Are characteristics of metropolis matters for structural transformation of provinces: A spatial approach in the case of Vietnam

Abstract

This study analyzes the impact of metropolises on the structural transformation of provinces based on specific metropolises’ characteristics. Moran I index on the share of non-agricultural sectors in the economic structure is firstly employed to examine the spatial autocorrelation between provinces in Vietnam. Furthermore, the relationship between metropolises and the change in the economic structure of provinces is analyzed quantitatively using spatial panel data models based on data from the statistics yearbook of all 63 areas in the period 2015 to 2018. The research results confirm the role of metropolises in the process of structural transformation of the provinces. Provinces with a high proportion of non-agricultural sectors in the economic structure show the concentration around metropolises. The destination for migration, the origin of remittance, and the market for neighboring provinces, are the function of metropolises. These roles are seen as driving forces for off-farm activity through changes in the incomes of neighboring provinces. Accordingly, focusing on specific metropolises characteristics as well as enhancing regional connectivity will contribute to promoting effective provincial structural transformation.

Keywords:

• Provincial structure transformation
• metropolis
• Moran I
• spatial panel data model
• Vietnam

PUBLIC INTEREST STATEMENT

Different from the existing theories on economic structural transformation, which assume that the factors promoting structural change occur within the locality, this study approaches from a different angle with the assumption that the driving forces of structural transformation come from outside. By considering Vietnam as a typical case, this study analyzes the impacts of metropolises characteristics on the economic structural transformation of neighboring localities through a panel data spatial model. The results from the study contribute to expanding the understanding of the process of structural change in localities.

1. Introduction

Structural transformation is one of the earliest and most essential literature issues on development economics. According to several definitions found in the literature, structural transformation refers to the reallocation of economic activity across the broad sectors of agriculture, manufacturing, and services (Kuznets, 1973). This process plays an essential role in the development of any economy, particularly developing countries that almost depend on agriculture with low productivity. A well-known observation is that countries that have experienced remarkable growth also had a significant change in economic structure (Rodrik, 2016; De Vries et al., 2015). Structural transformation helps to allocate resources efficiently and increase productivity sustainably (2014) and creates opportunities for workers to have access to better technologies and accumulate capabilities (McMillan et al., 2014; Mujeri & Mujeri, 2021). Structural transformation also helps to shape the modern world. Not only in economic terms, it contributes to the creation of classes, social organizations, and political movements and promotes urbanization, and creates new social habits (Rodrik, 2016). As for the policy aspect, structural transformation is an important instrument that enhances the economy’s resilience after the shock and disturbance across various disciplines (Van Aswegen & Retief, 2020). Therefore, the determinants of structural transformation have been investigated for a long time, and the policymakers in countries have affected them to develop their economy. However, these determinants are examined in the country’s whole economy while any country has different regions with different characteristics and development levels; therefore, this raises the question of whether characteristics of the area with high-level development matter for structural transformation of lower development areas. Whereas areas with a higher level of development are often known as urban areas, or metropolises with a robust urbanization process and large populations, less developed areas are often rural or provinces. Obviously, urbanization in urban areas always brings certain effects on many socio-economic aspects for the locality and the region as a whole (Guan et al., 2018; Kuddus et al., 2020; Liang & Yang, 2019; Molaei Qelichi et al., 2017).

Only 35 years since the Doi Moi reforms, Vietnam has undergone one of the most rapid structural transformations of any low-income country. From 38.06% in 1986, the share of GDP in agriculture fell to 13.96% in 2019, while the share of employment in agriculture fell from 70.88% in 1991 to 37.22% in 2019. This contributes significantly to the enormous progress of Vietnamese economic development. Starting as one of the poorest nations with a GNI of less than the US $500/year in 1990, Vietnam became a lower middle-income country with a GNI of around US $1900/year in 2019 (in constant 2010 dollars).¹ This structural transformation is continuous, especially at the prefecture level, where each area has different characteristics. With the similarities of developing countries, the investigation of Vietnam provides a valuable case study for understanding the structural transformation of other low-income countries.

The main goal of this study is to focus on examining the effects of characteristics from the metropolis on the structural transformation process in provinces. Based on the previous literature on structural transformation and regional linkages, this study establishes the hypotheses of determinants in changing the economic structure in provinces. Then spatial panel data model is employed to examine whether factors that explain the structural transformation of provinces originated from metropolises following the spatial approach. Based on the results of this investigation, the policymakers have a comprehensive view of determinants in structural transformation and launch more appropriate policies in certain regions. The remainder of this study is organized as follows. The second section provides the theoretical background and develops several hypotheses that inform the research question. The third section describes the data and methods, followed by a section fourth presenting the results. The fifth section is a discussion, and the final section draws the conclusion.

2. Theoretical backgrounds and hypothesis developments

2.1. Structural transformation

For a long time, the determinants of structural transformation have been studied in both contexts: closed economy and open economy. Among a large body of literature on structural transformation in a closed economy, two mechanisms through which structural change can occur are changes in income and productivity. The first mechanism derives from the non-homothetic preferences, which are cross-sector differences in income elasticity. The income elasticity of demand for agricultural products is lower than that of non-agricultural products (Kongsamut et al., 2001). Therefore, the increase in income raises the need for non-agricultural products, leading to the movement of labor to non-agricultural sectors (Foellmi & Zweimüller, 2008). Using panel data for 39 countries during the postwar period, Comin, 2021Truong, 2021 show that income effects account for the bulk of the within-country evolution of sectoral reallocation. The second mechanism is associated with unbalanced technological progress across sectors and capital deepening. In an economy with two sectors with different productivity, if the ratio of the outputs of the two sectors is held constant, more and more of the total labor force must be transferred to the non-progressive sector (Baumol, 1967). Gollin et al. (2002) argue that improvements in agricultural productivity can hasten the start of industrialization. Ngai and Pissarides (2007) point to the implications of different sectoral total factor productivity (TFP) growth rates for structural change and consider that there is a shift of employment away from sectors with a high rate of technological progress toward sectors with low growth. Bustos et al. (2019) also confirm that increases in agricultural productivity can reallocate labor toward the industrial sector. However, improvements in agricultural productivity have been positive in the short run because of generating specialization in less innovative industries. In the long run, this has adverse effects on manufacturing productivity. Herrendorf et al. (2014) develop a multi-sector model that encompasses the main existing theories to account for many salient features of structural transformation. This model shows that both mechanisms are essential. With the open economy, along with income and productivity effects, the role of external factors which come from outside of the country is investigated. McMillan et al. (2014) discuss the role of globalization through which countries get some opportunities to attract investment and achieve a kind of structural change. Święcki (2017) and Teignier (2018) argue that international trade significantly accelerated the transition out of agriculture and contributed to the structural transformation in some countries. Besides globalization and trade, Van Neuss (2019) finds strong support for the role of outsourcing in contributing to structural change and particularly deindustrialization in both developed and developing countries.

2.2. Relationship between geographical areas

Another theoretical field is regional linkage theories dealing with aspects of the relationship between areas. Basically, these theories attempt to explain regional linkages through the regional growth and shrinkage processes. Perroux (1955) introduced the growth poles theory, which indicates that development is unbalanced; it appears firstly in points or development poles and spreads along diverse channels and with varying terminal effects on the whole economy. Supporting the growth poles theory, Friedman (1966) develops the core-periphery model, which spatially shows how economic, political, and cultural authority is dispersed in core or dominant regions and the surrounding peripheral and semi-peripheral regions. The core regions refer to centers, which are usually metropolitan. These centers typically dominate power in the economy and have a high potential for innovation (improvement) and growth, while the periphery with a lower level of development is dependent. Hirschman (1958) and Myrdal (1957) show two effects in which the growth pole influences the development of surrounding areas. Hirschman’s trickling down effects (or spread effects in Myrdal’s terms) include the role of the core in the diffusion of investment, innovation, and absorption capacity for disguised unemployed in periphery areas. Hirschman’s polarization effects (or backwash effects in Myrdal’s terms) refer to the unfavorable effect of core economic growth on periphery economic development through selective migration of the young, skilled, educated people and the capital movement from nearby areas to the center. In addition, in Myrdal’s opinion also, spread effects are weaker than backwash effects, whereas Hirschman considers that in the end, the spread effects would gain the upper hand over the polarization effects if the core had to rely on an important degree of periphery products for its own expansion. Douglass (1998) and Tacoli (2003) analyze the regional connection from a spatial perspective between two areas: urban and rural. The linkages between them include a set of flows: people, production, commodities, capital, information, waste, and pollution, and these flows have different directions. Through spatial statistical analysis methods such as Ripley’s K and Moran’s I functions on the point of interest (POI) data taken from the Gaode map, Xue et al. (2020) explain why surrounding urban areas are often the focus of non-agricultural activities. In addition, some other studies also use the basis of Moran I or bivariate Moran I to show the distribution of non-agricultural activities around urban areas (Cong & Kim, 2022; Yuan et al., 2017). From the above experimental results, it can be seen that the spatial factor cannot be ignored in the analysis of changes in the economic activities of geographical areas.

2.3. Hypotheses of metropolis impacts

According to the above theoretical basis of structural transformation, it is clear that the determinants of economic structural transformation include changes in productivity and income in the locality. Meanwhile, the theories of regional linkages in the previous section support the existence of different development across regions, and their relationship leads to variation within each region. This shows a clear connection between the structural transformation of a locality and the link between that locality and other localities. In particular, the effects that regional linkages have on local incomes and productivity and boost off-farm activity are likely to spur structural transformation. Although there has yet to be a general quantitative agreement on the specific definition of metropolises and provinces concepts for all countries in the world, there is a fundamental commonality in the criteria to distinguish the two terms: the population’s size. This study analyzes the case of Vietnam, so the author uses the definition of the metropolis in this country. Following that, the metropolis is an area with a population of more than 1,500,000 and the neighboring areas of the metropolises have a smaller population, so they will be considered provinces or rural areas. Thus this study develops the following hypotheses.

2.3.1. Migration and remittances

Migration and remittances always exist together; however, their directions are opposite. The out-migration from provinces to metropolises will lead the remittances to move from metropolises to provinces where migrants’ household lives (Gray, 2009; Mobrand, 2012). Besides that, labor migration between provinces and metropolises has become the core impetus for provincial change, especially agricultural transformation in traditional farming areas (Caulfield et al., 2019; Ge et al., 2020). On the one hand, migration affects structural transformation in provinces due to losing labor in agricultural activities, which leads to agricultural productivity declines (Hussain et al., 2018; Shi, 2018; Taylor & Castelhano, 2016). On the other hand, remittances of migrants increase the household’s income in provinces (SamaratunD. L. Nguyen et al., 2017; Samaratunge et al., 2020). This financial resource supports agricultural productivity increase by using more chemical fertilizers and pesticides, applying new technologies (Caulfield et al., 2019), and leading to labor value reductions, thus removing labor from agriculture (Bhandari & Ghimire, 2016). Although remittance expenditure depends on the household’s characteristics (Adams & Cuecuecha, 2013; Garip, 2014), almost all remittances to provinces are used for food, clothing, health care, and education with inelastic demand for food and elastic demand for the rest of expenditure. Along with the increase in non-agricultural product demand, the opportunity for local industries and services is expanded (D. L. Nguyen et al., 2017). In practice, the factors shaping migration also shape remittances’ distribution and potential impacts. This leads to studying separated remittances or migration impacts so complicated (Taylor & Castelhano, 2016). Therefore, this study estimates the overall effects of migration and remittance instead of separating each effect by using the hypotheses based on migration reasons because there can be no migrant remittances without migration. In total impacts of migration and remittances, the previous studies show that loss in labor in agriculture is compensated by increased access to money (McCarthy et al., 2006; Taylor & Castelhano, 2016). Therefore, any factors that promote migration are proposed to affect structural transformation in provinces. As for the cause of provinces-metropolises migration, these motivations have often been analyzed under the push-pull theory (L. D. Nguyen et al., 2015; Ge et al., 2020). From a metropolis’s perspective, the social and economic opportunities in which job opportunities and higher income are the principal (Becic et al., 2019; Hoffmann et al., 2019; Otterstrom et al., 2021; Pitoski et al., 2021; Yu et al., 2019). Based on the above arguments, I propose the following hypotheses:

Hypotheses of metropolises’ monthly average income per capita

H1: Metropolises’ monthly average income per capita has a positive and significant influence on the structural transformation in provinces

Hypotheses of the underemployment rate in metropolises

H2: The underemployment rate in metropolises has a positive and significant influence on the structural transformation in provinces

2.3.2. Land-use expansion

As a metropolises expansion, their demand for surrounding provincial land grows, creating pressure on land prices (Diao et al., 2019) which leads to an increase in the opportunity costs of engaging in agriculture (Cali & Menon, 2013), thus motivating households to enter the rural non-farm sector (Cobbinah et al., 2015). The demand for metropolises’ land usage includes housing, manufacturing, and other various purposes (Aguilar et al., 2003); however, lots of research shows that provinces-metropolises migration effect farmland usage in provinces (Caulfield et al., 2019; Ge et al., 2020; Qin & Liao, 2016) so using the factors related to housing purpose could lead to the endogeneity with migration.

Thus, the hypothesis of land-use expansion effects is proposed as the following:

H3: The land used for non-agricultural production and business ratio in metropolises has a positive and significant influence on the structural transformation in provinces.

2.3.3. The market for goods and services

Metropolises which are characterized by a large population, play the role of the market for goods and services that comes from neighboring provinces (Otsuka, 2007). The demand for provincial production not only raises the incomes of the local household, which leads to the development of the provincial non-farm economy (Christiaensen & Todo, 2014; Haggblade et al., 1989) but also creates more opportunities for processed foods (Reardon et al., 2016).

And, the hypothesis of market effects is proposed as the following:

H4: Metropolises’ population has a positive and significant influence on the structural transformation in provinces.

2.3.4. Information and knowledge spillover

Metropolises are information centers (Wattenbach et al., 2005), which supply weather updates, price fluctuations, consumer preferences, knowledge, and technology information. This information helps farmers improve their productivity, yields, and profitability by managing risk from the market and restricting harm from natural hazards (Ajani, 2014; Ajani & Agwu, 2012). Increased access to information on different places has an important role in moving out of farming in favor of more modern types of employment in services and manufacturing (Tacoli, 2003).

Hence, the hypothesis of information effects is proposed as the following:

H5: The internet subscribers to total population ratio in metropolises have a positive and significant influence on the structural transformation in provinces.

2.3.5. Sectoral linkages

The manufacturing industry is one of the non-agriculture sectors, and it is also typical of urban areas (Monarca et al., 2019), while this industry has very high linkages with non-agriculture sectors in other localities (Chifamba & Odhiambo, 2015; Kaur, 2020). Therefore, when the manufacturing industry in metropolises develops, it will create connections and promote non-agricultural sectors in neighboring provinces to develop and expand, leading to structural transformation.

Therefore, the hypothesis of sectoral linkages is developed:

H6: The gross product of the manufacturing industry in metropolises has a positive and significant influence on the structural transformation in provinces.

Based on existing literature and empirical evidence, my research framework assumes that the roles of the metropolises on structural transformation in provinces are determined by the effects of metropolises (Figure 1).

Figure 1. Conceptual framework and hypotheses.

3. Methodologies and Data

3.1. Methodology

3.1.1. Global Moran I statistic

Spatial autocorrelation is defined as the correlation of a variable in one area with itself in nearby territories, and it reflects the degree of dependency between the variables. The spatial autocorrelation might be either positive or negative. It will be positive when the same values of the variable appear geographically together and negative when different values of the variable appear geographically together. Moran’s I is a well-known and commonly used to measure and test spatial autocorrelation (Getis, 2008). Moran (1950) proposed this index, which was derived as follows:

(1) $\mathrm{I}=\frac{n{\sum }_{i=1}^n{\sum }_{j=1}^n\left[w_{\mathit{\text{ij}}}\left(Y_i-\bar{Y}\right)\left(Y_j-\bar{Y}\right)\right]}{\left({\sum }_{i=1}^n{\sum }_{j=1}^n{w}_{\mathit{\text{ij}}}\right){\sum }_{i=1}^n{\left(Y_i-\bar{Y}\right)}^2}$(1)

Where: n denotes the number of provinces; $Y_i$, $Y_j$ is the non-agriculture sector shares at province i and j, respectively; $\bar{Y}$ is the mean value of the non-agriculture sector shares; $w_{\mathit{\text{ij}}}$ is a weight index of province i relative to j in spatial weights matrix. Spatial weights matrix W = ($w_{\mathit{\text{ij}}}$: i,j = 1, … n) describe spatial relations between n provinces. According to normal practice, self-influence was ruled out by assuming that $w_{\mathit{\text{ij}}}$= 0 (when i = j) for all i = 1,.,n (so spatial weights matrix W has a zero diagonal). Moran’s Index values range from +1 to −1, with +1 corresponding to strong positive spatial autocorrelation or clustering and −1 corresponding to high negative spatial autocorrelation or dispersion, and 0 corresponding to no spatial autocorrelation.

3.1.2. Spatial econometric models

When spatial autocorrelation occurs, neglecting spatial dependency and spatial heterogeneity might lead to bias in regression analysis due to breaches of common assumptions (Anselin, 1988). In addition, one of the main advantages of spatial regression models is their ability to quantify spatial spillover effects, which are difficult for other models to capture (LeSage & Pace, 2009). Endogenous interaction effects, exogenous interaction effects, and correlated effects are three forms of interaction effects that might explain the dependency of an observation in one specific location on observations in another location (Manski, 1993). The Manski model looks like this:

(2) $\mathrm{Y}=\lambda \mathrm{W}\mathrm{Y}+\alpha {\tau }_{\mathrm{N}}+\mathrm{X}\beta +\mathrm{W}\mathrm{X}\delta +\mu$(2)

$\mu =\rho \mathrm{W}\mu +\epsilon$

$\epsilon {\hspace{0.17em}}_{\sim \hspace{0.17em}}^{\mathrm{i}.\mathrm{i}.\mathrm{d}}\mathrm{N}\left(0,{\sigma }^2{\mathrm{I}}_{\mathrm{n}}\right)$

Where: $Y$ is a (N x 1) vector of non-agriculture sector share of provinces; $X$ is a (N x k) matrix of characteristics of metropolises; $\mu$ is an (N x 1) vector of the error term; reflects the endogenous interaction effects; $\mathit{\text{WX}}$ reflects the exogenous interaction effects; $W\mu$ reflects the correlated effects; $\lambda$ is a spatial autoregressive coefficient; $\rho$ is a spatial autocorrelation coefficient; ${\iota }_N$ is an associated (N x 1) vector with the constant term parameter $\alpha$; while $\beta$ and $\delta$ are associated (k x 1) vectors with unknown parameter. Various models are obtained by imposing constraints on one or more parameters in this model (Kelejian & Prucha, 2010). The characteristics of metropolises are used as explanatory variables in this research to examine the effects of metropolises on provincial structure transformation. At the same time, some provinces have impacts from the same metropolis; hence, there will not be exogenous interaction effects ($\mathit{\text{WX}}$) in this study. Our models are used for analysis include SAC, SAR, SEM. The model with both endogenous interaction effects ($\mathit{\text{WY}}$) and correlated effects ($W\mu$) is the spatial autoregressive confused model (SAC; Kelejian & Prucha, 2010). The model with only endogenous interaction effects is the spatial autoregressive model (SAR). And the model with only correlated effects is the spatial error model (SEM). The statistical approaches for determining the optimum regression model relied on the Hausman specification and Lagrange multiplier tests. In order to compare the random effects to the fixed effects, Hausman specification tests are used. On the other hand, Lagrange multiplier (LM) tests are used to test the absence of each spatial component without having to estimate the unconstrained model.

3.2. Data

Spatial data: This data includes geographical information on the localities of Vietnam extracted from The global administrative area database (GADM). This data will provide information about the area and location of the localities. This information is used to define the Spatial weights matrix in spatial autocorrelation and spatial regression analysis of panel data.

Socio-economic data: This data will provide information on the characteristics of the metropolises and the proportion of non-agricultural in provinces. These data are taken from the provincial statistical yearbooks with 252 observations of all provinces and metropolises in Vietnam during the four years from 2015 to 2018. In there: Non-agricultural sector share: The ratio of total value-added of industries and services to all economic activities’ total value-added. Metropolis per capita income is an indicator of excess wealth that can be remitted: Average monthly income per capita in the metropolis. The underemployment rate in the metropolis indicates the utilization of human capital: The ratio of underemployed people to the total number of people working in the metropolis economy. Underemployed people are those who work under 35 hours and are willing and ready to work additional hours in the reference period. They include: (i) want to do an extra job(s) to increase over time; (ii) want to replace one of the current job(s) with another job to be able to work overtime; (iii) want to increase the hours of one of the current jobs or a combination of the three types of desire above. Ready to work additional hours means if there is a chance to work overtime, they are ready to do it immediately. The rate of land area used for non-agricultural production and business is evidence of pressure on peri-metropolitan agricultural land.: The ratio of the land area used for non-agricultural production and business to the total land area of a metropolis. Metropolis population represents to market for provincial goods: The number of people in the metropolis. The rate of internet subscribers, which is an indicator of knowledge transfer to provinces: The ratio of the number of internet subscribers to the total population of the metropolis. An internet subscriber is the registration number entitled to access the internet. Each internet subscriber has an account to access the network provided by an Internet service provider (ISP). Internet subscribers include Indirect Internet subscribers (dial-up), broadband Internet (xDSL) subscribers, and direct Internet subscribers. The gross product of the manufacturing sector is evidence of sectoral linkages with provincial industries: The total value-added of the manufacturing sector.

4. Results

4.1. Metropolis characteristics and provincial economic structure

Table 1 shows descriptive statistics for the variables used in this study. For metropolises’ characteristics, the monthly per capita income of metropolises in Vietnam (Inc) is about VND 4.718 million. However, this value has a large difference between metropolises, with the largest difference up to 3.334 million VND. Metropolises’ underemployment rate (UDEm) ranges from 0.2 to 3.8%, with an average value of 0.95%. The average percentage of land area used for non-agricultural production and business (Land) in metropolises is 3.2%, while the average percentage of internet subscribers (Inter) in metropolises is 36.319%. The average population of metropolises (Pop) is 4.452 million people, of which the population in the largest metropolis is 7.650 million people, 13 times more than the population in the smallest metropolis. The total manufacturing industrial product of the metropolises fluctuates between 11.764 and 181.001 trillion VND, with an average value of 86.069 trillion VND. Regarding the economic structure of the provinces, the share of the non-agricultural sector (SnA) averaged 71.650%; however, there was a large disparity between localities. The proportion of non-agricultural sectors in the economic structure is highest in the localities, up to 93.140%, while the lowest value is only 45.02%.

Table 1. Descriptive statistics of variables

Variables	Description	Mean	SD	Min	Max
Independent variables
Inc	Average monthly income per capita in the metropolis (million VND)	4.718	0.984	2.843	6.177
UDEm	The underemployment rate in the metropolis (%)	0.950	0.917	0.200	3.800
Land	The rate of land area used for non-agricultural production and business (%)	3.200	1.220	0.878	5.262
Inter	The rate of internet subscribers (%)	36.319	22.054	9.055	67.813
Pop	Metropolis population (million people)	4.452	2.493	0.569	7.650
GPMan	The gross product of the manufacturing sector (trillion VND)	86.069	57.586	11.764	181.001
Dependent variable
SnA	Non-agricultural sector share (%)	71.650	10.177	45.020	93.140

Source: Authors’ calculation using data from GSO (2015–2018).

From a spatial perspective, changes in the spatial distribution of the value-added share in non-agriculture in Vietnam can be seen visually in Figure 2. The quantile map shows that the distribution is unevenly distributed for all 63 provinces. Almost all areas that have over 70% of non-agriculture share in the economic structure are located in the North of Vietnam, while regions in the South have a lower non-agriculture percentage. Incredibly, there are two areas in which the provinces with the highest non-agriculture share are located, and they are areas around Ho Chi Minh and Ha Noi City.

Figure 2. (a) Geographic locations of the study areas and metropolises (left panel). (b) Quantile map of non-agriculture share in study areas of Vietnam 2018 (right panel).

Through the Moran Scatter plot (Figure 3), each point on the scatter plot represents each observation’s value compared with the value of neighboring observations. The horizontal axis of the scatter graph presents the share of the non-agriculture sector of each province, and the vertical axis shows the weighted average (averaged values received in neighboring provinces). The slope of a least-squares regression line through the points is the value of Moran’s I. It can be seen that most of the observations are distributed in the upper right and the bottom left (states that the value for the non-agriculture sector share in the area and the value of its neighborhoods are similar—positive spatial autocorrelation). Therefore, the slope of regression lines is an upward slope. In other words, the almost non-agriculture share of provinces in Vietnam is clustering. Areas located near each other have a similar value to the non-agriculture share in economic structure.

Figure 3. Moran Scatter plot of non-agriculture share in provinces of Vietnam 2018.

Statistically, Moran’s I index in Table 2 proves that the non-agriculture share in the economic structure of provinces is not independently distributed over space. The values of this index fluctuate in a small range, with the lowest value of 0.37 and the highest value of 0.39 in the period from 2015 to 2018, indicating a strong positive spatial autocorrelation and the clustering phenomenon. Besides that, Moran’s Test results show that the p-value is always smaller than 0.05 in 4 years, which means the hypothesis $H_0$ (no spatial autocorrelation) can be rejected with significance.

Table 2. Moran’s test of the spatial autocorrelation of non-agriculture share in provinces of Vietnam

Year	2015	2016	2017	2018
Moran’s I	0.3919	0.3734	0.3914	0.3850
Standard deviate	4.6862	4.4698	4.6795	4.6042
Variance	0.0076	0.0076	0.0076	0.0076
P-value	1.39E-06	3.91E-06	1.44E-06	2.07E-06

4.2. Impacts of metropolises characteristics on provincial structural transformation

Based on the above results of Moran’ I statistics, there is the existence of spatial clustering in non-agriculture share in economic structure in Vietnam. Therefore, spatial economic models are employed in analyzing the relationship between provinces and metropolises. Besides that, to control better for individual heterogeneity or change over time but not across entities, this study collects panel data and analyzes the effects of metropolises and provinces.

The results of Lagrange multiplier tests (LM) in Table 3 show that it is impossible to conclude that the appropriate spatial effect will be used in the model when both LM-Lag and LM-Error tests give P-value < 0.05. Besides, the results of Robust Lagrange multiplier tests (RLM) show that both RLM-Lag and RLM-Error tests are not significant at 5% level even though the RLM-Lag test has P-value = 0.5608 is higher than the P-value = 0.0814 of the RLM-Error test. From here, it is possible to eliminate the case that there is a combination of both spatial effects in the model or the SAC model. However, it is not possible to choose between SAR and SEM models, so both SAR and SEM models are considered.

Table 3. Lagrange multiplier and Robust tests

Test	Null hypothesis	Value of Test	P-Value
Lagrange multiplier tests (LM)
LM-Lag	No spatial autocorrelation (λ = 0) assuming no random effects (${\sigma }_{\mu }^2$=0)	11.944	0.0005
LM-Error	No random effects (${\sigma }_{\mu }^2$=0) assuming no spatial correlation (λ = 0)	14.642	0.0001
Robust Lagrange multiplier tests (RLM)
RLM-Lag	No spatial autocorrelation (λ = 0) assuming the possible existence of random effects (${\sigma }_{\mu }^2$≥0)	0.3383	0.5608
RLM-Error	No random effects (${\sigma }_{\mu }^2$=0) assuming the possible existence of spatial autocorrelation (λ ≥ 0)	3.0366	0.0814

The robust Hausman test was used to compare random effects and fixed effects models for spatial panel data (Arbia, 2014; Elhorst, 2014). The test results in Table 4 show that the strong Hausman test for both the spatial lag of the dependent variable and the spatial lag error has a p-value < 0.01. This result leads to rejecting the null hypothesis about the absence of correlation between individual effects and explanatory variables (significant at the 0.01 level). Therefore, fixed effects will be considered in the selected spatial econometric models.

Table 4. Spatial panel data model

Variables	Spatial autoregressive model (SAR)		Spatial error model (SEM)
	Fixed effects	Random effects	Fixed effects	Random effects
Constant		34.9975*** (10.3858)		56.8755*** (15.6734)
Inc	0.6836*** (3.5314)	0.7056*** (3.9843)	0.9854*** (4.4822)	1.1138 (4.8709)
UDEm	−0.4412* (−1.8330)	−0.3658 (−1.3559)	−0.4691* (−1.7702)	−0.3164 (−1.0572)
Land	2.3137 (0.6653)	1.3088 (1.2735)	3.5912 (1.0261)	1.1981 (1.1032)
Inter	0.0725 (1.3041)	0.0850** (2.0981)	0.1090 (1.5980)	0.1254*** (2.6605)
Pop	0.4420** (2.0354)	0.3558* (1.7593)	0.5555** (2.1579)	0.3695 (1.4749)
GPMan	0.0158 (0.7727)	0.0027 (0.1841)	0.0140 (0.6263)	−0.0025 (−0.1459)
Spatial lag for the dep. variable (λ) Lambda	0.2946*** (3.9454)	0.3442*** (4.7738)
Spatial lag for error components (ρ) Rho			0.2765*** (3.5568)	0.3417*** (3.5832)
Robust Hausman Test	43.985***		89.844***
Log-Likelihood	−411.2406	−608.0976	−752.0547	−612.5632
AIC	964.4812		1646.1090
BIC	1215.0710		1896.6990

Notes: Standard deviations are shown in brackets. Significance: *p < 0.1, ** p < 0.05, *** p < 0.01.

The Akaike criterion of specifications (AIC) is used for model selection. With these criteria, the model with the lowest AIC value is selected, which is biased toward the SAR model (AIC = 964.4812). In other words, the most suitable model to simulate the structural transformation at the provincial level in Vietnam is the endogenous interaction effect (SAR) model.

Based on the results from Table 4, it can be seen that the variables Inc, UDEm, and Pop are all statistically significant. Specifically, variables Inc and Pop positively impact the proportion of non-agricultural sectors in the economic structure, while the variable UDEm has a negative impact. This helps to strengthen hypotheses H1, H2, and H4. In contrast, the variables Land, Inter, and GPMan are not statistically significant at the 5% level, so hypotheses H3, H5, and H6 are not supported.

In the SAR model estimation, the change of one explanatory variable not only has a direct impact on a defined region but may also indirectly affect other regions. Table 5 provides estimation results of direct, indirect, and total impacts according to the SAR model based on the recommendations of LeSage and Pace (2009). Based on the results of Table 5, the direct, indirect, and overall effects of the income variable (Inc) are positive and significant. This shows that the increase in income in metropolises affects the non-agricultural share of a locality and spreads to neighboring localities. However, indirect effects have a smaller value than direct effects. Meanwhile, the underemployment rate (UDEm) and metropolis’ population (Pop) have only a direct impact that is significant, while the spillover effect is not statistically significant at 5%.

Table 5. Direct, indirect and total effects (estimation results for SAR model)

	Direct effects		Indirect effects		Total effects
	Coeff	Standard errors	Coeff	Standard errors	Coeff	Standard errors
Inc	0.7000***	0.2072	0.2691**	0.1253	0.9691***	0.3069
UDEm	−0.4518*	0.2593	−0.1737	0.1266	−0.6255*	0.3697
Land	2.3693	3.5184	0.9109	1.4106	3.2802	4.8394
Inter	0.074221	0.058	0.0285	0.0283	0.1028	0.0838
Pop	0.4526**	0.2143	0.174	0.1132	0.6266**	0.3111
GPMan	0.0162	0.0208	0.0062	0.0084	0.0224	0.0286

Notes: Standard deviations are shown in brackets. Significance: *p < 0.1, ** p < 0.05, *** p < 0.01 .

5. Discussion

The share of the non-agricultural sector in the economic structure (SnA) is unevenly distributed among provinces in Vietnam. Most of the high-value SnA is concentrated around metropolises, and the more remote provinces tend to have lower SnA values. This is consistent with the view of Diao et al. (2019) that areas near the city have more opportunities to participate in the non-agricultural sector. However, there is a difference between metropolises. The SnA value in the adjacent provinces of Hanoi and Ho Chi Minh City is very high (over 80%), while this value in the neighboring areas of Da Nang and Can Tho is about 70%. (Figure 2). The difference between metropolises is the cause of this disparity. Although there are five metropolises in Vietnam, Hanoi and Ho Chi Minh City have distinct characteristics that allow them to play an important role in the SnA of neighboring provinces. Basically, the economic size of these two metropolises exceeds the economic size of the others. Among them, the difference in economies of scale is a wide range that includes many aspects and complex interrelationships. Population, monthly per capita income, the underemployment rate, percentage of land used for production and business purposes, percentage of Internet users, and gross manufacturing product are characteristics that play an important role in expressing the economic scale of metropolises.

Based on the spatial autocorrelation analysis through Moran I index, the experimental results in this study provide solid evidence of the spatial dependence of the non-agricultural sector share in localities of Vietnam. Therefore, ignoring the spatial factor in analyzing the distribution of economic activities or the economic structure of localities may lead to inaccurate conclusions.

From the results of the spatial model analysis with panel data, it can be seen that the impact of metropolises on the economic structure of neighboring provinces is most clearly shown through the process of migration and remittance. As a destination for labor from neighboring localities and as a source of funds through remittance (Gray, 2009; Mobrand, 2012), metropolises contribute to raising the income of households in provinces, promoting the development of non-agricultural activities, thereby helping to transform the economic structure. This is consistent with the results of studies by Truong Cong (2021), Ge et al. (2020), D. L. Nguyen et al. (2017). In detail, metropolises are characterized by high per capita income and a place that provides many jobs and diverse job characteristics, thus attracting a large number of workers in different areas who want to increase their income and find suitable work (Hoffmann et al., 2019; Pitoski et al., 2021). After leaving the locality to work in metropolises, workers will send money back to their families in their place of origin. The amount of money received from migrants will increase consumption of non-agricultural goods, thereby promoting non-agricultural activities to develop (D. L. Nguyen et al., 2017). Interestingly, the impact of metropolises’ income characteristics on structural change in localities is spillover. In other words, the increase in income in metropolises not only increases the proportion of non-agricultural industries in one locality but also spreads to other localities. However, the reduction in the metropolises’ underemployment rate only directly impacts the economic structure of specific localities while not showing an indirect effect on neighboring localities.

As a consumption market for products from the neighborhoods, metropolises with the characteristics of a large population and large consumer demand for goods will help to solve the output issues for products as well as raise the income of people in the regions (Christiaensen & Todo, 2014; Haggblade et al., 1989). In addition, the demand for products in metropolises is very diverse, especially the need for processed foods. This promotes the development of non-farm activities in neighboring provinces to integrate into production processes serving metropolises’ markets (Jayne et al., 2018; Reardon, 2015). However, the increase in population in metropolises also directly impacts the rise in the share of non-agricultural sectors in the economic structure of a particular locality while not showing an indirect effect on the economy of neighboring localities.

The need for metropolises’ land expansion has been shown to impact structural change in neighboring localities in previous studies (Cobbinah et al., 2015; Thuo, 2013). These studies use residential land as a proxy for metropolises’ land-expanding effects. This is acceptable when analyzing the single impact of metropolises’ land expansion on neighboring provinces because land use expansion in metropolises contains multiple purposes, including residential development and service production activities are typical. However, when considering the overall impact of metropolises on provincial structure transformation, if residential land is used as an explanatory variable, it is very likely to violate the phenomenon of correlation with another characteristic of the metropolises, population. Therefore, in this study, the proportion of land used for non-agricultural production and business instead of residential land is used to proxy for metropolises’ land expansion. The research results show that the expansion of land use in metropolises may affect the economic structure of neighboring localities; however, this impact may be mainly due to pressure on residential land demand, while there is no evidence that the increase in land use for non-agricultural production and business purposes in metropolises areas will affect the structural transformation in neighboring provinces.

The manufacturing industry is considered to be highly connected and one of the typical industries in metropolises (Kaur, 2020; Monarca et al., 2019). Therefore, the development of this industry in metropolises is expected to promote the development of non-agricultural activities and structural transformation in neighboring provinces. However, the results show that in the case of Vietnam, this impact is not significant. The main reason is that the linkages between industries are still very weak (Cong, 2022; Demont & Rutsaert, 2017). The lack of connectivity between sectors will limit the ability to integrate production activities in the value chain, leading to inefficient production support and the inability to promote non-agricultural activities.

Based on previous studies, it can be seen that the role of the information and knowledge hub of metropolises will make a significant contribution to the process of structural change in neighboring provinces (Mujeri & Mujeri, 2021; Tacoli, 2003). However, the results in this study were not statistically significant even though the effect was positive. This difference comes from adding spatial factors into the model, which previous studies did not mention. In fact, the role of metropolises as information and knowledge centers cannot be denied, but in the case of Vietnam, the contribution of this role may be insignificant when the system of connectivity and communication to transmit information is still limited.

From the above analysis, metropolises play an essential role in the structural change of neighboring provinces through particular functions. Through that, policymakers can effectively develop options to promote the structural transformation of localities. With limited resources, policymakers can focus on some specific characteristics in metropolises instead of investing and developing spread out across all localities. Policies on income and employment opportunities are remarkably suggested when the impact from income is spillover to the whole region. In addition, the development of commodity markets for neighboring localities should also be a concern. With available infrastructure, metropolises need to develop distribution systems in their markets and expand the exploitation of domestic markets as well as international markets. In addition, it is crucial to invest in and upgrade infrastructure and communication systems between provinces and metropolises. This will help people access timely information and knowledge, thereby making adjustments in production and business activities. In particular, it should be noted that any local development policy needs to be placed in regional linkages linked with metropolises to bring into full play the resources.

6. Conclusion

This study analyzes the effects of metropolises’ characteristics on structural transformation in neighboring provinces using a spatial regression model with panel data. Research results show spatial autocorrelation in the proportion of non-agricultural sectors between localities, with a high proportion of non-agricultural sectors in the economic structure concentrated around metropolises. Metropolises contribute to provincial structural change based on specific characteristics. In particular, the role of attracting migrant workers and providing remittances is very clear. In addition, the metropolises’ market also provides opportunities for the development of non-agricultural activities as well as promoting structural transformation in neighboring provinces. In general, this study has provided an extended scope in understanding the structural transformation of geographic regions. Different from existing theories on economic structural transformation, the results of this study indicate that the factors promoting structural change not only appear within the locality but also include impacts from neighboring localities, especially areas with a greater level of development. In addition, this study also adds empirical evidence on the impact of urban on rural areas with the combination of considering the interrelationships between localities, which have not been mentioned in the previous studies that analyzed only pairs of regions together. The results from the study also confirm the necessity of integrating spatial factors in analyzing the change in the structural transformation of specific geographical areas, which has often been overlooked in other studies before. Since this is the first study to analyze the impact of urban characteristics on the economic restructuring of neighboring localities, there are certain limitations. Firstly, this study only focuses on one-way impacts from urban areas. However, these impacts are potentially influenced by the characteristics of neighboring localities; in other words, the influence of urban areas could be analyzed in a two-way. Therefore, if data is available, the investigation of differentials between characteristics in the metropolis and provinces rather than absolute values in the metropolis will provide more robust evidence. Besides, this study uses the adjacency weight matrix in the correlation analysis between localities as well as simplifies the relationship when assuming the provinces will be affected by the nearest metropolises. Therefore, to improve the validity of the research, the proposed suggestion is a deeper examination of the effect of distance on specific metropolises impacts.

Acknowledgements

This research is funded by University of Economics and Law, Vietnam National University Ho Chi Minh City/VNU-HCM.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research is funded by University of Economics and Law, Vietnam National University Ho Chi Minh City/VNU-HCM.

Notes on contributors

Truong Cong Bac

Truong Cong Bac is Ph.D. Candidate in Economics at University of Economics and Law, Vietnam National University Ho Chi Minh City/VNU-HCM. His research interests are in the areas of economics, development economics, regional linkages, and economic policy.

Van Tran

Van Tran is a lecturer in Faculty of Economics at University of Economics and Law, Vietnam National University Ho Chi Minh City/VNU-HCM. His main research interest is in the field of Development Economics, economic growth, and household welfare.

Tran Thanh Long

Tran Thanh Long is a lecturer in Faculty of International Economic Relations at University of Economics and Law, Vietnam National University Ho Chi Minh City/VNU-HCM. His main research interest is in the field of Economic relations, international economic integration, and exporting enterprises.

Notes

1. Data are extracted from https://data.worldbank.org/