Mapping the dynamics of FDI in China: Convergence, divergence, and policy insights within free trade zones and beyond

Abstract

The uneven distribution of FDI contributes to inequality in economic development. Most studies on FDI in China use provincial data and parametric tools of analysis while delivering mixed findings regarding FDI’s distribution and free trade zones (FTZs) attractiveness. We investigate the convergence and transitional dynamics of relative FDI (RFDI) in China using visual tools of the distribution dynamics approach to a panel of 273 prefectural cities from 2003 to 2019. Convergence-divergence patterns are examined from inter- and intra-regional perspectives and FTZs vis-à-vis non-FTZ cities. In the long run, most entities within each region converge towards different RFDI values (from 0.05 to 0.25), far below the national average equal to one. FDI is the most unevenly distributed within the poorest western region. Furthermore, three (two) convergence clubs emerge among the non-FTZ (FTZ) entities around RFDI values of 0.05, 0.6 and 2.7 (2.5 and 10). The results corroborate the New Growth and New Economic Geography theories, indicating an absence of desirable spatial diffusion from affluent ‘core’ cities and FTZs to ‘peripheral’ entities outside the FTZs. Additionally, we provide a ‘policy priority list’ comprising western cities with above (below) average RFDI and the highest propensity to diverge further above (below) the national mean.

KEYWORDS:

• China
• transitional dynamics
• convergence
• foreign direct investment
• free trade zones

JEL CLASSIFICATIONS:

• F21
• F2
• F

1. Introduction

Since China initiated the open-up policy in 1978, the volume of FDI has experienced dramatic growth. According to China's Ministry of Commerce data, the flow of China's FDI jumped from 0.9 billion USD in 1984 to 149.3 billion USD in 2020, ranking first globally. Nevertheless, foreign capital remains unevenly distributed across space and time. For instance, the eastern region has been documented to attract the most foreign capital (Huang and Wei 2016; Zhang 2023). While the twentieth century witnessed increasingly more FDI flowing into the inland (central and western) areas, the gap between the affluent eastern/coastal regions and the poorer, underdeveloped inland regions continued to widen (Huang and Wei 2016). Furthermore, China is characterised by significant intra-regional inequality in terms of FDI (Cheong, Cheng, and Li 2019a; Zhao, Chan, and Chan 2012).

Chinese policymakers have gradually established free trade zones (hereafter FTZs) as a policy instrument designed to attract foreign capital for export industries, inflow of advanced technologies and maximisation of the country’s comparative advantages (Huang and Wei 2016). The first FTZ was established in Shanghai in 2013, but as of January 2024, the number has increased to 21, significantly promoting China’s international trade and inward FDI (Yuan 2021). However, FTZs also contribute to the uneven distribution of FDI by offering favourable tax rates and liberal, market-oriented policies to foreign enterprises such that cities within FTZs attract most foreign investments (Zhao, Chan, and Chan 2012). However, the spatial inequality of FDI contributes to uneven economic development (Liu 2011; Long, Kim, and Dai 2021). Only if these imbalances are effectively addressed will the growth of the Chinese economy be sustained and aligned with the goal of shared prosperity (Cheong, Cheng, and Li 2019a). Otherwise, sustained economic growth across Chinese regions may be hard to continue (Whalley and Xian 2010).

The existing studies mainly examine the influencing factors and effects of FDI, while few studies analyse the relationship between the spatial inequality of FDI and uneven economic development among regions (e.g. Huang and Wei 2016; Zhang 2023). In particular, the transitional dynamics and long-run ergodic distribution of FDI in China remain under-researched, with Cheong, Cheng, and Li (2019a) being the only related study using provincial data. Moreover, the empirical literature on inward foreign investment documents conflicting results regarding patterns in inter-regional distribution (Cheong, Cheng, and Li 2019a; Huang and Wei 2016) and the allure of Chinese FTZs over time (Huang 2018; Huang and Wei 2016; Yuan 2021).

Against this backdrop, helping policymakers achieve the goal of shared prosperity by conducting in-depth research on the distribution and trends of FDI in China is imperative. Consequently, our objective is to analyse the convergence-divergence patterns of FDI’s future distribution, which can help elucidate the dynamics driving economic inequality and deliver valuable policy insights for promoting balanced and steady growth of FDI. Specifically, we employ Quah’s (1993) distribution dynamics approach (hereafter DDA) and unveil the long-run evolution and transitional dynamics of relative FDI per capita (hereafter RFDI)¹ at the prefectural city spatial level from 2003 to 2019.

Our study contributes to the existing literature in the following aspects. First, we adopt a continuous dynamic distribution approach, which, unlike the traditional parametric tools (e.g. σ- and β-convergence, local Getis-Ord G, Local Indicators of Spatial Association statistics, etc.), can forecast FDI’s bi-dimensional distribution in its entirety (Liu et al. 2022; Maasoumi, Racine, and Stengos 2007). Specifically, the employed visual tools of the DDA, the ergodic distribution and the mobility probability plot (hereafter MPP) present the shape and trends of long-run convergence and detailed information about the relative movement of entities within the distribution. Second, unlike Cheong, Cheng, and Li (2019a), we analyse FDI’s distribution across prefectural cities. Compared with the provincial-level analysis, our study has two advantages: (1) captures the cross-city (inter- and intra-regional) FDI convergence-divergence patterns, and (2) overcomes the problem of a small sample that provincial data may encounter. Third, we conduct a novel, in-depth analysis comparing FDI distribution across the prefectural cities within China’s FTZs vis-à-vis the entities outside the FTZs.

This study reveals the following key findings. First, from the perspective of desirable convergence to the national mean, cross-prefectural cities’ FDI convergence has improved slightly over time. Second, most entities within each region converge toward inadequate and varied RFDI levels (0.05 to 0.25), indicating a long-term concentration of FDI in a few cities (outliers), particularly in the western region. Third, intra-regional RFDI convergence is most (least) significant in China's poorest western (richest eastern) region. Additionally, western cities with RFDI values around 0.6 (and 4.3) have a 38% (and 56%) probability of deviating below (above) the national average and thus merit the top places in the so-called policy priority list. Finally, conditional long-run convergence is more significant in non-FTZ than FTZ-located entities, with three (two) convergence clubs emerging among non-FTZ (FTZ) entities around RFDI values of 0.05, 0.6, and 2.7 (2.5 and 10).

The rest of the paper is organised as follows. We review the relevant literature in Section 2. The data preparation procedure is presented in Section 3. Section 4 offers the research method. Section 5 analyses the existing results. Conclusions and policy implications are provided in Section 6.

2. Literature review

The rapid growth of FDI has aroused the interest of researchers in examining the association between foreign investments and the host country's economic development (e.g. Alfaro et al. 2010; Apergis, Lyroudia, and Vamvakidis 2008). The New Growth (hereafter NG) theory holds that sustainable growth can be achieved by accumulating physical capital (Kottaridi and Thomakos 2007). FDI can directly increase the capital stock of the host country and alleviate the lack of funds, promoting economic growth (Tabassum and Ahmed 2014). In addition, FDI can accelerate the speed of economic development in the host country through the spillover effect from employment opportunities, new technologies, know-how, and managerial experience (e.g. Appiah et al. 2023; Ha and Giroud 2015; Hu, Fisher-Vanden, and Su 2020; Perri and Peruffo 2016; Wang and Liu 2017).

Notwithstanding, according to the NG theory, rich regions with abundant factors can maintain a high growth rate and further strengthen their advancement. The opposite is the case for poor areas unless they can escape the ‘poverty trap’ (Kottaridi and Thomakos 2007). The New Economic Geography theory (hereafter NEG) assumes virtuous (vicious) cycles will be created for core (peripheral) regions. It demonstrates that the agglomeration economy is one of the driving forces behind FDI concentration (Huang and Wei 2016). The interactions between foreign investors and firms in the host country (or region) can help the latter operate more effectively through spillover effects. This, in turn, attracts even more foreign entities, thus forming agglomeration zones and creating a virtuous circle (Kottaridi and Thomakos 2007). Therefore, these countries/regions will reap the benefits of FDI, but peripheral areas, less fortunate in attracting foreign investments, will increasingly lag. Summing up, the NEG theory suggests that foreign capital reinforces the spatial clustering of the economy and can be combined with the NG theory (Kottaridi and Thomakos 2007; Zhao and Zhang 2007).

Extant empirical literature offers mixed results regarding the relationship between FDI and economic growth in developing and emerging markets. For instance, Dinh et al. (2019) show that although FDI hurts economic growth in the short run in some developing countries, it stimulates economic growth in the long run. Udi, Bekun, and Adedoyin (2020) demonstrate that foreign capital increases economic expansion in both the short and long run in South Africa. Most recently, Rao et al. (2023) find that FDI positively influences South and Southeast Asia growth. On the contrary, Liu (2011) documents that direct foreign investments decrease China’s economic growth in the long run. More recently, Long, Kim, and Dai (2021) find evidence of a ‘crowding out’ negative effect of FDI on the productivity of Chinese cities from central, western and northeastern regions.

Other researchers focus on the impact of FDI on income equality but yield inconclusive results. Xu et al. (2021) find that foreign capital reduces income inequality in Sub-Saharan African countries. Le et al. (2021) show that FDI increases income inequality in Vietnam. Huang, Sim, and Zhao (2020) document that FDI is linked to increased inequality among the low-income group of countries, whereas decreased inequality within the high-income group.

Some scholars have tried to explain the uneven spatial distribution of FDI in China from the perspective of the location determinants (e.g. Belderbos and Carree 2002; Boermans, Roelfsema, and Zhang 2011). Specifically, regional differences in relative factor endowments, geographical location, and technology are the source of differences in production structures (Huang and Wei 2016). Additionally, comparative advantages such as large market size and abundant cheap labour can account for determinants and dynamic processes of FDI (Zhang 2023). Furthermore, the New Trade theory holds that differences in market size and market access of economies can explain the agglomeration of foreign production in specific environments (Kottaridi and Thomakos 2007). Moreover, according to the NEG theory, the market size of the host country, market access, and the agglomeration effect are the decisive factors affecting FDI flowing into China (e.g. Boermans, Roelfsema, and Zhang 2011; Jiang, Liping, and Sharma 2013).

Other studies suggest that the host country's policies are associated with the distribution of FDI. For instance, Wei, Luo, and Zhou (2010) argue that the differences in foreign investment and regional development policies across regions affect the location choice of FDI. Furthermore, the FTZ, as a policy instrument, reduces restrictions on overseas investment and financing, thereby successfully attracting foreign capital (Yuan 2021). Based on the synthetic control method, Huang (2018) examines the impact of establishing the Shanghai pilot FTZ on FDI and finds a significant positive effect. Furthermore, the author documents that the differences between FDI in FTZ vis-à-vis non-FTZ will continue to increase, hindering China’s development of a new pattern of comprehensive opening-up. On the contrary, Huang and Wei (2016) argue that the FTZs lost their comparative advantage, whereas the localisation and urbanisation agglomeration were the most significant factors attracting the FDI.

Summing up, ample anecdotal and empirical evidence indicates that FDI remains unevenly distributed across and within Chinese regions, which is unfavourable to the desirable balanced economic development (Huang and Wei 2016; Zhang 2023) and may lead to growing social disharmony (Cheong, Cheng, and Li 2019a). However, the existing research mainly focuses on the influencing factors and effects of FDI (Boermans, Roelfsema, and Zhang 2011; Jiang, Liping, and Sharma 2013; Whalley and Xian 2010). The scant empirical literature investigating the uneven spatial distribution of foreign capital delivers mixed findings concerning inter-regional inequality (Cheong, Cheng, and Li 2019a; Huang and Wei 2016) and the attractiveness of Chinese FTZs over time (Huang 2018; Huang and Wei 2016; Yuan 2021). Furthermore, there is insufficient research on the transitional dynamics and long-term distribution of FDI, with only one relevant study conducted at the provincial level (Cheong, Cheng, and Li 2019a). The authors find that convergence of foreign investments toward the national average remains elusive due to substantial cross-regional disparities.

Against this backdrop, we employ three visual tools of DDA to augment the extant knowledge about FDI's distribution and its future evolution in China. Unlike Cheong, Cheng, and Li (2019a), we examine inter- and intra-regional FDI distribution at the prefectural cities level. Furthermore, we are the first to investigate the transitional dynamics and long-run convergence-divergence patterns regarding cities within and outside China’s FTZs. The findings of this paper can be conducive to helping the government formulate relevant policies to achieve equal distribution of FDI across prefectural cities (convergence towards the national average), an essential ingredient of balanced regional, provincial and prefectural development.

3. Data issues

We constructed a panel dataset across 273 prefectural-level cities between 2003 and 2019. The data is compiled from the Provincial Statistical Yearbook of each of China’s provinces and the China City Statistical Yearbook. We exclude cities with missing values for more than five years. The missing values of the first and last years are calculated using the average growth rate of the recent four years, while the missing values of other years are calculated by linear interpolation and extrapolation methods.

Annual FDI per capita is compiled for each city. Subsequently, the national average is computed for every year of the analysis. In the next step, the relative value of FDI per capita (hereafter RFDI) is calculated by dividing the annual FDI value in each city by the national average value in the same year. The process is repeated for all the years such that the original (absolute) data are transformed into relative values. In other words, RFDI measures the intensity of FDI in each city. If the value of RFDI is greater than one, it indicates that the entity has a level of FDI higher than the national average in a given year, whereas a value less than one means a below-average level of FDI.

Furthermore, we separated the entire sample into smaller datasets based on three factors (time, regions, and FTZs). Specifically, distribution dynamics analysis has been conducted separately on several sub-samples to examine the transitional dynamics and long-run evolution of FDI in greater detail and enable comparisons over time and at different spatial levels.

4. Research methods

The nonparametric distribution dynamics approach (hereafter DDA) was first proposed by Quah (1993) and focuses on the evolution of the distribution, which represents inequality. DDA is considered highly effective in examining the changes in the shapes of the distribution over time and the associated underlying trends (e.g. the persistence or convergence clubs) behind the observed changes which could not be detected using traditional econometric methods (Maasoumi, Racine, and Stengos 2007; Wojewodzki et al. 2023). Moreover, econometric forecasting is inherently limited in that it cannot reveal the shape of the underlying distribution. In contrast, the DDA examines the overall shape of the distribution and its evolution over time. While standard econometric approaches focus on the slope parameters in analysing the impacts of driving variables, the DDA gives an avenue for examining the effects of determinants on the shape of the distribution (Colavecchio, Curran, and Funke 2011). Furthermore, due to its resistance to outliers, the DDA method has a significant advantage over conventional econometric analysis. This virtue results from the fact that the calculation of DDA is based on the likelihood of entities transitioning between distinct states, which is dependent on the occurrence of the entities rather than their measured values (Juessen 2009). In sharp contrast, traditional econometric analysis is sensitive to the impact of outliers since it mainly depends on the computation of slope parameters, which may be severely affected by outliers.

The stochastic kernel approach is adopted in this study to examine the transition dynamics of RFDI across prefectural-level Chinese cities. The bivariate kernel estimator employed in this study is in the form presented below. (1) $\hat{f}(x,y)=\frac{1}{n{h}_1{h}_2}\sum _{i=1}^{n}K\left(\frac{x-X_{i,t}}{h_1},\frac{y-X_{i,t+1}}{h_2}\right)$(1) Where K is the normal density function, n represents the total number of observations, X_i,t is an observed value of RFDI at time t, x is the variable of RFDI at time t, X_i,t+1 is the observed value of RFDI at time t+1, y is the variable of RFDI at time t+1. Additionally, factors h₁ and h₂ represent values of bandwidth computed under the procedure developed by Silverman (1986). Subsequently, the process involves a correction procedure for the spareness of the dataset, and a kernel density function is computed based on the transition of the data.²

Based on the kernel density function, the ergodic density (steady-state equilibrium distribution) can be derived in the long run. This, in turn, is combined with the Mobility Probability Plot (hereafter MPP) developed by Cheong and Wu (2018). The MPP is a measure of the probability of the future movement of the entities within the distribution, and its range is from −100 (100% probability of moving downwards in the following period) to 100 (100% probability of moving upwards in the next period). Thus, a positive value of MPP equal to, e.g. 25 indicates that the city has a probability of 25% moving upwards within the distribution in the following year. In contrast, a negative value equal to, e.g. −47 suggests that the city has a probability of 47% to move downwards within the distribution in the next period.

Due to its construct and simplicity, the MPP is considered to be a handy and powerful tool of the DDA and, in recent years, has been applied in various economic research areas, e.g. carbon emissions (Cheong, Wu, and Wu 2016; Wojewodzki et al. 2023), consumption of electricity (Cheong, Li, and Shi 2019b), credit ratings (Lee et al. 2021), and information transparency (Williams, Cheong, and Wojewodzki 2022). Interested readers can refer to Cheong and Wu (2018) for technical details related to the MPP.

5. Results and discussions

5.1. Relative FDI per capita during 2003–2010 versus 2010–2019 periods

The two-dimensional contour maps of the transition probability kernel for RFDI in 273 Chinese prefectural-level cities can be seen in Figure 1. More specifically, panel A (B) shows the contour map based on the 2003–2010 (2010–2019) period. The horizontal and vertical axes represent the RFDI values at time t (t+1). Additionally, the colours correspond to the kernel’s density (or its height), with the red/purple (blue) colour indicating the highest (lowest) density.

Figure 1. Contour maps for RFDI during the 2003–2010 (A) and the 2010–2019 (B) periods.

Note: The horizontal (vertical) axis represents time t (t+1). Source: Authors’ calculation.

In both panels, we can observe one major peak (the highest density) at highly deficient RFDI values around 0.05 to 0.1, i.e. far below the national average equal to one. This suggests that many prefectural Chinese cities clustered around inadequate levels of FDI during both periods. Furthermore, Figure 1 shows contour lines concentrated diagonally along the 45° line in both panels. At the same time, the kernel’s width is moderately dispersed (less so for the 2010–2019 period), suggesting persistence in relative FDI, somewhat more significant in the more recent period. In other words, assuming no changes in transitional dynamics, the convergence to the national average will be painfully slow, taking many years.

Lastly, we can observe that the outermost contour lines reach the value of around 1.2 (2) in panel A (B). This translates into smaller and larger inter-city differentiation in foreign capital attracted during the earlier and more recent periods. The results broadly align with Cheong, Cheng, and Li (2019a) based on the provincial-level RFDI for the 2004–2014 period. However, compared to their investigation, the prefectural city-level spatial divergence and persistence in RFDI are more pronounced. Thus, the results are alarming and require further analysis to reveal whether the observed trend remains in the long run and if the source of growing differentiation is mainly inter-regional, as suggested by Cheong, Cheng, and Li (2019a), intra-regional, as per Huang and Wei (2016) or occur at both spatial levels.

Figure 2 shows the ergodic distributions, i.e. the long-run steady-state equilibrium of RFDI, assuming no changes in transitional dynamics (Cheong and Wu 2018). This display tool allows us to examine whether the findings from Figure 1 persist in the long run. Based on the 2010–2019 period, we can observe that the distribution peak is much higher than that for the earlier period. The results suggest that, from a nationwide perspective, the cross-prefectural city long-run convergence in FDI has become more significant over time, which is a good sign. Such results support Huang and Wei (2016), who examine spatial inequality in FDI across Chinese prefecture cities between 1990 and 2010 and document inter-regional convergence over time.

Figure 2. Ergodic distributions for RFDI during the 2003–2010 and the 2010–2019 periods.

Note: The horizontal axis represents RFDI, while the vertical axis represents the proportion. Source: Authors’ calculation.

Moreover, the peak for the more recent period is located at a somewhat higher RFDI value of 0.1 compared to the peak for the earlier period, situated at an RFDI value of 0.05. This, in turn, signifies a miniscule improvement in the sought-after convergence to the national mean. Conversely, ergodic distribution based on the 2010–2019 period remains heavily right-skewed, indicating a polarisation in the long run; most entities trapped at deficient FDI levels coexist with the outliers attracting foreign capital significantly above the national average. Such findings corroborate prior studies on China (Zhang 2023; Zhao, Chan, and Chan 2012) and other developing countries (e.g. Farole and Winkler 2014; Jordaan 2008). Moreover, the results are congruent with the NG and NEG theories, suggesting spatial FDI inequality between rich ‘core’ and poor ‘peripheral’ entities, with the latter trapped in the vicious cycle of ‘poverty trap’ and deficient FDI (Kottaridi and Thomakos 2007).

Figure 3 shows the mobility probability plots (hereafter MPPs) for the two periods. It is worth reiterating that this tool was not employed by Cheong, Cheng, and Li (2019a) and, as such, delivers a novel contribution to extant literature. For the cities with RFDI values below one, the dashed line (2010–2019) MPP plots above the solid line (2003–2010) MPP. This suggests that the probability of convergence to the mean improved over time for Chinese prefectural cities with below-average foreign investment levels. Moreover, for the range of values between 1.7 and 2.3, the dashed MPP is located below the solid line plot. This, in turn, means that, more recently, the cities with such levels of FDI display a decreased probability of moving upward in the distribution (diverging from the national mean) in years to come. On the contrary, from the convergence to the mean perspective, the performance of the entities with RFDI values from 1 to 1.6 and from 2.4 to 3.4 worsened over time. This suggests that in the coming years, prefectural cities with RFDI values within such a range are more (less) likely to diverge away (converge towards) the average RFDI.

Figure 3. Mobility Probability Plots (MPPs) for RFDI during the 2003–2010 (A) and the 2010–2019 (B) periods.

Note: The horizontal axis represents RFDI, and the vertical axis represents the MPP. Source: Authors’ calculation.

5.2. Relative FDI per capita across China’s economic regions

Empirical evidence indicates significant inter- and intra-regional heterogeneity in FDI at provincial and prefectural spatial levels (Cheong, Cheng, and Li 2019a; Huang and Wei 2016; Zhang 2023; Zhao, Chan, and Chan 2012). Figure 4 presents stochastic kernels for RFDI in 273 prefectural cities located within four major economic regions: western (panel A), central (panel B), northeastern (panel C), and eastern (panel D).

Figure 4. Contour maps for RFDI in four Chinese regions: western (A), central (B), northeastern (C), and eastern (D).

Note: The horizontal (vertical) axis represents time t (t+1). Source: Authors’ calculation.

The outermost contour lines in panel D reach the highest RFDI values of around 4.8, i.e. approximately 12 times larger than in panel A. Such results suggest a prominent disparity in the intra-regional distribution of foreign capital between the poorest western and affluent eastern (coastal) Chinese regions. Furthermore, the observed disproportion is significantly more pronounced than the one documented by Cheong, Cheng, and Li (2019a) at the provincial spatial level. Thus, our findings highlight that the eastern (western) prefectural cities continue to attract significantly unequal volumes of foreign investments. This, in turn, suggests that the trend of decreasing relative gap (convergence) in FDI between the coastal and inland Chinese regions documented by Huang and Wei (2016) has either slowed down or was only transitory.

Furthermore, the transitional dynamics in western and eastern regions differ the most regarding the RFDI values around which many entities cluster. Specifically, in panel A, the peak is located around extremely low RFDI values ranging from 0.01 to 0.05. In contrast, the peak in panel D appears around significantly higher values (0.2 to 0.4). Additionally, in all panels, the observed peaks are located around RFDI values far below those documented by Cheong, Cheng, and Li (2019a), ranging from 0.1 to 1.6. Furthermore, Figure 4 shows only one peak in each panel, i.e. no signs of intra-regional convergence clubs at a prefectural level. This is at odds with Cheong, Cheng, and Li (2019a), who reported convergence clubs emerging in northeastern and eastern regions.

Figure 5 shows that while ergodic distributions in four regions are right-skewed, skewness decreases as we move from the western to the eastern (coastal) region. Moreover, the heights and values of the major peaks suggest that the long-run intra-regional convergence process is the most (least) significant for the entities in the western (eastern) region, followed by those from the central region. The above findings align with the provincial-level results by Cheong, Cheng, and Li (2019a). However, unlike Cheong, Cheng, and Li (2019a), the distributions in Figure 5 are unimodal, with the sole peaks located significantly below the national average RFDI. Specifically, peaks in the western, northeastern, central, and eastern distributions are centred around the RFDI values of 0.05, 0.1, 0.15, and 0.25. Such findings are mixed news from the perspective of desirable convergence to the national average. On a positive note, the lack of intra-regional convergence clubs suggests that most entities are expected to attract similar levels of FDI. However, the RFDI values towards which many prefectural cities are expected to converge are different in each region and highly deficient, i.e. significantly below the national average. This, in turn, highlights that instead of desirable spatial diffusion, FDI will remain concentrated in a few prefectural cities, which supports Zhao, Chan, and Chan (2012) findings of progressive polarisation in FDI.

Figure 5. Ergodic distributions and contour maps for RFDI in four Chinese regions: western, central, northeastern, and eastern.

Note: The horizontal axis represents RFDI, while the vertical axis represents the proportion. Source: Authors’ calculation.

Figure 5 also reveals a somewhat thicker right tail occurring over the western distribution's 4.3–5 range of RFDI values. Such results indicate that compared to the other regions, there will be more outliers in the western region with very high levels (four to five times above the mean) of foreign capital in the long run. Therefore, assuming no change in transitional dynamics, RFDI across the western prefectural cities will be the most unevenly distributed (polarised) in the future. Such results are pessimistic and support previous research on China. For instance, Huang and Wei (2016) show that in the western region, two high-FDI outliers (Chengdu and Chongqing) coexisted with 20 low-FDI clusters. Zhang (2023) finds that, as of 2018, only a few major western cities enjoyed increasing foreign capital inflows.

Overall, Figures 4 and 5 suggest highly uneven intra-regional distributions, with many entities converging towards inadequate (below the national average) RFDI values in the long run across all economic regions. This is bad news from the perspective of desirable convergence towards the national mean. Moreover, the situation in the western region appears to be the grimmest, i.e. the most unevenly distributed foreign investments. Specifically, most prefectural (peripheral) cities will experience a vicious cycle of deficient FDI, while a few core agglomerations will enjoy high FDI levels (four to five times the national average).

On the one hand, the lack of a spillover effect to the other peripheral prefecture cities, which aligns with the NG and NEG theories, might also be ascribed to China’s increasing economic and decision-making centralisation (Zhang 2023) coupled with the growing role of the state in the economy (Verma 2022). On the other hand, the attractiveness of the outliers might be associated with inter-regional fiscal transfers, large market size, an abundance of cheap labour (Boermans, Roelfsema, and Zhang 2011; Zhang 2023) and China’s major development strategies, such as the Great Western Development and the Belt and Road Initiative (Huang and Wei 2016). At the same time, the observed outliers could be the major prefectural cities (e.g. Chengdu and Chongqing) located within the western region’s FTZs offering multifaceted incentives (e.g. favourable tax rates and liberal, market-oriented policies) to foreign enterprises.

The MPP visual tool can help policymakers focus their attention and resources on the most problematic entities by formulating pragmatic prefectural city-specific policies to mitigate the forecast of intra-regional polarisation in FDI. Specifically, the MPP unveils the entities with (1) RFDI values that are the furthest above (and below) the national average and (2) the highest propensity to diverge further from the average in the years to come. Accordingly, Figure 6 shows that the western prefectural cities with RFDI values between 3.7 and 5.1 should be included in the so-called ‘policy priority’ list. In particular, the entities with RFDI values of around 4.3 have the highest probability of moving upward (56%) and in the distribution in the coming years, thus meriting the top place in the policy priority list. The same holds for the entities in the western region with RFDI value below 0.6 and the highest likelihood (38%) of drifting further down in the distribution. Furthermore, in the central (eastern) region, the entities with values above one (2.4) and below 2.6 (3.3) display a tendency (albeit smaller than the western entities) to move upward in the distribution and, as such, should be added to the policy priority list. Overall, Figures 5 and 6 imply in unison that the least affluent western region of China will be the most unequally distributed at a prefectural-city spatial level regarding the future FDI levels.

Figure 6. Mobility Probability Plots (MPPs) for RFDI in four Chinese regions: western, central, northeastern, and eastern.

Note: The horizontal axis represents RFDI, and the vertical axis represents the MPP. Source: Authors’ calculation.

5.3. Relative FDI per capita and the free trade zones (FTZs)

In this section, we split the sample into prefectural cities within China’s FTZs vis-à-vis those outside the FTZ. Most scholars argue that FTZs are highly effective in attracting FDI (Huang 2018; Yuan 2021). However, Huang and Wei (2016) argue that the FTZs have lost their advantage over other areas. We are the first to investigate this topical issue using visual tools of the DDA to reveal transitional dynamics and long-run convergence-divergence patterns in RFDI across FTZ and non-FTZ entities. The analysis also aims to examine whether the FTZ can be associated with the highly uneven (polarised) intra-regional distributions, especially in China's western region, as documented in section 5.2.

Figure 7 shows three striking differences between the transitional dynamics of the RFDI variable in panels A (non-FTZ cities) and B (FTZ cities). First, we can observe that many non-FTZ entities congregate tightly around deficient RFDI values of around 0.05. On the contrary, most FTZ-located prefectural cities cluster over a broader range of much greater values (from one to 3.5). Second, the outermost contour lines in panel A (B) stretch to the RFDI value of 1.5 (15). This highlights a tenfold disparity in the inter-city RFDI variability between these two groups of Chinese prefectural-level cities. Third, the width of the stochastic kernel presented in panel B is significantly larger than that of the kernel depicted in panel A. Such results mean that the city-specific changes in FDI are more significant and less persistent under the FTZ arrangements.

Figure 7. Contour maps for RFDI in the Chinese cities without (A) and within (B) the free trade zone.

Note: The horizontal (vertical) axis represents time t (t+1). Source: Authors’ calculation.

Figure 8 presents the long-run steady-state equilibrium of RFDI across prefectural-level cities outside (panel A) and within China’s FTZs (panel B). We can notice prominent disparities in their main features. Specifically, the non-FTZ and FTZ distributions have three and two peaks, which signals the emergence of three and two convergence clubs in the long run. Additionally, the distribution for non-FTZ cities is much less spread out, with a thin, tall (a height of 1.02) primary peak around an inadequate RFDI value of 0.05. However, the distribution for the FTZ entities is characterised by a substantially shorter major peak (a height of 0.13), centred at an RFDI value of 2.5, i.e. significantly above the national mean. Furthermore, the minor peaks are located around 0.6 and 2.7 (10) in the non-FTZ (FTZ) distribution.

Figure 8. Ergodic distributions and contour maps for RFDI in the Chinese cities without (A) and within (B) the free trade zone.

Note: The horizontal axis represents RFDI, while the vertical axis represents the proportion. Source: Authors’ calculation.

The observed differences suggest that the long-run convergence is, at best, conditional for both groups of entities and much more significant across the cities outside of the FTZs. While most (some) of the non-FTZ (FTZ) entities will congregate around deficient (far above the national average) RFDI values, only a few non-FTZ outliers will attract foreign capital above the national mean. Furthermore, Figures 7 and 8 suggest that the preferential policies offered in the FTZ will remain highly attractive to foreign enterprises, contrary to Huang and Wei (2016). Thus, despite efforts undertaken by local governments of non-FTZ prefectural cities (Zhao, Chan, and Chan 2012), there is no evidence of spatial diffusion of FDI from FTZs to other areas happening anytime soon. Quite the opposite, the long-run distribution of RFDI appears to be polarised and characterised by convergence clubs far from the national average. Such results corroborate Huang (2018), who finds a widening gap between FDI flowing into FTZ vis-à-vis non-FTZ.

Figure 9 shows that the MPP for FTZ-located entities slopes steadily downward and intersects the horizontal axis at a value of 2.5. Such observation corroborates the results gathered from Figures 7 and 8, indicating prefectural cities within the FTZs clustering around those high levels of RFDI. Moreover, over the range of values from 2.5 to 9, the net mobility probabilities oscillate around zero, meaning that FTZ entities with RFDI within this range are unlikely to move down in the distribution in years to follow.

Figure 9. Mobility Probability Plots (MPPs) for RFDI in the Chinese cities with and without the free trade zone.

Note: The horizontal axis represents RFDI, and the vertical axis represents the MPP. Source: Authors’ calculation.

As for the non-FTZ prefectural cities, the MPP intersects with the horizontal axis at RFDI values corresponding to the three peaks observed in Figure 8, i.e. 0.05, 0.6 and 2.7. Beyond that point, the MPP remains visibly below the horizontal axis. In other words, Figure 9 substantiates the results presented in Figure 8 and suggests the emergence of three convergence clubs in the future FDI among the non-FTZ prefectural cities.

6. Conclusion and policy recommendations

Empirical and theoretical literature argues that the uneven distribution of FDI can significantly contribute to regional disparity in terms of economic development. However, extant studies are usually based on provincial data, and with a noticeable exception (Cheong, Cheng, and Li 2019a) employ parametric tools of distribution analysis. Additionally, prior research on Chinese FDI delivers mixed findings regarding inter-regional spatial diffusion and the attractiveness of FTZs (Cheong, Cheng, and Li 2019a; Huang 2018; Huang and Wei 2016).

Given the above backdrop, this paper adopts three visual tools of the DDA to investigate the long-run convergence-divergence patterns and transitional dynamics of RFDI per capita across 273 prefectural-level Chinese cities. To get more insights into the evolution and trends of RFDI, the analyses are conducted for two periods (2003 to 2009 and 2010 to 2019) and at different spatial levels: inter- versus intra-regional and FTZs vis-a-vis non-FTZs entities.

This study offers five major findings. First, from the perspective of desirable convergence to the national mean, cross-prefectural cities’ FDI convergence has improved slightly over time. Second, the intra-regional convergence process in RFDI will be the most (least) significant among entities located in China's poorest western (richest eastern) region. Third, most prefectural cities within each region converge towards inadequate and different RFDI levels (from 0.05 to 0.25). Furthermore, we find that the western cities with RFDI values around 0.6 (and 4.3) have the highest 38% (and 56%) probability of further diverging below (above) the national average FDI and thus merit the highest places in the so-called policy priority list. Such results suggest that in the long run, foreign capital will remain highly concentrated across a few prefectural cities (outliers) and the most unevenly distributed (polarised) within the western region. Finally, the documented long-run convergence is much more significant across non-FTZ than FTZ-located cities, with three (two) convergence clubs emerging among the non-FTZ (FTZ) entities around RFDI values of 0.05, 0.6 and 2.7 (2.5 and 10).

Summing up, the results are congruent with the new growth (NG) and the new economic geography (NEG) theories and imply a lack of spatial diffusion in FDI from rich ‘core’ to poor ‘peripheral’ entities, as well as from the FTZs to other parts of China. Instead, while few outliers enjoy a virtuous cycle of high growth and high FDI, most prefectural cities remain trapped in the vicious cycle of deficient foreign investment and under-development.

From the policy perspective, the following implications may be drawn and encompassed by policymakers pursuing the overarching ‘shared prosperity’ goal through the policies aiming at balanced and stable growth of FDI. First, we advocate the amalgamation of ergodic distribution and MPP tools with existing forecasting tools employed in the governmental analysis of FDI. Second, extant central and provincial-level FDI-oriented policies might have failed to consider the heterogeneities in cities. Thus, city-specific policies should be formulated to alleviate the uneven distribution of FDI, promoting spatial diffusion instead. Third, the top echelons of leadership in China should reevaluate the merits of increasing economic and decision-making centralisation coupled with the growing role of the state in the economy. Furthermore, the central government should cooperate with (and motivate) local governments to reform their FDI policies to prioritise the spatial diffusion to peripheral, non-FTZ-located prefectural cities, especially in the least affluent western region. This could be done by enhancing existing (and creating new) comparative advantages and optimising the employment of local resources to attract foreign enterprises beyond the core and FTZ-located prefectural cities.

A noteworthy limitation of our study is the inability to capture the role of technology in determining the transitional dynamics of FDI. Therefore, further research is required to investigate how technological progress contributes to the distribution of FDI. Moreover, like any other research method, the DDA has shortcomings, especially the restrictive assumption of no changes in transitional dynamics of the studied variable. Given the unprecedented detrimental effects of the COVID-19 pandemic on the Chinese and global economy and the FDI flows, a follow-up study covering the most recent period is warranted.

Acknowledgements

Open Access funding provided by the Qatar National Library.

Disclosure statement

The authors report no competing interests to declare.

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Notes

1 RFDI is a ratio of a city’s annual FDI per capita to the sample’s average FDI in the same year. Given the nationwide coverage of our data, we consider the sample’s average FDI value equivalent to the national average. Due to its construct, the sample average RFDI is equal to one and thus, RFDI values greater (smaller) than one is above (below) the national mean.

2 Please refer to the Appendix for technical details.

Appendix

In this study, a two-step procedure suggested by Silverman (1986) is adopted to consider the potential problem of spareness in the dataset. Assuming that the data is time-invariant and first-order, the future distribution of RFDI can be represented by equation (2). (A1) $f_{t+\tau }(z)={\int }_0^{\mathrm{\infty }}{g}_{\tau }(z|x)f_t(x)\mathrm{dx}$(A1) Where $f_t(x)$ represents the kernel density function of the distribution of RFDI at time t, $f_{t+\tau }(z)$ stands for the τ-period-ahead density function of z conditional on x, $g_{\tau }(z|x)$ is the probability kernel mapping the distribution from time t to t +τ. The ergodic distribution can be derived by repeated use, thereby transforming equation (2) into equation (3) as follows. (A2) $f_{\mathrm{\infty }}(z)={\int }_0^{\mathrm{\infty }}{g}_{\tau }(z|x)f_{\mathrm{\infty }}(x)\mathrm{dx}$(A2)

Where $f_{\mathrm{\infty }}(z)$ is the ergodic density function and the steady-state equilibrium distribution in the long run. It is worth noting that Cheong and Wu (2018) developed a new visual tool, the Mobility Probability Plot (MPP), which enables the detailed investigation of the future movements of the entities within a distribution. The MPP is defined as $p(x)$ which is the net probability of upward movement in RFDI of individual cities within the distribution as captured by equation (4) below. (A3) $p(x)={\int }_x^{\mathrm{\infty }}{g}_{\tau }(z|x)\mathrm{dz}-{\int }_0^x{g}_{\tau }(z|x)\mathrm{dz}$(A3)

Accordingly, the MPP shows the net probability of an increase in the RFDI for the cities within the distribution. It is conveniently denoted in percentages ranging from −100 to 100.