The World Bank Economic Review, 37(3), 2023, 460–478 https://doi.org10.1093/wber/lhad010 Article Domestically “Flying Geese”: Regional Manufacturing Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 Investment Flows within China Jialiang Zhang and Xiaobo Zhang Abstract This paper examines the evolving patterns of bilateral city-to-city manufacturing investment flows from 2000 to 2015 in China, which are aggregated from detailed firm-level investment transactions based on the adminis- trative business registration database. The coastal regions were a more favorable destination for manufacturing investment prior to 2006 despite their higher wage levels. Since then, the trend has reversed, that is, the inland regions have attracted a growing share of manufacturing investment. The pattern is more pronounced for labor- intensive manufacturing industries. The wage gap between coastal and inland cities is the main driver behind the giant “flying geese”—the relocation of manufacturing firms from coastal to inland areas. JEL classification: O18, O53, R11 Keywords: wage gap, flying geese, investment flows, industrial transfer 1. Introduction After joining the World Trade Organization, China has become the largest exporter of labor-intensive manufactured goods in the world. However, the country’s exports are facing the headwinds of rising wages. Real wages have increased almost sevenfold over the past two decades. China’s real wage level in 2015 was about 3.37 times that in India, an economy whose labor force is nearly the same size as China’s.1 Rising labor costs have put the brakes on China’s phenomenal manufacturing export growth, which has slowed significantly compared with the preceding two decades (Autor et al. 2020). In the wake of rising labor costs, Chinese manufacturing firms face several options (Wei, Xie, and Zhang 2017): “Up,” “Out,” and “In.” The first option is Up: that is, upgrading manufacturing production with Jialiang Zhang is an assistant professor at the School of Economics, Central University of Finance and Economics, Beijing, China; her email address is jlzhang@cufe.edu.cn. Xiaobo Zhang (corresponding author) is a chair professor of economics at the Guanghua School of Management, Peking University, Beijing, China, and is a senior research fellow at the International Food Policy Research Institute (IFPRI), Washington, DC, USA; his email address is x.zhang@gsm.pku.edu.cn. The research for this article was financed by China Natural Science Foundation (#71603057, #71874008, #71950011, and #72192844). The authors thank Nina Pavcnik (the editor), two anonymous reviewers, Lixing Li, Qihong Liu, Fudong Zhang, Xiaodong Zhu, and seminar participants at PBC School of Finance, Tsinghua University, and participants at the 19th China Economics Annual Conference in 2019, the 2nd China Development Economist Forum in 2019, and the Chinese Economists Society 2021 Annual Conference for helpful discussions and valuable comments. The authors declare no conflict of interests. 1 China’s average monthly wage in 2015 was US$391 (estimates based on National Bureau of Statistics, National Data of China), while that of India was US$116 (estimates based on Database on Indian Economy, Reserve Bank of India; Wages and Statistics, Ministry of Labour & Employment, Government of India). © The Author(s) 2023. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com The World Bank Economic Review 461 labor-saving technologies, such as automation and industrial robots. China has already become the largest and fastest-growing market for industrial robotics in the world (Cheng et al. 2019). As of September 2020, there were 140,500 units of industrial robots installed in China, accounting for 37 percent of the global market (International Federation of Robotics 2020). The second option is Out: that is, conducting outward foreign direct investment (FDI) in other countries with lower labor costs. This mechanism is essentially the “flying geese” model. The theory postulates Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 that as labor costs rise, labor-intensive industries in developed areas lose their comparative advantage and are more likely to relocate to less-developed regions with lower factor prices (Akamatsu 1962). The phenomenon of labor-intensive industries relocating across economies has been well documented in the literature, beginning with Japan and progressing to the East Asian Tigers (including the Republic of Korea, Singapore, Hong Kong, and Taiwan), and then to mainland China, India, and Vietnam (Puga and Venables 1996; Kojima 2000; Chiang 2008; Kumagai 2008). In the case of China, although the number of firms that conduct outward FDI has been increasing in recent years, as of 2012, only 5,501 of the country’s 283,018 manufacturing firms (or 1.94 percent) in the Annual Survey of Industrial Firms were engaged in outward FDI (Chen, Tian, and Yu 2019). The annual outflow of FDI from China increased from US$87.8 billion in 2012 to its peak of US$196.15 billion in 2016 before showing a declining trend (MOFCOM, various years). Overall, China’s outward FDI stock is relatively low compared with that of developed countries, including the United States and Japan (Shen et al. 2020). The relocation of labor-intensive manufacturing to other developing countries is only slowly getting under way (Hanson 2020). The third option is In: that is, relocating manufacturing factories from coastal to inland regions in China, to take advantage of lower labor and land costs in the hinterlands. Considering the large regional variation in resource endowments and the varying levels of economic development within China, Hanson (2020) speculates that “China may be on the brink of major changes in its spatial distribution of man- ufacturing production.” Although it is a well-documented fact that the flying geese pattern (successive transfer of labor-intensive industries across space) has happened across countries in Asia, evidence of do- mestically flying geese (spatial relocations of manufacturing activities within China) is lacking (Hanson 2020). Have domestically flying geese begun to migrate in China? This paper aims to answer this question by examining firm investment flows across regions. The data of firm investment flows were obtained from the administrative business registration database maintained by the China State Administration for Industry and Commerce (SAIC). Aggregated invest- ments between all the prefecture-level city pairs by year and by industry are computed by tracing the original locations of the legal representatives and shareholders of all registered firms. The aggre- gate data show a clear pattern of flying geese in the manufacturing sector to the hinterlands since the mid-2000s. Next, the paper examines the role of wage differences in driving regional manufacturing investment flows. To operationalize the analysis, firm-level investment flows (from a shareholder in city A to a firm in city B) are aggregated to bilateral investment flows between city pairs by year based on the administrative business registration database. The rich information about bilateral city-to-city investment flows makes it possible to examine how the wage gap across cities directs firm investment flows, and how this pattern changes over time. This study documents a positive link between the lagged wage gap and cross-city investment flows in the manufacturing sector, following the gravity equation framework, which is widely used in the trade literature. A 1 percent increase in the difference in the lagged average wage between the origin city and the destination city is associated with increases of 0.04 percent in the number of cross-city investments and 0.25 percent in the amount of investment. However, the lagged wage gap may be time persistent and be correlated with unobserved conditions, such as the difference in time-varying business environments, which may also matter to investment flow. To address this potential endogeneity problem, the difference in lagged minimum wages between cities is 462 Zhang and Zhang employed as an instrument for the difference in lagged average city wages. The minimum wage is external to cities because it is determined by the upper province-level government instead of by the city-level gov- ernment. As expected, the difference in lagged minimum wages is shown to be uncorrelated with underly- ing economic links at the city-pair level. The instrumental variable (IV) estimates support the positive link between the wage gap and cross-city investment flows in the manufacturing sector, with the pattern being more pronounced for labor-intensive manufacturing firms. Poisson pseudo-maximum likelihood (PPML) Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 estimation (Santos Silva and Tenreyro 2006) is also applied to address heteroskedasticity problems asso- ciated with a large number of zero values. The results are similar to those based on the IV approach. This paper directly contributes to the “flying geese” literature by demonstrating the existence of do- mestically “flying geese” in China. There is limited empirical evidence on the relocations of labor-intensive industries within a country, except for Qu, Cai, and Zhang (2013) and Ruan and Zhang (2014). Qu, Cai, and Zhang (2013), who document that there was a trend of increasing industrial concentration in the eastern coastal areas up to 2003, before leveling off. Their sample covers only large manufacturing firms (encompassing all state-owned enterprises and private businesses with yearly sales over 5 million RMB). The sample of this paper includes all the registered manufacturing firms regardless of size over a longer period. Additionally, this paper directly examines the effect of wage gaps on firm investment flows across regions, showing that the effect has become larger since the mid-2000s. This paper’s finding can help explain the weakening agglomeration effect in the coastal region since 2004 as identified in Qu, Cai, and Zhang (2013). Ruan and Zhang (2014) zoom in on domestically flying geese in the textile and apparel industry based on aggregated data at the provincial level. Using firm-level administrative data, the present paper goes one step further, confirming that the pattern of flying geese has occurred in China for the manufacturing sector as a whole and for labor-intensive industries in particular. Thanks to the more disaggregate data, the present paper provides more solid empirical evidence on the effect of regional wage differences on the relocation of manufacturing firms than Ruan and Zhang (2014). The present paper is also related to the emerging body of literature on the impacts of China’s rising la- bor costs, including firm-level adjustments (Mayneris, Poncet, and Zhang 2018), firm export performance (Gan, Hernandez, and Ma 2016), and outward FDI from China (Fan, Lin, and Tang 2018). These studies focus exclusively on the impact on the performance of above-scale manufacturing firms in China, largely due to data constraints. The vast numbers of small and medium-size enterprises are not considered. This paper fills the gap by including firms of all sizes and focusing on the impact on regional investment flows. In addition, this paper is associated with the literature on industrial relocation in relation to agglomera- tion (Wen 2004; Long and Zhang 2012) and environmental regulations in developing countries (Eskeland and Harrison 2003; Dean, Lovely, and Wang 2009). This work has some policy implications. According to the theory of “flying geese,” rising labor costs would induce labor-intensive industries in developed areas to relocate to less-developed regions with lower factor prices. Lin (2011) predicts that the next major destination of the “flying geese” will be Africa. The “flying geese” pattern provides a great opportunity for less-developed economies to accommodate the relocated labor-intensive manufacturing industries, which can potentially generate employment and contribute to local economic growth. However, the prediction may not materialize in the near term if domestically “flying geese” occur first on a large scale. China is a large country with vast regional differences in resource endowment and economic devel- opment. China witnessed a rising regional gap between the mid-1980s and the late 1990s (Kanbur and Zhang 2005). In 2016, the GDP per capita of the coastal provinces was almost twice as high as that of the western provinces, and their attracted foreign direct investment (FDI per capita) was as high as nine times.2 Of course, in the face of rising labor costs since the mid-2000s, some firms have considered going 2 In 2016, the average GDP per capita of coastal provinces and western provinces was 81,000 RMB and 43,000 RMB, respectively, while the average attracted foreign direct investment in (FDI) per capita in the two regions was 10,000 US The World Bank Economic Review 463 abroad. Yet many firms in developed regions actually first sought investment opportunities in lagging regions within China. The domestically flying geese have likely served as a potent market force to help reduce the enormous regional gap in economic development since the late 2000s (Kanbur, Wang, and Zhang 2021). The remainder of the paper is organized as follows. Section 2 provides background on the rising labor costs and minimum wage regulation in China. Section 3 describes the construction of cross-city invest- Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 ments. Section 4 presents the empirical strategy. The empirical results are discussed in section 5. Section 6 concludes. 2. Background Rising Labor Costs China’s wage levels remained low for a long time but have risen rapidly over the past two decades. This can be explained by a combination of factors. First, thanks to the family policy beginning in the late 1970s, China has enjoyed more than two decades of the demographic dividend with an increasing share of working-age population. Second, the reform and opening-up policies have boosted investment and entrepreneurship, which in turn have generated hundreds of millions of jobs. As the labor market tightens, wages increase. There is a debate in the literature on whether the Lewis turning point has arrived in China. Lewis (1954) postulates that wages tend to be low when there is surplus labor in the less-productive agriculture sector, and a turning point of rising wages occurs when surplus labor is absorbed by the more productive manufacturing sector. Some studies provide empirical evidence that the Lewis turning point was reached in China in the mid-2000s (Cai and Wang 2010; Cai and Du 2011; Zhang, Yang, and Wang 2011). Others, however, counter that the increase in wages was not caused by a shortage of unskilled labor but rather by institutional barriers that prevent migrant workers from staying in cities (Golley and Meng 2011; Meng 2012). According to this argument, if China removes the institutional barriers, there is still abundant labor available. Putting aside the debate on whether China has reached the Lewis turning point as defined by the exhaustion of surplus labor, the fact of rising real wages since the mid-2000s has been well established. The era of cheap labor in China is over (Li et al. 2012; Hanson 2020). The Minimum Wage System in China Minimum wage regulations were first introduced in China in 1994. China does not have a national min- imum wage, however, because of the large variations in the standards of living across provinces. Instead, each provincial government sets its own minimum wage following the national guidelines. According to the guidelines, several local factors, such as the cost of living, family size, average wages, labor produc- tivity, unemployment, and economic development level, should be considered. In practice, the regulations provide substantial latitude for provincial governments to set their minimum wages (Wang and Gunder- son, 2015). A common practice is that the provincial government divides cities within its jurisdiction into several groups, according to their level of economic development, and then sets the minimum wage for each group of cities. In some provinces, the grouping of cities is fixed over time. For example, in Yun- nan province, cities were divided into three minimum-wage groups, which have never changed. In other provinces, the classification of cities changes over time. For example, the cities in Inner Mongolia province were divided into three groups before 2005 but four groups since 2006. The minimum wage system was reinforced by the Chinese Ministry of Labor and Social Security through new regulations in 2004. This reform aimed to moderate the growing inequality across cities and to improve the standard of living in less-developed areas. The reform introduced several changes. dollars and 1.1,000 US dollars, respectively. Data source: National Bureau of Statistics, National Data, Regional-Annual by Province. https://data.stats.gov.cn/english/easyquery.htm?cn=E0103. 464 Zhang and Zhang First, it strengthened the enforcement of minimum wage regulation. The penalties for violations were in- creased from 20–100 percent of the owed wages to 100–500 percent. And since 2004, the State Council has conducted labor inspections to improve firm compliance. Second, the new regulation requires provin- cial governments to renew the standard minimum wage at least once every two years. Since the reform was implemented, local minimum wages have been adjusted more frequently and have spiked more rapidly. Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 3. Data and Documented Facts Firm Investment Data Set The business registration database maintained by China’s SAIC covers the whole universe of enterprises in China, which included more than 17 million registered firms as of 2015. The mandatory information at the time of registration includes firm location, industry code, ownership type, legal representatives, shareholders, executives, the value of registered capital, and the year of establishment. This article uses the records in the business registration database to measure firm-to-firm investment, that is, a firm serving as a shareholder of another firm through investment. Such firm-to-firm investments are required to be reported to the SAIC within the same calendar year. As a result, the business registration database includes all the records of firm-to-firm investment. The sample for this article is limited to 2000– 2015. There were 1.42 million firm-to-firm investments in the database during the sample period. To study firm relocations, the sample is restricted to investments in which the receiver is a new firm: that is, the investment activity occurs within the calendar year in which the receiver firm was first registered with the SAIC, which leaves 1.30 million firm-to-firm investments. With this restriction, the sample essentially ex- cludes changes in shareholdings of incumbent firms, which are recorded with less precision in the database and are more likely subject to measurement errors. Thereby, this study’s estimate represents a lower bound of regional investment flows. City-Level Average Wage and Minimum Wage In the analysis, the sample is restricted to prefecture-level cities, excluding prefecture-level autonomous regions. The sample ends up with 284 cities. The prefecture-level social-economic variables, such as aver- age wage, gross domestic product (GDP), and population, were obtained from the China City Statistical Year Books from 2000 to 2015. There is no single source of information on the minimum wages enacted by the municipal governments. The study gathered the information via local government websites and statistics bulletins, as well as from local labor and civil reports available on the internet. Ten cities that lacked minimum wage data from 2000 to 2004 are dropped from the IV regressions. Geographic features of a particular prefecture-level city pair are calculated directly from ArcGIS, in- cluding distances between any two cities and whether the two cities share a boundary. First, the longitude and latitude of each city’s geometric center are obtained using spatially referenced shapefiles of Chinese government borders. Second, the spherical distance transformation formula is used to determine the dis- tances between any two cities. Third, ArcGIS’s overlap algorithm is used to identify whether two cities share common borders. Documented Facts This subsection documents the evolving patterns of investment flows across regions in China. First, the direction of investment flows between the coastal and inland regions3 in the manufacturing sector, partic- ularly labor-intensive manufacturing,4 has reversed. As shown in panel A of fig. 1, the share of investments 3 Coastal regions include the following 12 provinces: Beijing, Liaoning, Tianjin, Hebei, Shandong, Jiangsu, Shanghai, Zhe- jiang, Fujian, Guangdong, Guangxi, and Hainan. Inland regions include the other 19 provinces. 4 There is not a unique definition of labor-intensive manufacturing industries. The classification may change over time and vary across countries. Here the study follows Chen, Tian, and Yu (2019) to define the following seven 2-digit industries The World Bank Economic Review 465 Figure 1. Firm Investments in the Manufacturing Sector in Inland Regions Panel A. Share of investments in the manufacturing sector in inland regions. Panel B. Share of investments in the labor-intensive and capital-intensive manufacturing industries in inland regions. Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 Source: Authors’ analysis based on data of firm investments from China’s business registration database administered by SAIC. Note: Panels A displays the share of investments to inland regions from 2000 to 2015 in the manufacturing sector. The solid line in panel A is the share to inland regions by number of investments, while the dotted line stands for the share to inland regions by amount of investments. Panel B displays the share of investments to inland regions in the labor-intensive and capital-intensive manufacturing industries by number of firm investments. The solid line in panel B is the share of investments in labor-intensive industries to inland regions, while the dotted line refers to the share of investments in capital-intensive industries to inland regions. in manufacturing industries to inland regions initially dropped in number and amount before 2002–2003. The tide has shifted to a continuous increase since then. The turning point for labor-intensive manufactur- ing industries was around 2002 to 2004 (panel B), corresponding to the timing of the Lewis turning point documented in the literature (Cai and Wang 2010; Cai and Du 2011; Zhang, Yang, and Wang 2011). The relocation of labor-intensive manufacturing firms from the coastal to inland regions is more pronounced than that of capital-intensive ones. This fact is consistent with the hypothesis that rising labor costs are a major driver for firm relocations. The labor-intensive manufacturing firms in the coastal regions are more sensitive to the rising labor costs than their capital-intensive counterparts. Furthermore, the pathways of the largest investments in manufacturing industries are plotted to track the changing spatial distribution over time. Firm-to-firm investments in the manufacturing sector are aggregated to province-to-province pair-level to provide a clear visualization of the investment routes across regions. Table A1 contains the top 15 province-pair investment flows by amount in 2003 and 2013. As shown in the table, in 2003, among the top 15 province-to-province investments, only 3 destinations belong to inland provinces. By 2013, the number of inland provinces as the destination of the largest 15 province-to-province investments increases to 9. Clearly, in recent years, manufacturing firms have turned to inland regions for investment. 4. Empirical Strategy This study applies the gravity equation framework, which is widely used in the trade literature, to examine the role of factor prices and distance in shaping regional investment flows. Empirical studies based on the gravity equation for international trade find that bilateral trade between two countries is proportional to their respective sizes and inversely proportional to the geographic distance between them (Chaney 2018). Firm portfolio investment can be regarded as a form of capital. The flow of capital between regions is vital for improving the efficiency of spatial capital allocation and promoting the convergence of regional economic growth (Barro, Mankiw, and Sala-i-Martin 1992). Similar to the flow of tradable goods, the as labor-intensive: processing of foods (13), manufacture of foods (14), beverages (15), textiles (17), apparel (18), leather (19), and timber (20). 466 Zhang and Zhang flow of firm investment across regions is expected to follow a gravity pattern. Three distinct estimate approaches are presented in this section: OLS, IV, and PPML estimations. The most commonly used, OLS, might suffer from endogeneity and heteroskedasticity issues (due to the presence of zero values). The IV method addresses the potential problem of endogeneity at the city-pair level by employing differences in lagged minimum wages as an instrument. PPML deals with the heteroskedasticity problem inherent in the dependent variable due to the presence of many zeros. Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 OLS Estimation The empirical model of the regression design based on the gravity equation is specified as follows: yi j,t = α0 + β1 Di f fAverageWagei j,t −1 + α1 Di f fXi j,t −1 + γ Di j + θt + η j + δip + η jp,t + δip,t + i jt (1) where i, j, and t denote investor city, destination city, and year, respectively. The firm-level portfolio in- vestment records in the business registration database are aggregated to the city-pair level in each year. There are two measures of the dependent variable yi jt : Log (number )i jt , which is the number of unique investments from city i to city j in year t, and Log (inv estment )i jt , which is the total investment flow from city i to city j in year t. Dependent variable yi jt is transformed using the inverse hyperbolic sine transformation (Burbidge, Magee, and Robb 1988) log(yi jt + y2 it + 1), to handle observations with zero yi jt values. The main coefficient of interest is β1 , which measures the effect of the wage gap between two cities on cross-city investments, Di f fAverageWagei j,t −1 . It equals the difference in the log average wage between investor city i and destination city j in year t−1. Altogether, the sample consists of 284 prefecture-level cities and 1,205,580 annual city-pair observations from 2001 to 2015. Admittedly, one concern to identification is that the OLS estimates are vulnerable to reverse causality: when capital flows to the destination city, the wages there will go up and close the gap between a city pair. If the wages at the destination city quickly adjust to a higher level, the observed wage gap between the two cities at the current period will be smaller than the initial wage gap that shaped the investment decision. Because the estimated effect should be associated with a larger pre-investment wage difference, the regression on the observed wage gap to the observed investment inflow would inflate the actual effect, causing an upward bias. The lagged wage gap between cities is used to partly mitigate the potential reverse causality problem. Differences in GDP per capita and total population between the investor city and the destination city in year t−1 are included in Xi j,t −1 . Log geographical distances for a given city pair and a dummy variable Di j , indicating whether two cities share borders, are also controlled in the regressions, following the standard gravity equation specification. Time fixed effects, θt , are included to control for the time trend; destination city fixed effects, η j ; and origin province fixed effects, δip , to control for all unobserved and non-time-varying heterogeneity at the destination city level and origin province level in all specifications. Furthermore, origin province × year fixed effects and destination province × year fixed effects as η jp,t and δip,t are included in stricter specifications. One concern is that there might be some unobserved factors at the origin or destination province level, such as local growth potential, which affect both local wages and investment flows. Omitting these factors may bias the results. To address this concern, the sets of origin province × year fixed effects and destination province × year fixed effects are included. These time-varying dummy variables capture most of the unobserved local factors that change over time. IV Strategy Although fixed effects have been included, the difference in lagged wages may still stand for some un- observed factors, such as the difference in time-varying business environment, which may also matter to investment flow. To further mitigate the endogeneity problem at the city-pair level, the lagged difference of minimum wages is used as an instrument for the lagged difference in average wages for a given city The World Bank Economic Review 467 Table 1. Summary Statistics at the City-Pair Level Variable N Mean Std. dev Min Max Panel A: All city pairs Number of investments 1,205,580 0.039 0.515 0 93 Total investment amount 1,205,580 4.851 251.029 0 21,210 (million yuan) Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 Panel B: City pairs where the destination city belongs to coastal regions Number of investments 488,175 0.064 0.756 0 93 Total investment amount 488,175 7.333 338.484 0 21,210 (million yuan) Panel C: City pairs where the destination city belongs to inland regions Number of investments 717,405 0.022 0.238 0 29 Total investment amount 717,405 3.162 167.111 0 12,037 (million yuan) Source: Authors’ analysis based on data from the business registration database administered by the China State Administration for Industry and Commerce. Note: The number of investments refers to the total number of cross-city investments in the manufacturing sector between each city pair in each year, while the total investment amount (in million yuan) is defined as the total amount of cross-city investments in the manufacturing sector between each city pair in each year. The analysis includes 284 prefecture-level cities. Thereby, the total number of city pairs is 284 × 283 = 80,372. The sample covers the period from 2001 to 2015. In panel C, when adding up all city-pair investment, the total number of investments to inland cities is 1,052.19 per year (=0.022 × 47,827 city dyad). Over the whole sample period, the numbers and amounts of investment to inland regions account for 33.6 percent (=(717405 * 0.022)/(1205580 * 0.039)) and 38.8 percent (=(717405 * 3.162)/(1205580 * 4.851)), respectively. pair. Minimum wages are external to cities because they are set by the upper province-level government, rather than by the city-level government (Fan, Lin, and Tang 2018). The difference in minimum wages in city pairs is not directly correlated with their underlying economic links. In the instrumental variable (IV) approach, equation (1) is estimated employing Di f fMimumum Wagei j,t −1 , the difference in minimum wages in city pairs, as instrumental variable for the Di f fAverageWagei j,t −1 . On the other hand, there exist large regional variations in the degree of overlay. The minimum wage not only determines the lower bound of wages in a city, but also shifts the whole wage distribution upward, which even affects firms with wages already above the minimum wage (Berg, 2003). Therefore, minimum wages and average wages are supposedly strongly correlated. Indeed, the correlation coefficient between the two variables is as high as 0.928. Yet, the within-province variation in average wages accounts for 81.021 percent of the total variation, much larger than that of minimum wages (47.935 percent). This is not surprising given that each province sets the minimum wage according to a few groups of cities. Therefore, the difference in lagged minimum wages does not necessarily reflect the difference in local economic conditions between the city pairs that firms take into account when making cross-city investment decisions.5 These features justify using the difference in lagged minimum wages as an instrument for the difference in average wages. 5. Empirical Findings Descriptive Statistics Panel A in table 1 reports the summary statistics for all the city-pair observations. In panels B and C, ob- servations are restricted to whether the destination city is in a coastal region or inland region, respectively. 5 Conditional plots are used to check if the minimum wage differences are correlated with other city-level differences in the residuals of the OLS regressions on other control variables without minimum wages. First, the same regressions are run as in table 3 except for excluding the wage variable. Next, the analysis graphs a nonparametric plot of the residuals from these regressions versus difference in minimum wages, showing that there is not a clear relationship between the difference in minimum wage and other city-level differences in the residuals. To save space, the graphs are not included, but are available upon request. 468 Zhang and Zhang Table 2. Summary Statistics at the City Level Variables N Mean Std. dev Min Max Panel A: All cities Yearly average wage (1,000 yuan) 4,544 26.248 15.977 4.046 113.073 Yearly minimum wage (1,000 4,544 7.868 4.340 2.040 24.360 yuan) Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 Population (million) 4,544 4.283 3.016 0.160 33.752 GDP per capita (1,000 yuan) 4,544 29.497 37.340 1.637 493.052 Panel B: Cities in coastal regions Yearly average wage (1,000 yuan) 1,840 28.459 16.809 5.728 113.073 Yearly minimum wage (1,000 1,840 8.484 4.421 2.160 24.360 yuan) Population (million) 1,840 4.638 2.617 0.475 14.430 GDP per capita (1,000 yuan) 1,840 39.622 48.025 2.396 493.052 Panel C: Cities in inland regions Yearly average wage (1,000 yuan) 2,704 24.723 15.193 4.046 86.358 Yearly minimum wage (1,000 2,704 7.443 4.232 2.040 20.040 yuan) Population (million) 2,704 4.037 3.240 0.160 33.752 GDP per capita (1,000 yuan) 2,704 22.516 25.427 1.637 268.633 Source: Average wage, population, and gross domestic product (GDP) per capita are from the China City Statistical Yearbooks (National Bureau of Statistics, var- ious years). China City Statistical Year books can be accessed through CNKI database: https://navi.cnki.net/knavi/yearbooks/YZGCA/detail?uniplatform=NZKPT. City-level minimum wage data are from local government websites, statistical bulletins, and local reports on labor and civil rights that are available online. “Ren She Tong,” a social security information service platform, maintained by Xi’an Baiyu IT company, have collected the above-mentioned public information online: https://m12333.cn/zuidigongzi/. Note: The sample covers 284 prefecture-level cities from 2000 to 2015. On average, city pairs with destination cities in the coastal area are 3 times more likely to receive intercity investment in the manufacturing sector, and the amount of investment is 2.3 times larger. The sample covers 284 prefectural level cities from 2001 to 2015. Firms in each of the 284 cities in the sample have 283 other cities to choose from. In Panel C, the sample is restricted to city pairs where the destination city belongs to inland regions, which include 47,827 city dyad × 15 year = 717,405 city-pair by year observa- tions. The average incidence of yearly inter-city investment with inland cities as a destination city is 0.022. When adding up all city-pair investments, the total number of investments to inland cities is 1,052.19 per year (=0.022 × 47,827 city dyad). Over the whole sample period, the number and amount of investment to inland regions account for 33.6 percent (=(717405 * 0.022)/(1205580 * 0.039)) and 38.8 percent (=(717405 * 3.162)/(1205580 * 4.851)), respectively, consistent with the pattern shown in fig. 1. Table 2 presents the summary statistics for the key variables at the city level. In general, cities in the coastal areas have higher GDP per capita, average wages, and minimum wages, and larger population than cities in the inland areas. OLS Results The OLS estimates corresponding to equation (1) are reported in table 3. Columns (1) and (2) show the effect of the wage gap on the number of cross-city investments. Column (1) includes year fixed effects, destination city fixed effects, and origin province fixed effects. The more stringent specification in column (2) controls for destination province × year fixed effects and origin province × year fixed effects, which allows for a flexible functional form of the destination and origin province time trend. As reported in column (2), the coefficient for the wage gap variable is 0.040, significant at the 1 percent level. It implies that when the ratio of the average wage in the investor city relative to the destination city increases by 1 percent, the number of cross-city investments to the destination city rises by 0.04 percent. The World Bank Economic Review 469 Table 3. OLS Estimates: The Impact of Wage (lag) Differences on Cross-City Investment (1) (2) (3) (4) VARIABLE Log (number) Log (investment) Diff in lag wage 0.022*** 0.040*** 0.125*** 0.224*** (0.002) (0.002) (0.009) (0.011) Diff in lag GDP pc 0.021*** 0.025*** 0.134*** 0.158*** Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 (0.001) (0.001) (0.004) (0.004) Diff in lag population 0.015*** 0.018*** 0.094*** 0.114*** (0.001) (0.001) (0.003) (0.004) Ln (distance) −0.084*** −0.084*** −0.488*** −0.488*** (0.003) (0.003) (0.015) (0.015) Border 0.133*** 0.133*** 0.707*** 0.707*** (0.012) (0.011) (0.050) (0.050) Observations 1,205,580 1,205,580 1,205,580 1,205,580 Adjusted R-squared 0.151 0.160 0.132 0.140 Year FE Yes Yes Yes Yes Destination City FE Yes Yes Yes Yes Origin Prov FE Yes Yes Yes Yes Destination Province × Year FE No Yes No Yes Origin Province × Year FE No Yes No Yes Source: Firm investment data is obtained from the business registration database administered by the China State Administration for Industry and Commerce. Average wage, population, and gross domestic product (GDP) per capita are from the China City Statistical Yearbooks (National Bureau of Statistics, various years). Note: The table reports estimation results from equation (1). The main dependent variables are Log(number )i jt , which is the number of unique investments from city i to city j within year t; Log(inv estment )i jt , which is the total investment flow from city i to city j in year t. Each of the yi jt dependent variables is transformed using the inverse hyperbolic sine transformation (Burbidge, Magee, and Robb 1988): log(yi jt + y2 it + 1), to handle observations with zero yi jt values. Diff in lag wage is defined as log average wage of investor’s origin city i in year t−1 minus log average wage of destination city j in year t-1. Diff in lag GDP pc equals log GDP per capita of origin city i in year t−1 minus log GDP per capita of destination city j in year t−1. Diff in lag pop is the difference in log total population between city i and city j. Robust standard errors clustered at the city-pair level are reported in parentheses. The 1 percent, 5 percent, and 10 percent statistical significance levels are denoted by asterisks ***, **, and *, respectively. Columns (3) and (4) in table 3 present the effect of the wage gap on the amount of cross-city in- vestments. In the more stringent specification, which includes origin and destination province-year fixed effects in column (4), the coefficient of the wage gap is 0.224, implying that the amount of intercity in- vestment would increase by 0.251 percent6 if the ratio of the average wage in the investor city relative to that in the destination city goes up by 1 percent. Sharing a border promotes investment flows across cities. Based on the coefficient for border, as shown in column (4), the amount of intercity investment increases by 102.79 percent if the two cities share a common border. In addition, the coefficient on distance in column (4) is −0.488, meaning that the amount of intercity investment decreases by 0.629 percent on average as the distance between the two cities increases by 1 percent. In a meta-analysis of 103 papers on the distance coefficient of gravity equations on trade flows, the weighted mean of the effect is −1.07, with 90 percent of the estimates ranging from −0.28 to −1.55 (Disdier and Head 2008). The present study’s estimated coefficient for distance lies in the ballpark of the literature. On the one hand, the significant coefficient suggests that distance still plays an important role in shaping capital flows in China. On the other hand, compared with iceberg-type trade costs induced by the physical transportation costs of goods, the distance elasticity of cross-city investments is smaller. IV Results The first stage of the IV estimation is presented in table 4. The difference in lagged average wages between city pairs and the instrument (the difference in minimum wages between city pairs) is positively correlated with the coefficient, ranging from 0.197 to 0.408. The F-tests rule out the possibility of weak IVs. 6 This study calculates the implied growth rate using exp (β )−1 for all the significant coefficients larger than 0.1. 470 Zhang and Zhang Table 4. First Stage of the Instrumental-Variable Estimate on the Impact of Wage (Lag) Differences on Cross-City Investment (1) (2) VARIABLE Difference in lag wage (log) Diff in lag minimum wage (log) 0.197*** 0.408*** (0.002) (0.004) Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 Diff in lag GDP pc (log) 0.172*** 0.164*** (0.001) (0.001) Diff in lag population (log) 0.025*** 0.026*** (0.001) (0.001) Log (distance) −0.001 −0.002 (0.002) (0.001) Border −0.003 −0.002 (0.003) (0.003) F statistics 270.564 462.981 Observations 1,205,580 1,205,580 Adjusted R-squared 0.725 0.803 Year FE Yes Yes Destination City FE Yes Yes Origin Prov FE Yes Yes Destination Province × Year FE No Yes Origin Province × Year FE No Yes Source: Average wage, population, and gross domestic product (GDP) per capita are from the China City Statistical Yearbooks (National Bureau of Statistics, var- ious years). China City Statistical Year books can be accessed through CNKI database: https://navi.cnki.net/knavi/yearbooks/YZGCA/detail?uniplatform=NZKPT. City-level minimum wage data are from local government websites, statistical bulletins, and local reports on labor and civil rights that are available online. “Ren She Tong,” a social security information service platform, maintained by Xi’an Baiyu IT company, have collected the above-mentioned public information online: https://m12333.cn/zuidigongzi/. Note: The table reports the first stage of the IV estimation. The endogenous variable is the difference in average wages between city i and city j in year t; the instrument is the difference in minimum wage between city i and city j in year t. The analysis includes all the controls as in the baseline specification. Robust standard errors clustered at the city-pair level are reported in parentheses. The 1 percent, 5 percent, and 10 percent statistical significance levels are denoted by asterisks ***, **, and *, respectively. Table 5 reports the second stage of the IV estimation. The dependent variable in columns (1) and (2) is the number of investments across city pairs. The coefficient of the wage variable is larger in the IV estimates than that in the OLS regressions. Taking the coefficient in column (2), if the difference in lagged average wages increases by 1 percent, the number of cross-city investments to the destination city would increase by 0.317 percent. Columns (3) and (4) in table 5 report the effect on the amount of investment. The coefficient for the wage gap variable is 1.636 in column (4). This suggests that the investment flow would rise by 4.135 percent when the relative wage gap widens by 1 percent. The coefficients in table 5 are around seven times the OLS estimates in table 3, suggesting that OLS yields a downward-biased estimation. The coefficients of the geographical factors, that is, distance and border, are similar between the OLS and IV estimations. The analysis uses back-of-the-envelope calculations based on parameter estimation from the IV ap- proach to get the magnitude of the flying geese effect. Given that the sample means of the average wages were 7,758 RMB for inland cities and 10,880 RMB for coastal cities in 2000, the corresponding log of the wage gap in equation (2) is 0.147. In 2003, the sample means of the average wages were 11,031 RMB for inland cities and 16,079 RMB for coastal cities, with the log of the wage gap equaling 0.164. The main variable of interest, Diff in lag wage, increases by 11.56 percent from 2001 to 2004. According to the estimated coefficients in columns (2) and (4) in table 5, the increase associated with the widening wage gap corresponds to a 3.66 percent increase in the number of investments and a 47.80 percent increase in the amount of investment. Given that the sample mean of cross-city investment flow is 4.85 million RMB, The World Bank Economic Review 471 Table 5. Instrumental Variable Estimates of the Impact of Wage (Lag) Differences on Cross-City Investment (1) (2) (3) (4) VARIABLE Log (number) Log (investment) Diff in lag wage 0.093*** 0.275*** 0.526*** 1.636*** IV (lag min wage) (0.006) (0.011) (0.032) (0.055) Diff in lag GDP pc 0.008*** −0.022*** 0.058*** −0.123*** Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 (0.001) (0.002) (0.006) (0.009) Diff in lag population 0.012*** 0.009*** 0.082*** 0.060*** (0.001) (0.001) (0.003) (0.004) Ln (distance) −0.084*** −0.084*** −0.488*** −0.487*** (0.003) (0.003) (0.015) (0.015) Border 0.133*** 0.134*** 0.708*** 0.712*** (0.011) (0.011) (0.050) (0.049) Observations 1,205,580 1,205,580 1,205,580 1,205,580 Adjusted R-squared 0.039 0.006 0.0349 0.002 Year FE Yes Yes Yes Yes Destination City FE Yes Yes Yes Yes Origin Prov FE Yes Yes Yes Yes Destination Province × Year FE No Yes No Yes Origin Province × Year FE No Yes No Yes Source: Firm investment data are obtained from the business registration database administered by the China State Administration for Industry and Commerce. Average wage, population, and gross domestic product (GDP) per capita are from the China City Statistical Yearbooks (National Bureau of Statistics, various years). City-level minimum wage data are from local government websites, statistical bulletins, and local reports on labor and civil rights that are available online. Note: The table reports the second stage of the IV estimation. The endogenous variable is the difference in average wage between city i and city j in year t, and the instrument is the difference in the minimum wage between city i and city j in year t. The main dependent variables are Log(number )i jt , which is the number of unique investments from city i to city j in year t, and Log(inv estment )i jt , which is the total investment flow from city i to city j in year t. Each of the yi jt dependent variables is transformed using the inverse hyperbolic sine transformation (Burbidge, Magee, and Robb 1988): log(yi jt + y2 it + 1) to handle observations with zero yi jt values. Diff in lag wage is defined as the log average wage of investor’s origin city i in year t−1 minus log average wage of destination city j in year t−1. Diff in lag GDP pc equals log GDP per capita of origin city i in year t−1 minus log GDP per capita of destination city j in year t−1. Diff in lag pop is the difference in log total population between city i and city j. Robust standard errors clustered at the city-pair level are reported in parentheses. The 1 percent, 5 percent, and 10 percent statistical significance levels are denoted by asterisks ***, **, and *, respectively. the increase in the wage gap corresponds to 2.32 million RMB of capital flow between any random city pair, or 186.46 billion RMB for all city pairs during 2001–2004. The simple back-of-the-envelope calcu- lations show that the flying geese pattern has large economic significance in shaping cross-city investment flows. PPML Results This study has tried to address the potential endogeneity problem in the OLS regressions using an IV approach. However, the large number of zero values in the highly granular data poses an estimation challenge, as it creates a heteroskedasticity problem, which makes the OLS estimates inconsistent (Santos Silva and Tenreyro 2006). A popular way to address the heteroskedasticity problem in the presence of many zeros is to employ the technique of PPML regression (Santos Silva and Tenreyro 2006). PPML has been widely applied to estimate gravity equations in the international trade literature. In addition, PPML regression provides a natural way to deal with a large number of zero values inherent in the dependent variable, as it does not require a log transformation. Table 6 reports the PPML estimations in which the dependent variables are in levels. The results are broadly unchanged across the different specifications. The coefficients in table 6 are closer to the IV estimates in table 5 than to those in table 3 with the OLS approach. 472 Zhang and Zhang Table 6. Impact of Wage (Lag) Difference on Cross-City Investment (PPML) (1) (2) (3) (4) VARIABLE Number Investment Diff in lag wage 0.609*** 1.676*** 0.440*** 1.473*** (0.079) (0.131) (0.085) (0.132) Diff in lag GDP pc 0.435*** 0.750*** 0.480*** 0.862*** Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 (0.028) (0.063) (0.030) (0.054) Diff in lag population 0.669*** 1.056*** 0.584*** 0.981*** (0.049) (0.048) (0.047) (0.046) Ln (distance) −2.323*** −2.293*** −2.015*** −1.985*** (0.095) (0.090) (0.090) (0.085) Border 0.393*** 0.337*** 0.359*** 0.317*** (0.107) (0.105) (0.104) (0.099) Observations 1,201,335 1,189,480 1,201,335 1,189,480 Year FE Yes Yes Yes Yes Destination City FE Yes Yes Yes Yes Origin Prov FE Yes Yes Yes Yes Destination Province × Year FE No Yes No Yes Origin Province × Year FE No Yes No Yes Source: Firm investment data are obtained from the business registration database administered by the China State Administration for Industry and Commerce. Average wage, population, and gross domestic product (GDP) per capita are from the China City Statistical Yearbooks (National Bureau of Statistics, various years). Note: The table reports Poisson pseudo maximum likelihood (PPML) regression estimation results from equation (1). The main dependent variables are the number of unique investments from city i to city j in year t and the total investment flow from city i to city j in year t. Diff in lag wage is defined as log average wage of investor’s origin city i in year t−1 minus log average wage of destination city j in year t−1. Diff in lag GDP pc equals log GDP per capita of origin city i in year t−1 minus log GDP per capita of destination city j in year t−1. Diff in lag pop is the difference in log total population between city i and city j. Robust standard errors clustered at the city-pair level are reported in parentheses. The 1 percent, 5 percent, and 10 percent statistical significance levels are denoted by asterisks ***, **, and *, respectively. Discussions on the Limitations of the Estimation Methods As indicated in equation (1), in the above three estimation approaches, destination city fixed effects, η j ; and origin province fixed effects, δip , are controlled. Although there is no econometric reason not to include origin-city fixed effects, δi , incorporating them will absorb most variations inherent in the variable of wage differences between the city pairs. Because the objective of this study is to link firm investment behavior with the differences in wages between cities, only origin-province fixed effects are included. Similarly, for a panel data set with a time dimension, one could theoretically include bilateral FE σi j , and identify effects from changes over time at the city-pair level. However, this highly demanding specification would absorb all time-invariant bilateral frictions for each city pair. Moreover, controlling for the pair- wise fixed effects would result in a severe multicollinearity problem with the key independent variable, wage difference, particularly under the PPML specification. In summary, the identifying power primarily stems from the cross-sectional variation at the destination side, that is, to which city firms are locating. Although this article’s discussions on the mechanism driving relocations have focused on wage increases at the originating side of city-pairs, the relatively lower wage levels at the destination cities are equally important for firms to make cross-city investments. Effects over Time The results so far are consistent with the hypothesis that wage gaps played an important role in driving cross-city investment flows from 2001 to 2015. The specifications make a crucial assumption that the role of wage differences is constant over time. Next, this restriction is relaxed by allowing the coefficient of wages to change over time. The analysis first examines the time-varying effects of the wage gap on manufacturing firms’ cross-city investment behavior. The study first uses the data of the entire manufacturing sector before limiting the The World Bank Economic Review 473 Figure 2. Impact of Wage Differences on Investment in the Manufacturing Sector (by Year) Panel A. Log (Number of Investments)Panel B. Log (Amount of Investments) Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 Source: Firm investment data is obtained from the business registration database administered by the China State Administration for Industry and Commerce. Average wage, population, and gross domestic product (GDP) per capita are from the China City Statistical Yearbooks (National Bureau of Statistics, various years). Note: The figure visualizes coefficient β1t for variable Di f f in lag wage × Yeart in equation (2) using PPML estimation, with data of the entire manufacturing sector. Panels A and B refer to the number of investments and the total amount of investments of yearly cross-city investment flows, respectively. The coefficients are presented in dots with their 95 percent confidence intervals. sample to the labor-intensive manufacturing industries. The specification is shown in equation (2). yi j,t = α0 + β1 Di f fAverageWagei j,t −1 × Yeart + α1 Di f fXi j,t −1 + γ Di j + θt + η j + δip + i jt (2) where i denotes origin city, j denotes destination city, and t denotes year. yi jt stands for the log value of the number of investments, or the amount of investment flow from origin city i to destination city j in year t, transformed by log(yi jt + y2 it + 1). The interaction term of Di f f Av erageWagei j,t −1 and the dummy variable for year is the main explanatory variable of interest; Xi j,t −1 controls for the difference in GDP per capita and total population between the investor city and the destination city in year t−1. Di j includes log geographical distance and the border effect. Similar to equation (1), time fixed effects, θt , are included to control for the time trend; destination city fixed effects, η j ; and origin province fixed effects, δip , in all the specifications. Figure 2 shows the PPML estimation results of equation (2) using the data of the entire manufacturing sector. The estimated coefficients of the interaction terms as well as their 95 percent confidence intervals are plotted in fig. 2, as the specification allows for the effects of wage gaps to vary across every year. The outcome variables in panels A and B are the number of investments and the amount of investment. Interestingly, the effect of the wage gap on both the number and amount of investments is initially sig- nificantly negative until 2005–2006, before turning positive in 2007, reaching the highest values in 2012, and leveling off during 2013–2015. The shift in the sign of the wage variable over time is related to the trade-off between agglomeration and “congestion.” In the economic geography and urban economic literature, the spatial distribution of economic activity is driven by two opposing forces: namely, agglomeration and dispersion, under a general equilibrium framework (Krugman 1991; Fujita and Thisse 1996; Puga and Venables 1999; Desmet and Rossi-Hansberg 2009). Agglomeration forces include increasing return to scale in production technology, knowledge spillover, labor pooling, and market linkages, which often result in the formation of industrial clusters. However, when more industrial activities are concentrated in a narrowly defined location, factor prices tend to go up and create congestions, causing a dispersion. Dispersion forces are primarily com- posed of factor prices, such as higher land costs and wages in areas with high industrial concentration. Specifically, Puga and Venables (1996) theorize that industries first concentrate in developed countries and then spread to less developed ones, drawing from the facts that manufacturing industries have relo- cated from Japan to other eastern Asian economies. For reviews on the large amount of theoretical and 474 Zhang and Zhang Figure 3. Impact of Wage Differences on Investment in the Labor-Intensive Manufacturing Sector (by Year) Panel A. Log (Number of Investments) Panel B. Log (Amount of Investments) Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 Source: Firm investment data are obtained from the business registration database administered by the China State Administration for Industry and Commerce. Average wage, population, and gross domestic product (GDP) per capita are from the China City Statistical Yearbooks (National Bureau of Statistics, various years). Note: The figure visualizes coefficient β1t for variable Di f f in lag wage × Yeart in equation (2) using PPML estimation, where the sample is restricted to labor-intensive manufacturing industries. Panels A and B refer to the number of investments and the total amount of investments of yearly cross-city investment flows, respectively. The coefficients are presented in dots with their 95 percent confidence intervals. empirical literature on the source of agglomeration externalities in urban economics, see Duranton and Puga (2004). Prior to 2005–2006, investments in the manufacturing sector mostly went to the coastal areas with relatively higher wages (though still low in absolute levels), probably to take advantage of agglomeration forces and the Open Door Policy, which favor the coastal areas. In the first two decade after China’s opening up, the coastal areas attracted massive foreign direct investment, mostly in the labor-intensive manufacturing sector, which in turn generated huge demand for workers. Accordingly, many industrial clusters were formed in the coastal areas (Long and Zhang 2011). The rapid growth in the manufacturing sector in the eastern coastal regions eventually bid up wages in mid-2010s (Zhang, Yang, and Wang 2011). It is not surprising that as of 2007, manufacturing firms started chasing cheap labor in the less-developed hinterlands, exhibiting a flying geese pattern. It is likely that labor-intensive manufactories responded to the labor shortage in the coastal areas more strongly. To check this out, this article repeats the estimation of equation (2) by zooming in on the subgroup of labor-intensive manufacturing industries. Figure 3 plots the estimated coefficients for the interaction terms for this subgroup together with their 95 percent confidence intervals. The temporal pattern is similar to that in fig. 2, except that the coefficient estimates in fig. 3 are larger in magnitude. The flying geese pattern is more pronounced in labor-intensive manufacturing industries than in the manufacturing sector as a whole. Figure 4 shows another heterogeneity analysis by dividing the sample into two periods, 2001–2006 and 2007–2015. The analysis runs separate regressions for the entire sample of manufacturing industries and the subsample of labor-intensive manufacturing firms. Figure 4 displays the coefficients in bars with their 95 percent confidence intervals, showing the impact of wage difference on cross-city investment in the two periods. Panel A reports the PPML estimates, while panel B presents the IV estimates. As shown in fig. 4, the impacts of the wage difference on cross-city investments in both the manufacturing and labor- intensive manufacturing sectors are mainly driven by the effects in the later period, after China passed the Lewis turning point. As shown in both panel A and panel B (manufacturing 2001–2006), the coeffi- cients are slightly negative and significant in the first period for the entire manufacturing sector, indicat- ing that the agglomeration forces dominated cost-saving motives, as manufacturing clusters formed in the coastal regions. The pattern is consistent with the findings by Wen (2004) and Long and Zhang (2012). By The World Bank Economic Review 475 Figure 4. Impact of Wage (Lag) Differences on Cross-City Investment (by Time Period) Panel A. Coefficients by time period: PPML Panel B. Coefficients by time period: IV Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 Source: Firm investment data is obtained from the business registration database administered by the China State Administration for Industry and Commerce. Average wage, population, and gross domestic product (GDP) per capita are from the China City Statistical Yearbooks (National Bureau of Statistics, various years). City-level minimum wage data are from local government websites, statistical bulletins, and local reports on labor and civil rights that are available online. Note: The figure reports the estimated coefficient for the wage variable in separate regressions for the entire sample of manufacturing industries and the subsample of labor-intensive manufacturing firms in two periods, 2001–2006 and 2007–2015. Panels A and B display the coefficients by PPML estimates and IV estimates, separately. The coefficients are presented in bars with their 95 percent confidence intervals. comparison, the coefficients are insignificant in the first period for the labor-intensive manufacturing sec- tor (labor-intensive 2001–2006), probably because these firms were more sensitive to labor costs even in the early period. In both panels, the coefficients in the later period (2007–2015) are more statistically significant and larger in magnitude, when the cost-saving motive came to dominate. Clearly, firms started to make cross-city investments, mostly in inland areas, by chasing cheap labor. The effect is noticeably larger in labor-intensive industries than those for the manufacturing sector as a whole after 2006. 6. Conclusion The flying geese pattern has been well-documented across countries in East Asia (Kumagai 2008). Yet, relatively little attention has been paid to examine the possibility that this industrial relocation could happen across different regions within a large country. Using the administrative business registration database, this paper shows that the flying geese pattern has occurred in China. The shifting tide happened in the mid-2000s when China reached the Lewis turning point. China is a large country with considerable regional variations in wage levels, creating room for domes- tic flying geese manufacturing firms. Through relocations from the coastal regions to the inland regions, the manufacturing sector still has potential in the near future. Continuous flying geese will likely help China to achieve more balanced regional growth. Studying the pattern of flying geese is instrumental in understanding the processes of structural change and industrial relocation in China as well as their implications. Data Availability Statements The business registration database of firm investments was provided by the China State Administration for Industry and Commerce by permission. Data will be shared on request to the corresponding author with permission of the China State Administration for Industry and Commerce. Average wage, population, and gross domestic product (GDP) per capita are from the China City Statistical Yearbooks (National Bureau of Statistics, various years). China City Statistical 476 Zhang and Zhang Yearbooks can be accessed through CNKI database: https://navi.cnki.net/knavi/yearbooks/YZGCA/detail? uniplatform=NZKPT. City-level minimum wage data are from local government websites, statistical bulletins, and local re- ports on labor and civil rights that are available online. “Ren She Tong,” a social security information service platform, maintained by Xi’an Baiyu IT company, have collected the above-mentioned public in- formation online: https://m12333.cn/zuidigongzi/. Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 Appendix Table A1. Top 15 Investment Flows at the Province-pair Level Panel A. Top 15 Investment flows at the province-pair level in 2003 Origin province Origin region Destination province Destination region Investment amount (billion yuan) Shanghai Coastal Jiangsu Coastal 597.068 Zhejiang Coastal Jiangsu Coastal 272.792 Guangdong Coastal Jiangsu Coastal 270.604 Liaoning Coastal Beijing Coastal 241.408 Beijing Coastal Shanxi Inland 227.317 Fujian Coastal Jiangsu Coastal 207.567 Guangxi Coastal Zhejiang Coastal 200.190 Beijing Coastal Liaoning Coastal 186.778 Beijing Coastal Guangxi Coastal 182.224 Beijing Coastal Guangdong Coastal 178.335 Hunan Inland Henan Inland 172.077 Shandong Coastal Beijing Coastal 169.947 Beijing Coastal Inner Mongolia Inland 166.111 Anhui Inland Guangdong Coastal 165.708 Beijing Coastal Jiangsu Coastal 151.991 Panel B. Top 15 Investment flows at the province-pair level in 2013 Heilongjiang Inland Jilin Inland 4028.251 Beijing Coastal Inner Mongolia Inland 1922.521 Guangdong Coastal Beijing Coastal 1773.476 Beijing Coastal Fujian Coastal 1417.471 Liaoning Coastal Inner Mongolia Inland 1126.088 Beijing Coastal Xinjiang Inland 974.126 Jiangsu Coastal Shanghai Coastal 814.178 Guangdong Coastal Hunan Inland 800.425 Beijing Coastal Guangdong Coastal 735.983 Beijing Coastal Sichuan Inland 729.094 Guangdong Coastal Henan Inland 681.566 Beijing Coastal Jiangxi Inland 664.611 Sichuan Inland Shanghai Coastal 476.935 Jiangsu Coastal Anhui Inland 469.764 Guangdong Coastal Jiangsu Coastal 467.939 Source: Firm investment data is obtained from the business registration database administered by the China State Administration for Industry and Commerce. Note: The table displays the top 15 province-pair investment flow by amount, at the province-pair level, as plotted on map in fig. 2. Panel A shows the top 15 province- pair investment flow in 2003 and panel B in 2013. The firm-to-firm investments in the manufacturing sector are aggregated to province-to-province pair level. Firm investment is obtained from the business registration database administered by the China State Administration for Industry and Commerce. The World Bank Economic Review 477 References Akamatsu, K. 1962. “A Historical Pattern of Economic Growth in Developing Countries.” Developing Economies 1: 3–25. Autor, D., D. Dorn, L.F. Katz, C. Patterson, and J.V. Reenen. 2020. “The Fall of the Labor Share and the Rise of Superstar Firms.” Quarterly Journal of Economics 135 (2): 645–709. Barro, R.J., N.G. Mankiw, and X. Sala-i-Martin. 1992. “Capital Mobility in Neoclassical Models of Growth.” Amer- Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 ican Economic Review 85 (1): 103–15. Berg, G.J.V.D. 2003. “Multiple Equilibria and Minimum Wages in Labor Markets with Informational Frictions and Heterogeneous Production Technologies.” International Economic Review 44 (4): 1337–57. Burbidge, J., L. Magee, and A.L. Robb. 1988. “Alternative Transformations to Handle Extreme Values of the Depen- dent Variable.” Journal of the American Statistical Association 83 (401): 123–7. Cai, F., and Y. Du. 2011. “Wage Increases, Wage Convergence, and the Lewis Turning Point in China.” China Economic Review 22 (4): 601–10. Cai, F., and M. Wang. 2010. “Growth and Structural Changes in Employment in Transition China.” Journal of Com- parative Economics 38 (1): 71–81. Chaney, T. 2018. “The Gravity Equation in International Trade: An Explanation.” Journal of Political Economy 126 (1): 150–77. Chen, C., W. Tian, and M. Yu. 2019. “Outward FDI and Domestic Input Distortions: Evidence from Chinese Firms.” Economic Journal 129 (624): 3025–57. Cheng, H., R. Jia, D. Li, and H. Li. 2019. “The Rise of Robots in China.” Journal of Economic Perspectives 33 (2): 71–88. Chiang, H.H. 2008. “The ‘Flying Geese Development’ Model of the IT Industry in East Asia.” Journal of the Asia Pacific Economy 13 (2): 227–42. Dean, J.M., M.E. Lovely, and H. Wang. 2009. “Are Foreign Investors Attracted to Weak Environmental Regulations? Evaluating the Evidence from China.” Journal of Development Economics 90 (1): 1–13. Desmet, K., and E. Rossi-Hansberg. 2009. “Spatial Growth and Industry Age.” Journal of Economic Theory 144 (6): 2477–502. Disdier, A.-C., and K. Head. 2008. “The Puzzling Persistence of the Distance Effect on Bilateral Trade.” Review of Economics and Statistics 90 (1): 37–48. Duranton, G., and D. Puga. 2004. “Chapter 48 - Micro-Foundations of Urban Agglomeration Economies.” In Hand- book of Regional and Urban Economics. Edited by J. V. Henderson and J-F. Thisse, 2089–2097. Amsterdam: Elsevier. Eskeland, G.S., and A.E. Harrison. 2003. “Moving to Greener Pastures? Multinationals and the Pollution Haven Hypothesis.” Journal of Development Economics 70 (1): 1–23. Fan, H., F. Lin, and L. Tang. 2018. “Minimum Wage and Outward FDI from China.” Journal of Development Eco- nomics, 135: 1–19. Fujita, M., and J.F. Thisse. 1996. “Economies of Agglomeration.” Journal of Japanese and International Economies 10 (4): 339–78. Gan, L., M.A. Hernandez, and S. Ma. 2016. “The Higher Costs of Doing Business in China: Minimum Wages and Firms’ Export Behavior.” Journal of International Economics 100: 81–94. Golley, J., and X. Meng. 2011. “Has China Run Out of Surplus Labour?” China Economic Review 22 (4): 555–72. Hanson, G.H. 2020. “Who Will Fill China’s Shoes? The Global Evolution of Labor-Intensive Manufacturing.” NBER Working Paper No. 28313. National Bureau of Economic Research. Cambridge, MA, USA. International Federation of Robotics. 2020. World Robotics 2020 Report, Germany. https://ifr.org/ifr-press- releases/news/record-2.7-million-robots-work-in-factories-around-the-globe. Kanbur, R., and X. Zhang. 2005. “Fifty Years of Regional Inequality in China: A Journey Through Central Planning, Reform, and Openness.” Review of Development Economics 9 (1): 87–106. Kanbur, R., Y. Wang, and X. Zhang. 2021. “The Great Chinese Inequality Turnaround.” Journal of Comparative Economics 49 (2): 467–82. Kojima, K. 2000. “The ‘Flying Geese’ Model of Asian Economic Development: Origin, Theoretical Extensions, and Regional Policy Implications.” Journal of Asian Economics 11 (4): 375–401. 478 Zhang and Zhang Krugman, P. 1991. “Increasing Returns and Economic Geography.” Journal of Political Economy 99 (3): 483–99. Kumagai, S. 2008. “A Journey Through the Secret History of the Flying Geese.” IDE Discussion Paper 158. USA International Development Enterprises. Lewis, W.A. 1954. “Economic Development with Unlimited Supplies of Labour,” Manchester School 22 (2): 139–91. Li, H., L. Li, B. Wu, and Y. Xiong. 2012. “The End of Cheap Chinese Labor.” Journal of Economic Perspectives 26 (4): 57–74. Downloaded from https://academic.oup.com/wber/article/37/3/460/7077007 by World Bank and IMF user on 14 September 2023 Lin, J.Y. 2011. “From Flying Geese to Leading Dragons: New Opportunities and Strategies for Structural Transfor- mation in Developing Countries.” Policy Research Working Paper No.5702. World Bank. Washington, DC, USA. Long, C., and X. Zhang. 2011. “Cluster-Based Industrialization in China: Financing and Performance.” Journal of International Economics 84 (1): 112–23. ———. 2012. “Patterns of China’s Industrialization: Concentration, Specialization, and Clustering.” China Economic Review 23 (3): 593–612. Mayneris, F., S. Poncet, and T. Zhang. 2018. “Improving or Disappearing: Firm-level Adjustments to Minimum Wages in China.” Journal of Development Economics 135: 20–42. Meng, X. 2012. “Labor Market Outcomes and Reforms in China.” Journal of Economic Perspectives 26 (4): 75–102. Puga, D., and A.J. Venables. 1996. “The Spread of Industry: Spatial Agglomeration in Economic Development.” Jour- nal of the Japanese and International Economies 10 (4): 440–64. ———. 1999. “Agglomeration and Economic Development: Import Substitution vs. Trade Liberalization.” Economic Journal 109 (455): 292–311. Qu, Y., F. Cai, and X. Zhang. 2013. “Has the ‘Flying Geese’ Occurred in China? An Analysis on the China’s Manu- facturing Industries from 1998 to 2008.” (in Chinese) China Economic Quarterly 12 (3): 757–76. Database on Indian Economy. Reserve Bank of India, India (accessed October 2021), https://dbie.rbi.org.in/ DBIE/dbie.rbi?site=home. Ruan, J., and X. Zhang. 2014. “‘Flying Geese’ in China: The Textile and Apparel Industry’s Pattern of Migration.” Journal of Asian Economics 34: 79–91. Santos Silva, J. M. C., and S. Tenreyro. 2006. “The Log of Gravity.” Review of Economics and Statistics 88 (4): 641–58. Shen, L., P. Koveos, X. Zhu, F. Wen, and J. Liao. 2020. “Outward FDI and Entrepreneurship: The Case of China.” Sustainability 12 (13): 5234. Wages and Statistics, Ministry of Labour & Employment, Government of India (accessed October 2021), https://labour.gov.in/wages-and-statistics. Wang, J., and M. Gunderson. 2015. “Adjustments to Minimum Wages in China: Cost-Neutral Offsets.” Industrial Relations 70 (3): 510–31. Wei, S.-J., Z. Xie, and X. Zhang. 2017. “From ‘Made in China’ to ‘Innovated in China’: Necessity, Prospect, and Challenges.” Journal of Economic Perspectives 31 (1): 49–70. Wen, M. 2004. “Relocation and Agglomeration of Chinese Industry.” Journal of Development Economics 73 (1): 329–47. Zhang, X., J. Yang, and S. Wang. 2011. “China Has Reached the Lewis Turning Point.” China Economic Review 22: 542–54.