For example, Driffield (2006) states that the role of agglomeration and proximity is very important in models analyzing the spillover effects. In this paper, the author uses the dataset stratified by industry and region in the U.K. from 1984 to 1992 and develops a spatial GMM estimator to examine the local intra-industry spillovers and interregional spillover effects from FDI. The local intra-industry spillovers are confirmed.
However, no interregional spillover effects from FDI are found. The reason discussed is that the opposite directions of the national and local intra-industry spillovers from FDI on productivity might cancel off each other.
Fischer and Griffith (2008) introduce a comparison between the spatial econometric approach and the eigenfunction spatial filtering approach to account for spatial autocorrelation among flow residuals. This paper also uses the data of patent citations, including 12,432 knowledge flows across 112 European regions to illustrate an application of these approaches. The results show that both methods are not significantly different from each other and lie within the 95 percent confidence limits of the least squares estimates.
Blonigen et al. (2007) adopt the spatial econometric model on a panel data of the U.S.
outbound FDI into 35 countries from 1983 to 1998. They find the significantly positive spatial impact of FDI and the significantly negative spatial impact of the market potential. Regelink and Elhorst (2015) show that results in Blonigen et al. (2007) are not consistent with any of the FDI motivations. They use the same dataset as Blonigen et al. (2007), but exclude countries after considering their neighbors (their sample includes 20 European countries) from 1999 to 2008 and apply the spatial Durbin model. They show that there is competition in attracting U.S. companies among European countries. The results correspond to the pure vertical and export-platform FDI.
38
Ho et al. (2013) aim to examine the augmented Solow model by applying the dynamic spatial Durbin model with a sample of 26 OECD countries over the period 1971-2005. They find that there is a positive spillover effect of growth from one country to its trade partners.
Benos et al. (2015) introduce a non-linear regression model using the annual panel data for 1,273 NUTS III (Nomenclature of Units for Territorial Statistics) regions in seven EU member- states, namely the U.K., France, Germany, Sweden, Italy, Spain and the Netherlands over the 1990- 2005 period. In their model, geographical, economic and technological proximity weights are analyzed to show that the existence and magnitude of interregional spillovers play a key role in the process of EU regional development.
Abate (2016) applies the spatial Durbin Ramey-Ramey model by using the dataset of 78 countries for the period 1970-2010 to examine the effects of volatility on growth in the framework of spatial interactions. In contrast to previous papers, the results show that the spatial effect plays an important role that influences the relationship between volatility on the growth of a particular country.
Gutierrez-Portilla et al. (2018) use the data of Spain's FDI outflows to the top 50 host countries for the period 1996-2014 and apply the dynamic spatial Durbin model to examine whether the Spanish FDI outflows to a host country are influenced by the economic growth and the Spanish FDI outflows of other host countries in the sample. The empirical results show that in the pre-crisis period, there is a positive impact of direct investment in neighboring countries to the host country. However, in the crisis period, this impact disappears.
Gutierrez-Portilla et al. (2019) apply the panel spatial Durbin model by using the data of FDI inflows to Spain, at both the aggregate and sectoral levels, over the period from 1996 to 2013. Their findings reveal that inward FDI in one region is complementary to that in neighbor regions.
39
Kim (2020) adopts the dynamic spatial Durbin model for the Schumpeterian technology diffusion model by using the dataset of 11 Asian countries including China, Hong Kong, Indonesia, India, Korea, Malaysia, Philippines, Singapore, Sri Lanka, Thailand and Taiwan over the period 1970–2014. The significant and negative results of the total effect on the lagged relative income confirm the hypothesis about the conditional convergence among Asian countries.
Amidi and Majidi (2020) use the dataset of 25 EU countries from 1992 to 2016 and adopt the spatial dynamic panel data model to confirm the positive effect of a country’s growth on its neighbor countries. Basile (2008) aims to test for the presence of spatial externalities on the process of economic growth of the European region. Based on a dataset of 155 European NUTS II regions for the period 1988-2000, this paper applies a semiparametric spatial Durbin model to show that regions surrounded by richer regions have higher expected growth rates than regions surrounded by poorer regions.
Ezcurra and Rios (2020) use the data of the European Quality of Government Index (EQI) and a dataset including GDP per capita, area, population, trust, latitude, tertiary education, high- tech sectors, patents, R&D expenditure, unemployment and a dummy variable for capitol region of 192 EU regions in three years 2010, 2013 and 2017 to calculate the hybrid spatial weights matrix combining geographical, technological and social distances. This paper confirms that there is a positive and statistically significant association between the quality of government in one region and governance in neighbor regions.
Hortas-Rico and Rios (2020) apply the spatial Durbin model based on the dataset of 5,556 Spanish municipalities for the period 2003-2011 to test the spatial effects on the optimal size of local governments. The results confirm that the spending decisions of a local government are influenced by its neighbor local governments’ spending decisions.
40
Stojčić and Orlić (2020) explore the intra- and interregional effects of foreign firms’ presence on 217 NUTS III regions regional productivity of indigenous firms in eight countries of Central and East Europe (Czech Republic, Hungary, Poland, Slovak Republic, Slovenia, Croatia, Romania and Bulgaria) for the period 2007-2011. The results from a spatial Durbin model reveal that FDI generates positive productivity effects on downstream firms within and across regions, while horizontal spillovers are negative. These effects become stronger for MNCs from neighboring regions and increase with distance.
The literature shows that not only the characteristics of a country but also the characteristics of its neighboring countries play a key role in the economic growth, FDI and trade of this country.
Most articles above investigate spatial effects among host countries from the perspective of home countries. In this chapter, we follow them to set up the spatial model but examine how FDI inflow is attracted to ASEAN countries from the perspective of host countries.