Effects of global trade integration on skill polarization and wage disparity

Abstract

This paper examines how the wage gap and skill distribution among workers with different skills respond to trade integration. Using Chinese city-level data for 2010–2020, we find that trade has a negative impact on the wage gap between middle and primary skilled workers based on their education, and a positive impact on the wage gap between high and middle skilled groups. The former has a negative effect on the mean value and distribution of labor skills, while the latter has a positive effect. Therefore, international trade activities will induce the problem of skill polarization, and the wage gap can explain 50% of the effect of trade on skill polarization. In addition, in the middle-aged group of workers and high-tech intensive industries, the income gap, especially the income gap between high and middle skill groups, brings about the problem of skill polarization.

Introduction

Throughout the global economic history, the phenomenon of middle class being hit due to economic downturn has appeared several times. In the post-pandemic era, ordinary salaried workers also face significant social pressure, especially the middle class with vocational education and professional skills who are most affected by the macro environment (Lin and Wu 2010). The disappearance of the middle class will be accompanied by a widening income gap and a downgrade in consumption (Blanchard and Willmann 2016), which is the problem of “polarization” of labor skills and wages. The skill polarization refers to the continuous increase in the employment share of high-skilled and low-skilled workers, while the employment share of middle-skilled workers continues to decline, presenting a phenomenon of “upward at both ends and downward in the middle” (Goos et al. 2009).

In recent years, many scholars have conducted relevant research on labor market issues in the United States, the European Union, and China (Goos et al. 2009; Autor et al. 2006; Lv and Zhang 2015; Chen et al. 2019). Unlike the path of “skill premium-skill polarization” dominated by the diffusion of service industry technologies in developed countries such as the United States (Autor et al. 2006), and also different from the “low-skilled labor trap” triggered by trade liberalization in developing countries like Mexico, China has shaped an asymmetric dynamic structure of skill demand by deeply integrating into the global value chain at the low-end links coexists with the technological breakthrough at the high-end links. This dynamic structure is characterized by the tripartite polarization of the accelerated expansion of high-skilled labor, the drastic shrinkage of medium-skilled labor, and the relative stickiness of low-skilled labor (Chen et al. 2019). In addition, the existing literature has the following viewpoints on the formation mechanism of the special tripartite polarization of China’s labor market. Firstly, the elasticity of skill supply under policy intervention. The elasticity of skill supply represents the labor market’s dynamic response capacity to changes in skill demand, which could be divided into two categories: short-term inelasticity and long-term elasticity. Short-term inelasticity means that due to the time lag in skill acquisition (such as long education cycles, limited training resources, etc.), it is difficult to adjust the skill distribution in the short term (Acemoglu 2003); long-term elasticity is mainly due to policy interventions (such as vocational training subsidies, expansion of higher education) and the dynamic adjustment of individual human capital investment (Autor 2014). For example, the expansion of higher education enrollment and vocational education subsidies have increased the proportion of high-skilled labor in the short term, while the labor market segmentation caused by the household registration system has exacerbated the skill mismatch among regions (Autor 2014; Li et al. 2018; Xia and Lu 2015). Secondly, the coordinated impact of export competition and import substitution. The technology upgrading oriented towards exports enables export industries to increase the high-skill premium through the “learning by doing” effect, while imported intermediate goods which replace the debugging and maintenance tasks of local medium-skilled positions, will cause medium-skilled labor to fall into a negative cycle of wage decline due to technological squeeze (Grossman and Rossi-Hansberg 2008; Li 2018; Bustos 2011; Lv and Zhang 2015). Thirdly, due to the vulnerability of the middle class, the employment of the middle class is highly concentrated in the standardized positions in the manufacturing industry, such as mechanical operation. Therefore, their employment is simultaneously faced with the dual impacts of automation and offshore outsourcing (Autor et al. 2006), while the low-skilled service industry with policy protection such as the minimum wage regulation, shows the characteristics of “non-marketization” (Li et al. 2017).

Various studies also attempt to demonstrate the reasons for the occurrence of labor polarization from different perspectives, such as skill-biased technological change (SBTC) (Berman et al. 2018; He and Liu 2023), routine-biased technological change (RBTC) (Hershbein et al. 2018), and production outsourcing (Grossman and Rossi-Hansberg 2008). These theories collectively weave a theme that skill polarization originates from exogenous demand shocks, such as shocks caused by trade integration that increase relative demand for specific types of labor. Existing theories often assume that skill supply is inelastic, but this is clearly not the case in a dynamic environment.

Acemoglu’s (2003) analysis suggests that when skill supply can continuously respond to changing skill demands, the economy will stay at a specific level on the relevant demand curve. Based on this research framework, this article will analyze the adjustments made by the skill supply side of labor market under exogenous shocks and how they contribute to the problem of skill polarization by the evidence from China market. This paper will position this exogenous shock in the context of global integration, analyze the impact of international trade on labor market. To some extent, the assumption in previous literature about the inelasticity of skill supply is reasonable because acquiring skills is a slow process, so in the short term, the skill distribution will remain relatively stable. However, in a dynamic environment, skill supply may not necessarily be inelastic. In particular, to cope with the foreign trade shocks brought about by global integration, governments will allocate a certain proportion of fiscal revenue as expenditure for labor re-education and skills training, including vocational training, short-term projects, online learning, and formal education subsidies (Autor 2014). Considering the continuous effectiveness of these policies, skill supply may adjust accordingly and may have an impact on the skill distribution of the entire society.

Through systematic analysis, this paper aims to answer the following questions: What is the impact of external demand shocks caused by trade on wage gaps between workers with different skills? How does trade shock affect the skill distribution of workers? To what extent does trade, through the role of wage gaps, affect the skill distribution of workers?

This paper uses the Chinese Family Panel Studies (CFPS) data and combines it with China’s foreign trade data for prefecture-level cities to study the above issues. China has made significant progress in the training of skilled talents. The education sector has enriched the team of skilled talents, enhanced innovation capabilities, and provided a strong impetus for high-quality development through top-level education system design, improved skills training, and talent development platforms. At the same time, the government is also striving to provide labor mobility with greater flexibility through policy operations such as building and cultivating mature production factor markets, breaking down “iron rice bowls,” de-bureaucratizing, and reducing the threshold for urban settlement in large cities (Bai et al. 2016; Xia and Lu 2015; Du et al. 2014). In the long run, these efforts have made skill supply more elastic, helping to adjust the skill level of the workforce in response to demand shocks caused by trade.

In terms of empirical design, the analysis in this paper is conducted in two steps. In the first step, the paper examines the impact of foreign trade activities on wage disparities among different skilled workers in different regions and their influence on the distribution of skills among workers. Education level is used as a representation of skills, and wage disparities are divided into two categories: income differences between workers with higher education (bachelor’s degree or above) and those with middle-level education (high school/technical school/vocational school), as well as income differences between workers with middle-level education and those with only primary education (compulsory education, illiteracy, semi-literacy). To address endogeneity concerns, the paper employs instrumental variables using changes in the import and export values of various product categories under the SITC classification system in the United Nations UNcomtrade database (UNCOMTRADE) to identify the impact of trade activities. In the second step, the paper investigates whether the impact of trade on skill distribution occurs through changes in wage disparities. Specifically, in the first step, the paper constructs wage disparity changes resulting from exogenous trade shocks and then links these changes to the distribution of skills. The aforementioned analysis is based on data at the city level. This paper evaluates the changes in wage disparities caused by changes in foreign trade experienced by cities and how they affect the distribution of skills in the labor market of the next period at the city level. The empirical analysis of this paper is based on a partial equilibrium analysis model. This model enables us to determine how exogenous changes in the wage gap affect individuals’ motivation to upgrade their skills, and how individual-level decisions, when aggregated, influence the distribution of the entire skill set at the macro level.

This study finds that exogenous trade changes have a negative effect on income disparities between middle-skill workers and low-skill workers, but they have a positive impact on income disparities between high-skill and middle-skill workers. Additionally, trade leads to an increase in both the mean and dispersion of skills measured by educational level, indicating the presence of skill polarization. Finally, the empirical analysis demonstrates that changes in wage disparities resulting from trade integration are an important component of the overall impact of trade on skill allocation. The predicted changes in wage disparities from exogenous trade shocks can explain around 50% of the changes in labor skill distribution at the city-level.

The main contributions of this article are reflected in the following: Firstly, this article investigates the impact of international trade on the wage gap among skilled workers. Currently, empirical research literature has not reached a consensus on the impact of trade globalization on wage inequality among workers. Research conducted in Denmark (Hummels et al. 2014) and the United States (Lopresti et al. 2016) suggests that imports have a negative impact on wages regardless of workers’ skill levels. On the other hand, some studies have affirmed the promoting effect of exports on the wages of high-skilled workers (Munch and Skaksen 2008; Li 2018). In the context of the domestic labor market, there are also differences in conclusions. For example, Liu et al. (2021) believe that offshore outsourcing (i.e., export) in a region mainly leads to wage polarization among workers with different skill levels through improving the productivity of high-skilled labor and technology spillover. However, Yang’s study (2014) argues that international trade, whether through imports or exports, is not the dominant cause of labor polarization, and its main mechanism stems from biased technological progress towards specific skills. To address the differences in the above research conclusions, this article supplements the field by examining the impact of foreign trade scale on wage inequality in local labor markets.

Secondly, in the empirical process, this article not only examines the impact of trade on the average skill level but also assesses its influence on skill dispersion (i.e., the degree of skill distribution variance). Existing literature provides some empirical explanations for the interregional differences in skill dispersion from various perspectives, mostly starting from the government’s education policies or investment in scientific research and development (Friesen and Krauth 2007; Rycx et al. 2015; Lv and Zhang 2015; He and Liu 2023). However, there is a lack of systematic exploration from the perspective of trade, especially regarding the labor polarization caused by skill dispersion. This issue is particularly relevant in China’s current context with a significant increase in social pressure on the middle class. Moreover, at the societal level, the dispersion of labor skills directly affects a country’s comparative advantage and thus affects the flow of trade (Gallipoli et al. 2014). This article supplements existing research by empirically evaluating the impact of trade on skill distribution.

Thirdly, this article refines the specific channels through which exogenous trade shocks affect the distribution of labor skills, namely changes in wage gaps. Only a small portion of existing research addresses the impact of trade shocks on the average skill level of a country or region. For example, Atkin’s study (2016) demonstrates how the North American Free Trade Agreement (NAFTA) leads to an increase in high school dropout rates in Mexican cities. Blanchard and Olney (2017), through empirical evidence, find that educational attainment is influenced by exogenous drivers composed of a country’s exports, providing insights into understanding how human capital investment evolves with changing trade patterns. Although these studies are insightful, they do not touch upon the channels that cause changes in wage inequality among workers. In comparison, this article examines the impact of trade on the distribution of different skilled labor by investigating changes in wage gaps, thereby exploring how changes in wage disparities between different skills further affect the distribution of workers with different educational levels based on the influence on the average skill level.

The remainder of the paper is structured as follows: the second part “Theoretical mechanisms” provides theoretical analysis on how individuals made decisions on skill by shocks on wages. Part “Data and model” describes our model and data. Part “Empirical analysis” presents the regression estimation and robustness tests, and part “Conclusion and suggestion” summarizes the paper.

Theoretical mechanisms

The core logic of the theoretical model in this paper is to reveal how exogenous changes in the wage gap affect the skill distribution through individuals’ skill-upgrading decisions. In this section, a simple partial equilibrium model is introduced to link exogenous wage gap shocks with the skill decisions made by heterogeneous individuals. This section links the impact of exogenous wage gap shocks to skill decisions made by heterogeneous individuals through the introduction of a simple partial equilibrium model. Unlike the main focus of the subsequent empirical analysis on the impact of exogenous trade shocks on wage gaps, this model does not explicitly model the origins of wage gap shocks (i.e., the impact of trade on wages and labor skills) (Blanchard and Olney 2017; Blanchard and Willmann 2016; Bustos 2011; Costinot and Vogel 2010) in order to maintain the simplicity of the theoretical framework. This theoretical framework provides an economic intuition for how exogenous changes in wage gaps affect the skill distribution (i.e., the mean and variance of skill distribution) of a country. In addition, the rationality of the partial equilibrium model lies in the fact that the analysis within the partial equilibrium framework focuses on the transmission channels of the wage gap by stripping away the complex impacts of trade on the factor market (such as price effects, industrial restructuring, etc.). This simplification is in line with the classic framework of Acemoglu and Autor (2011), that is, first analyze the local mechanism and then empirically test the overall effect of external shocks. This model allows us to determine how exogenous changes in the wage gap affect individuals’ motivation to upgrade their skills, and how individual-level decisions, when aggregated, influence the distribution of the entire skill set at the macro level. Besides, in the setting of this model, the wage gap is assumed to be an exogenous variable, and this setting also reserves space for the introduction of international trade shocks.

Previous literature has primarily focused on the influence of expected wage premiums for higher education on individual decisions to attend college, there has been a lack of systematic theoretical analysis on how wage gaps affect the distribution of different skilled labor in the economy and society. As mentioned in the introduction, understanding the trends in the skill distribution of a country’s labor force is crucial as it can shape economic inequality, consumption capacity, long-term stable growth, and trade patterns. This theoretical framework elucidates the distribution effects of the labor force, and the predicted outcomes of the model will guide the empirical hypotheses regarding the impact of wage gaps on the mean and dispersion of the skill distribution of the labor force in the subsequent sessions.

Assume that a country’s labor force consists of numerous workers with different but continuously distributed abilities, with individual i’s ability being a_i, which has a boundary of 0 < a ≤ a_i ≤ $\bar{a}$ and remains unchanged throughout their lifetime. The ability a is a continuous distribution with a cumulative distribution function F(a) and a corresponding density function f(a). Each person’s life is divided into three stages, each lasting for one unit of time. In stage t, individual i decides whether to strive for a higher skill level or maintain their current skill level. An indicator function I_it is used to represent the decision made by the individual. When I_it = 1, it means individual i decides to upgrade their skills in stage t, and will obtain a higher wage level w(s_it + $\bar{e}$) in the next stage corresponding to their new skill level s_it + $\bar{e}$. If I_it = 0, it means they will continue to receive the same wage level based on their current skill level, and their wage level in the next stage will be w(s_it). However, obtaining a higher level of skills requires obtaining education $\bar{e}$ with a cost of opportunity, which is the time spent upgrading personal skills and affects their labor income in the same period. This opportunity cost increases as the amount of energy required for education increases and decreases with an increase in an individual’s innate ability, and therefore can be designated as $\bar{e}$/a_i. An individual’s ability a determines the opportunity cost required for skill upgrading. The higher the ability (the larger the value of a), the lower the opportunity cost, and the more inclined the individual is to carry out skill upgrading. Therefore, the ability a directly affects the frequency and number of times of skill upgrading, and further determines the cumulative process of the skill level.

Each individual maximizes their lifetime utility based on their own consumption c_it, and we assume that individuals can use their lifetime income W (W ≡ Σ³_t=1w(s_it)p_t(1 − I_it$\frac{\bar{e}}{{a}_{i}}$)) completely smoothly throughout their lifetime. We represent individual i’s skill selection problem with the following equation:

$${V}_{i}={\max }_{{I}_{it}}U({c}_{i1})+\beta U({c}_{i2})+{\beta }^{2}U({c}_{i3})$$

(1)

The constraint conditions are:

$$\mathop{\sum }\limits_{t=1}^{3}{c}_{it}{p}_{t}=\mathop{\sum }\limits_{t=1}^{3}w({s}_{it}){p}_{t}\left(1-{I}_{it}\tfrac{\bar{e}}{{a}_{i}}\right)$$

(2)

$${s}_{it}=\underline{s}+\mathop{\sum }\limits_{k=1}^{t-1}\overline{e}{I}_{ik}$$

(3)

U(c_it) represents the utility brought by consumption, p_t represents the price index for each period, s is the minimum skill level each individual possesses in the first stage, and skill level improvement can only be achieved through obtaining education with an additional factor e. It is assumed that wages will increase with skill level improvement ($\frac{dw}{ds}$ > 0) (Lemieux 2006; Acemoglu and Autor 2011). Finally, it is assumed that the inflation rate is non-negative, i.e., p1 ≤ p2 ≤ p3.

Based on the previous setting, two skill thresholds A₁ and A₂ can be defined (assuming a < A₁ < $\frac{a+\bar{a}}{2}$ < A₂ < $\bar{a}$). The entire labor force can be divided into three groups based on these skill thresholds: the primary workers who have not acquired any skills, the intermediate workers who possess a higher level of labor skills (education) compared to the primary workers, and the advanced workers who have a higher level of skills compared to the intermediate workers. The specific formula for the skill thresholds is as follows:

$${A}_{1}=\tfrac{w(\underline{s}){p}_{1}\overline{e}}{({p}_{2}+{p}_{3}){\Delta }_{1}},{A}_{2}=\tfrac{[w(\underline{s})+{\Delta }_{1}]{p}_{2}\overline{e}}{{p}_{3}{\Delta }_{2}}$$

(4)

In the above equation, Δ₁ ≡ w(s + $\bar{e}$)-w(s) represents the wage gap between intermediate workers and primary workers, while Δ₂ ≡ w(s + 2$\bar{e}$)-w(s + $\bar{e}$) represents the wage gap between advanced workers and intermediate workers. If an individual’s innate ability is below the first threshold (a_i ≤ A₁), their optimal choice for maximizing utility throughout their lifetime is not to upgrade their skills. However, if a worker’s innate ability falls between the two skill thresholds (A₁ < a_i ≤ A₂), the optimal choice for maximizing utility is to upgrade their skills once. For individuals whose innate ability exceeds the second threshold (a_i > A₂), their optimal choice is to upgrade their skills twice in order to attain the higher income associated with the high skill level.

By using this framework, it is easy to demonstrate how exogenous changes in wage gaps (Δ₁ and Δ₂) affect the level of ability thresholds (A₁ and A₂) and consequently impact skill distribution. On one hand, according to Eq. (4), an increase in Δ₁ reduces A₁ and increases A₂. The intuitive rationale behind this is that as Δ₁ increases, low-skilled individuals experience a higher return on acquiring skills, while the opportunity cost of skill acquisition for intermediate-skilled individuals rises. As a result, the gap between thresholds A₁ and A₂ widens. This is because some low-skilled workers who previously had no intention of upgrading their skills might attempt a single skill upgrade to reach an intermediate skill level. However, due to the raised second threshold, many intermediate-skilled workers are less willing to invest effort in further skill upgrading to enter the high-skill group. Therefore, assuming that individual abilities follow a uniform distribution and that Δ₂ remains stable, the threshold A₂ is more sensitive to an increase in Δ₁ compared to A₁ (i.e., $|\tfrac{\alpha {A}_{1}}{\alpha {\Delta }_{1}}| < |\tfrac{\alpha {A}_{2}}{\alpha {\Delta }_{1}}{|}_{{\Delta }_{2}}$). In this case, the space for upward development towards the top level becomes smaller, making it more difficult, thus resulting in a larger number of individuals staying at the intermediate level compared to those who move up from the primary to the intermediate level. As a consequence, both the average skill level and the skill dispersion of the labor force decline when Δ₁ increases. Specifically, when Δ₁ increases, the joint effect of the decrease of A₁ (the threshold between the primary level and the intermediate level is lowered) and the increase of A₂ (the threshold between the intermediate level and the advanced level is raised) leads to the expansion of the intermediate skill group (the interval [A₁, A₂] becomes wider) and the contraction of the advanced group (the interval [A_2, $\bar{a}$] becomes narrower). For a uniform distribution, the extent of the expansion of the intermediate group (the increment of A₂−A₁) must exceed the extent of the contraction of the advanced group (the decrement of $\bar{a}$ - A₂), because the increase of A₂ is restricted by the elastic constraint of Δ₁ on A₂. As a result, the relative growth of the proportion of the intermediate group exceeds the absolute decrease of the advanced group, ultimately reducing the overall average skill level. At the same time, since the skill distribution shifts from being concentrated at the primary level to being concentrated at the intermediate level, the variance (as E(s²) − [E(s)]²) decreases due to the increase in the proportion of the intermediate group. Therefore, the following can be stated: An increase in Δ₁ reduces the average value of the labor force skills and the dispersion of their skills.

On the other hand, according to Eq. (4), when Δ₁ remains constant, an increase in Δ₂ does not directly impact A₁ but lowers A₂, i.e., $||\tfrac{\alpha {A}_{2}}{\alpha {\Delta }_{2}}{|}_{{\Delta }_{1}} < 0$. Due to the increased returns brought by skill promotion for intermediate-skilled workers, more people will attempt skill upgrades at this point, while it has no effect on primary skilled workers. Specifically, when Δ₂ increases, the decrease of A₂ leads to the expansion of the advanced group (the interval [A₂, $\bar{a}$] becomes wider). At the same time, with Δ1 remaining unchanged, the intermediate group remains stable. Due to the increase in the weight of the advanced group, the average value will inevitably rise, and the variance will increase because the gap between high-skilled and intermediate-level skills widens and the proportion of the advanced group increases. Therefore, this will naturally increase the overall average skill level and the dispersion of skills of the labor force. Thus, the following conclusion can be drawn: An increase in Δ₂ raises the average value of the labor force skills and the dispersion of their skills.

Based on the above models, as long as the empirical results in the following sessions can prove that trade shocks can lead to a decrease in Δ₁ and an increase in Δ₂, it can be proven that the average skill level and variance of workers will both increase. This paper will use data from individual workers to verify the above inference.

Data and model

Model design

The empirical methodology in this paper relies on three key identification assumptions driven by existing literature: Firstly, for labor skills, exogenous trade shocks are expected to generate a pure skill transfer demand (Wei and Li 2017). Secondly, in the short term, skill supply is considered to be inelastic, and the price effects caused by trade shocks primarily manifest in changes in relative wages across different skills (Acemoglu 2003). There is an old saying in China that “it takes a hundred years to cultivate a tree” indicating that acquiring high skills takes time. Therefore, trade shocks ultimately only affect changes in wage disparities without impacting skill supply. Thirdly, although skill supply is fixed in the short term, it is not constant in a dynamic long-term environment (Acemoglu 2003). Under these assumptions, exogenous growth in trade activity may lead to changes in wage disparities and potentially influence individuals’ motivation to improve or maintain their current skill levels. Aggregating individuals at the macro level, individual-level skill decisions result in overall changes in skill distribution. The empirical analysis in this paper is built upon a partial equilibrium analytical model, which enables us to examine how the exogenous changes in wage disparities affect individuals’ motivation to enhance their skills and how individual-level decisions, when aggregated, impact the overall distribution of skills at the macro level.

Next, this paper will first examine the impact of exogenous trade shocks on wage differentials between workers with different skills and the distribution of skilled labor. Then, it will determine to what extent the effect of trade shocks on the distribution of skilled labor is realized through the channel of wage differentials.

• (1) The impact of trade on wage differentials and skill distribution of labor

Regarding the impact of trade on wage differentials, this paper estimates the following city-level equation:

$${\Delta }_{{mt}}=\alpha +{\beta }_{1}{{Export}}_{{mt}}+{\beta }_{2}{{Import}}_{{mt}}+{\gamma }_{m}+{\gamma }_{t}+{\varepsilon }_{{mt}}$$

(5)

In the above equation, Δ_mt includes wage differentials between workers in the middle and low-skill groups and between workers in the high and middle-skill groups (i.e., Δ₁ and Δ₂) in city m and year t. The two wage differentials at the city level are calculated using the Eq. (12), with the natural logs of the city’s export (Export_mt) and import (Import_mt) variables introduced as indicators of the city’s foreign trade intensity. As mentioned earlier, this data comes from the EPS database, and the above model is implemented by controlling for city (γ_m) and year (γ_t) fixed effects, with ε_mt as the error term.

However, it should be noted that Export_mt and Import_mt may have strong endogenous characteristics, and there may be some unobservable city-specific shocks that are simultaneously related to wage differential variables as well as trade. For example, a city’s level of trade openness may be caused by shocks to local productivity or factor demand and supply, but this will also affect the wage differentials of its workers. To address reverse causality and endogeneity issues caused by omitted factors, instrumental variable methods need to be used to proxy Export_mt and Import_mt.

Instrumental variables for city-level trade shocks are constructed through two steps: Firstly, following the method of Hummels et al. (2014), the UNCOMTRADE database is used to aggregate China’s imports and exports by industry from and to other economies. Weighting is done using the average proportion of each country’s foreign trade with China during the period 2005–2009 as weights. At the industry level, the standard used is the SITC first-digit (0–9) industry code, which is then aggregated to the industry level in China to form instrumental variables. Specifically, the export shock variable faced by China in year t is calculated as follows:

$${ex}{p}ortshoc{k}_{t}=\mathop{\sum }\limits_{c=1}^{C}\mathop{\sum }\limits_{j=1}^{J}\tfrac{{exp }_{jc\_2005-2009}}{{exp }_{j\_2005-2009}}\times {I}_{cjt}$$

(6)

The formula for calculating the import shock variable in year t is as follows:

$$importshoc{k}_{t}=\mathop{\sum }\limits_{c=1}^{C}\mathop{\sum }\limits_{j=1}^{J}\tfrac{im{p}_{jc\_2005-2009}}{im{p}_{j\_2005-2009}}\times {E}_{cjt}$$

(7)

In the equation above, I_cjt (E_cjt) represents the total amount of imports (exports) from (to) country c and industry j in year t, subtracted by the country’s imports (exports) with China. This treatment ensures the exogeneity of this portion with respect to China’s foreign trade. exp_{jc_2005–2009} (imp_{jc_2005–2009}) is the average value of China’s exports (imports) from (to) country c in industry j during the period 2005–2009 (5 years before the sample period), while exp_{j_2005–2009} (imp_{j_2005–2009}) represents the average value of China’s exports (imports) to (from) the rest of the world during the same period. These calculations determine the weights of country c in China’s foreign trade activities. Next, we need to allocate the intensity of import and export trade shocks at the national level to individual cities. We adopt the approach used by Hummels et al. (2014), where the share of the city’s labor force in the total national labor force during the period 2005–2009 is used to allocate the city-level export (import) volume. This process forms the instrumental variable exportshock_mt (importshock_mt). As pointed out by Pierce and Schott (2020), the share of city labor force before the sample period maintains an exogenous relationship with the simultaneous impact on trade and income disparities due to intertemporal technological changes.

The construction of the above instrumental variable has a significant correlation with the trade variable, which effectively meets the requirement of correlation. In addition, the core idea of this instrumental variable is to capture the global trade fluctuations that are exogenous to the Chinese urban labor market. Specifically, I_cjt (E_cjt) represents the foreign trade scale of industry j in country c in year t, which is calculated by subtracting the imports (exports) from China from the total imports (exports) of that industry from the whole world. This approach excludes the trade activities between country c and other countries. Thus, it is not affected by the economic characteristics of Chinese cities (such as labor force skills and wage levels). Moreover, by setting the weights of the historical averages before the sample period, this instrumental variable further enhances its exogeneity. Therefore, we believe that this instrumental variable also satisfies the characteristic of exogeneity. And the construction process of this instrumental variable is consistent with that in most relevant literature. For example, Autor (2014) used the growth of China’s exports to the United States as an instrumental variable for the U.S. labor market, and its core logic is also to utilize the exogenous international demand shock. Hummels et al. (2014) constructed instrumental variables through the changes in the demand of the export destinations of Danish enterprises to verify the impact of trade on wages, which also indirectly demonstrates the scientificity and feasibility of this instrumental variable.

Using a similar econometric model, this study can also examine the impact of trade on skill distribution. Specifically, it considers the average skill level (education) of workers (M-skill_mt) at the city level, as well as the dispersion of skills (Sd-skill_mt), which serves as a measure of skill diversity. The specific econometric model is as follows:

$${M-{skill}}_{{mt}}=\alpha +{\beta }_{1}{{Export}}_{{{mt}}\hbox{-}1}+{\beta }_{2}{{Import}}_{{{mt}}\hbox{-}1}+{\gamma }_{m}+{\gamma }_{t}+{\varepsilon }_{{mt}}$$

(8)

$${Sd}-{{skill}}_{{mt}}=\alpha +{\beta }_{1}{{Export}}_{{{mt}}\hbox{-}1}+{\beta }_{2}{{Import}}_{{{mt}}\hbox{-}1}+{\gamma }_{m}+{\gamma }_{t}+{\varepsilon }_{{mt}}$$

(9)

It should be noted that in the above two equations, the explanatory variables of import and export values are both lagged by one period. This is because there is a certain time lag in the impact of current trade shocks on skill distribution. Similarly, the instrumental variable approach mentioned earlier is employed here to address endogeneity concerns.

• (2) Foreign trade affects skill distribution through the change of wage gap

In the previous section, we established that trade affects wage inequality on one hand, and skill distribution on the other hand. We aim to estimate the impact of trade on skill distribution by examining the changes in wage inequality caused by trade at the city level. In this regard, we utilize the predicted actual changes in wage inequality as derived from Eq. (5) in the previous section. These changes in wage inequality, resulting from exogenous trade shocks, serve as the main explanatory variable to quantify the effects of trade on skill distribution mediated by changes in wage inequality. Specifically, this study constructs the following estimation model at the city level:

$$M-skil{l}_{mt}=\alpha +{\delta }_{ave}{\mathop{\Delta }\limits^{\frown {}}}_{mt-1}+{\gamma }_{m}+{\gamma }_{t}+{\varepsilon }_{mt}$$

(10)

$$Sd-skil{l}_{mt}=\alpha +{\delta }_{disp}{\mathop{\Delta }\limits^{\frown {}}}_{mt-1}+{\gamma }_{m}+{\gamma }_{t}+{\varepsilon }_{mt}$$

(11)

The variable $\mathop{\Delta }\limits^{\frown {}}$ _mt-1 in the equation above includes two components of wage inequality between two groups, namely Δ1 and Δ2. Both Δ1 and Δ2 are values predicted by the Eq. (5), and it is important to note that the predicted wage inequality is lagged by one period. This consistency in using lagged values for predicting wage inequality aligns with the model framework used to determine the impact of trade on skill distribution. The rationale behind this choice is also consistent, as it accounts for the lagged effect of wage inequality on skill distribution.

Data collection

The data related to the labor force is sourced from the China Family Panel Studies (CFPS), a biennial dataset released by the Chinese Household Tracking Survey. The data collection for CFPS is organized and implemented by the China Social Science Survey Center at Peking University. It is a nationwide, comprehensive social tracking survey project aiming to reflect the changes in Chinese society, economy, population, education, and health by collecting data at the individual, household, and community levels. CFPS adopts a stratified, multi-stage, probability proportion sampling method, covering 25 provinces and cities across the country, excluding Hong Kong, Macao, Taiwan and some border provinces. Taking 2010 as an example, it involves visits to 14,960 households with 42,590 individuals. The interviewed individuals are followed up across different periods in the survey. Moreover, the CFPS sample includes both urban and rural families. Through weight adjustment, such as the proportion of urban and rural populations, age structure, etc., it ensures statistical representativeness at the national and urban levels. It should be noted that the CFPS data will be integrated and processed based on city level to generate indicator data on the labor force situation specific to each city. As for the foreign trade indicators, the main source is the EPS database, from which foreign trade data at the city-level is obtained. This is used to calculate the foreign trade intensity of cities. Additionally, relevant data on China’s imports and exports are constructed using the UNCOMTRADE database to create instrumental variables. Furthermore, in the process of constructing instrumental variables, this study also requires obtaining urban labor force data, which is sourced from the “China Urban Statistical Yearbook.” Based on the above sources, attention is also paid to the control of data quality during the data matching process. For example, when calculating the wage gap at the urban level, the upper and lower 1% of individual wage incomes are winsorized, so as to eliminate the interference of extreme values. In addition, during the process of aggregating data at the urban level, we follow the one-to-one matching principle, and exclude the data with missing key variables or obvious biases in variable data. Finally, this paper obtained a total of 665 observations in 6 survey years from 127 cities in 25 provinces, thus forming an unbalanced panel.

Variables

Table 1 presents the descriptive statistics of the main variables in this study during the sample period from 2010 to 2020. The first two rows report the average wage gap between workers with intermediate education (high school/secondary vocational school/technical school/professional high school) and those with primary education (compulsory education and illiteracy/half-literacy) at the prefecture-level city level (denoted as Δ₁), as well as the average wage gap between workers with advanced education (bachelor’s degree or higher) and those with intermediate education (denoted as Δ₂). Instead of simply calculating the difference in average income between different groups of people, we follow the method proposed by Li (2018) and use the Ordinary Least Squares (OLS) to estimate individual-level wage income in cross-sectional data of “year-city.” This is used to determine the wage gap between workers with different skills. The specific calculation formula is:

$${{lnw}}_{{imt}}={\alpha }_{{imt}}+{\Delta }_{1{mt}}{{Sec}}_{{it}}+{\Delta }_{2{mt}}{{Ter}}_{{it}}+{\varepsilon }_{{imt}}$$

(12)

Table 1 Descriptive statistics.

In the above equation, w_imt represents the wage level of individual i in city m in a specific year t. It estimates the wage level based on cross-sectional data. Sec denotes a dummy variable indicating whether the individual has obtained intermediate education or higher, while Ter represents a dummy variable indicating whether the individual has obtained advanced education. Coefficients Δ_1mt and Δ_2mt measure the wage premium for intermediate-skilled workers compared to primary-skilled workers and for advanced-skilled workers compared to intermediate-skilled workers, respectively. For simplicity, the following discussion uses Δ₁ and Δ₂ to represent the wage gap between workers with different education (skills) levels. During the sample period, the average wage gap between intermediate and primary skilled workers is 14.8%, while the average wage gap between advanced and intermediate skilled workers is 24.7%.

The core logic of the above wage gap determination equation is based on the additivity of the returns to education (Autor 2014; Lemieux 2006). Specifically, the wage gap Δ₁ reflects the wage premium of secondary education relative to primary education. The wage gap Δ₂ reflects the marginal additional premium of higher education on the basis of secondary education. This setting assumes that the skill hierarchy is continuous, that is, the wage premium of high-skilled labor force is based on secondary education. For example, an individual with a bachelor’s degree (Ter = 1) must have completed high school (Sec = 1) education. Their wage consists of two parts, one is the benchmark premium (Δ₁) of high school education and the additional premium (Δ₂) of undergraduate education. In addition, the mutual exclusivity of the dummy variables and the independence of the parameters also ensure that the calculation of the wage gap is error-free. In this paper, Sec represents the dummy variable indicating whether the individual worker has obtained an intermediate or higher academic degree, and Ter represents the dummy variable indicating whether the individual has obtained a higher academic degree. The above dummy variables are strictly mutually exclusive in the model, which can effectively avoid overlap. The estimated coefficients of Δ₁ and Δ₂ are economically independent, effectively ensuring the independence of the parameters. In the field of labor economics, there are extensive precedents for the design of such models. When studying skill-biased technological progress, Autor et al. (2013) used a similar method to distinguish the cumulative effects of secondary and higher education on wages. Lemieux (2006) analyzed the marginal returns of different education levels through an equation that includes multiple education dummy variables, which also verified the rationality of the above model.

Through the above processing, the data of labor force at the urban level (that is, the wage gap at the urban level) was finally obtained. In addition, as mentioned before, the data of foreign trade at the prefecture-level city level was obtained through the EPS database, so as to calculate the foreign trade intensity of the prefecture-level city. And the relevant data on China’s imports and exports was obtained through the UNCOMTRADE database to construct instrumental variables. The above variables were integrated and matched to the prefecture-level city level, and finally the urban-level data required for this paper was obtained.

The main skill variables in this study are represented by the average (Edumean) and standard deviation (Edusd) of the educational level of non-agricultural registered workers in each city over the years. The rationality of this division lies in the fact that the existing literature generally holds that education is the core channel for skill acquisition (Autor et al. 2013). For example, higher education significantly improves individuals’ “observable skills” such as analytical ability and technical proficiency through systematic knowledge transfer and cognitive training. Therefore, directly mapping educational upgrading to skill improvement in the model is a reasonable simplification of the real labor market. In addition, this setting is also in line with the core view of the human capital theory, that is, education is an explicit carrier of skills, and ability indirectly shapes the skill distribution by influencing educational choices.

The classification of education levels is based on the level of education received by the workers. The specific classifications are as follows: (1) Workers with primary skills. The educational experience of such workers is compulsory education or below (that is, those who have graduated from primary school or junior high school, or have not completed compulsory education, including illiterates or semi-illiterates). Such workers usually engage in low-skill-intensive jobs, such as manual labor, simple operation positions, etc. Their skill acquisition mainly depends on basic education and practical experience, and they lack a systematic vocational training background or higher education background. (2) Workers with intermediate skills. The educational experience of such workers is senior high school, technical secondary school, technical school or vocational high school education (that is, those who have completed secondary vocational education or general senior high school education). Such workers possess certain professional knowledge and operation skills and are capable of taking up positions that require a medium technical level, such as machinery operation, primary management, technical staff, etc. Secondary vocational education directly meets the needs of industries, and its curriculum design usually includes vocational skill training. Therefore, there is a high consistency between educational attainment and skill level. (3) Workers with advanced skills. The educational experience of such workers is a bachelor’s degree, junior college degree or above (including general higher education and higher vocational education). Such workers possess a high level of professional knowledge or complex technical capabilities and usually engage in high-skill positions such as management, research and development, design, etc. Their skill formation depends on systematic higher education and the ability of continuous learning. Specifically, individuals with compulsory education or below are categorized as 1, individuals with high school/secondary vocational school/technical school/professional high school education are categorized as 2, and individuals with bachelor’s degree or higher are categorized as 3. Based on this classification, the average education level in the sample is 1.4, with a standard deviation of 0.59. Additionally, this study also calculates the proportions of primary-skilled workers (sharep), intermediate-skilled workers (shares), and advanced-skilled workers (sharet) among the labor force in each city. It can be observed that over 70% of the labor force still belongs to the primary-skilled category, indicating that their education level does not exceed compulsory education.

Finally, the primary measure of trade activity at the city level is the value of exports (Export) and imports (Import). However, in the empirical process, endogeneity issues are addressed by constructing instrumental variables. The basic idea is to allocate the national-level export and import values to various cities based on the industry employment shares in a reference year (Autor et al. 2013). Additionally, from the data in Table 1, it can be observed that the overall export volume of cities in China is greater than the import volume.

Empirical analysis

In this section, we will first discuss in detail the impact of foreign trade on wage inequality and skill distribution in China’s intercity labor market. We will then test the mediating role of wage inequality changes caused by foreign trade on skill distribution. Next, we will examine whether alternative mechanisms, such as changes in overall educational structure of workers in cities, rather than just those in the manufacturing sector, could also affect our estimation results. In addition, we will conduct various robustness checks to confirm the reliability of the estimation results.

• (1) The impact of foreign trade on wage gap and labor skill distribution

In Table 2, we report the impact of foreign trade on income gaps between low- and medium-skilled workers (Δ1) and high- and medium-skilled workers (Δ2). Controlling for city and year fixed effects, we use instrumental variables to address endogeneity issues. The estimated coefficient of the first-stage IV is significant, and the F-statistic is much larger than 10. The results indicate that a 10% increase in the intensity of export shocks at the city level leads to a 4.7 percentage point decrease in Δ1 and a 6.7 percentage point increase in Δ2. These findings are consistent with previous studies by Munch and Skaksen (2008) and Li (2018), which suggest that wage inequality between high- and medium-skilled workers widens due to export shocks, resulting in higher wages for skilled workers. However, at the intercity level, the scale of import trade does not significantly affect the wage gap between high- and medium-skilled workers. Instead, it has a significant negative impact on the wage gap between medium- and low-skilled workers, with a 10% expansion in import scale leading to an average increase of 7.85% in income disparity among workers with lower educational levels. This finding confirms a long-standing conclusion in the trade literature that imports have a negative effect on wages, especially for relatively low-skilled workers (Hummels et al. 2014; Li et al. 2018).

Table 2 Skill distribution of foreign trade, wage gap and labor force (2SLS estimation method).

Columns (3)–(4) in Table 2 also report the impact of foreign trade on labor skill distribution. Foreign trade has a significant positive effect on both the mean and dispersion of labor skill distribution. Specifically, a 10% increase in export intensity in the previous year leads to an increase of 8.8 (6.4) percentage points in the mean (dispersion) of worker skills in the current period. Similarly, a 10% increase in imports in the previous year results in a 7.6 percentage point increase in skill mean for the following year. However, there is no significant impact on the skill dispersion of workers in the subsequent year.

Next, we examine the impact of foreign trade on the proportion of workers in three skill categories in columns (5)–(7). We find that foreign trade decreases the proportion of medium-skilled workers while increasing the proportions of both low- and high-skilled workers, with a larger effect on high-skilled workers. A 10% increase in export (import) scale leads to a 1.5 (1.2) percentage point increase in the proportion of highly-educated employees. Overall, the estimates from columns (3)–(7) in Table 2 suggest that foreign trade shifts the skill distribution of China’s labor force to the right but with a polarized distribution trend.

(2) The impact of foreign trade on the distribution of labor skills through the wage gap

Table 2 confirms that foreign trade can simultaneously affect wage disparities and skill distribution. Next, we examine Eqs. (10) and (11) to analyze whether changes in inter-group wage disparities caused by foreign trade in the previous period have an impact on the characteristics of labor skill distribution in the current period. Table 3 presents the results of tests conducted for this concept. In columns (1) to (2), for every 10% increase in the predicted wage gap ($\mathop{\Delta }\limits^{\frown {}}$₁) between medium and low-skilled workers, the urban-level skill mean (dispersion) decreases by 0.037 (0.014) respectively. However, the predicted wage gap ($\mathop{\Delta }\limits^{\frown {}}$₂) between high and medium-skilled workers has a positive and significant effect on skill mean and dispersion. Specifically, a 10% increase in ($\mathop{\Delta }\limits^{\frown {}}$₂) leads to a 0.041 (0.034) increase in the average skill mean (dispersion) across cities. It is worth noting that these results are consistent with the conclusions proposed in the previous mechanistic analysis.

Table 3 Effects of wage gap on labor skill distribution.

Combined with the estimation results from Table 2 and the first two columns of Table 3, it is evident that changes in wage disparities caused by foreign trade have a unidirectional effect on the skill distribution in the local labor market. Specifically, an expansion of foreign trade leads to a reduction in income disparities between medium and low-skilled groups, which in turn widens the skill distribution. Additionally, the expansion of foreign trade increases incomes for high and medium-skilled groups, further widening the skill distribution. Consequently, the impact of foreign trade shifts the skill distribution of Chinese workers to the right while simultaneously generating a two-tiered differentiation phenomenon.

To better understand the results of the tests, the quantitative analysis is as follows: The coefficient of the impact of exports on skill mean (dispersion) is 0.0088 (0.0064), while the coefficient of the impact of exports on Δ1 (Δ2) is −0.0047 (0.0067). Specifically, the coefficient of the impact of Δ1 on skill mean (dispersion) is −0.3721 (−0.1435), and the coefficient of the impact of Δ2 on skill mean (dispersion) is 0.4128 (0.341). This implies that the impact of exports on skill mean can account for 19.88% of the total effect through Δ1, while the impact on skill mean through Δ2 can account for 31.42% of the total effect. In summary, wage disparities can explain over 50% of the impact of exports on skill mean levels.

Columns (3)–(5) in Table 3 report the impact of wage changes caused by previous foreign trade on the proportions of employment in the labor market for low, medium, and high-skilled workers. For every 10% increase in Δ1, there is a 0.38% increase in the proportion of medium-skilled workers and a 0.51% decrease in the proportion of low-skilled workers. This confirms the conclusion proposed in the previous mechanistic analysis that the increase in Δ1 widens the gap between skill thresholds A1 and A2. Specifically, in this theory, although the increase in Δ1 raises the marginal return for low-skilled individuals to acquire intermediate skills (i.e., secondary education), it also raises the marginal cost for intermediate-skilled workers to acquire higher skills (i.e., tertiary education) when considering the opportunity cost of losing current wages during skill upgrading. On the other hand, for every 10% increase in Δ2, there is a 0.7 percentage point increase in the proportion of high-skilled workers, which aligns with the theoretical prediction that only the skill threshold A2 decreases when Δ2 increases. Particularly, the increase in Δ2 leads to an increase in the number of intermediate-skilled workers seeking higher skill returns, but it does not affect low-skilled workers.

Overall, in the CFPS data, as trade integration develops, $\mathop{\Delta }\limits^{\frown {}}$₁ decreases and $\mathop{\Delta }\limits^{\frown {}}$₂ increases. As a result, the skill distribution responds by shifting to the right and becoming polarized. Specifically, low-skilled workers are not encouraged to upgrade their skills, while intermediate-skilled workers are encouraged to do so. When the proportion of upgrading workers exceeds the proportion of workers who choose not to upgrade, it leads to a higher average skill level.

• (3) Placebo test

So far, we have interpreted the previous estimation results as follows: changes in wage inequality caused by trade impact the incentives for skill upgrading, leading to a polarization of the skill distribution. However, the observed changes in the skill distribution in our data could also be a result of other mechanisms, such as an overall increase in educational skills among the labor force in the region, not just in the manufacturing sector. If this is the case, the reason for the expansion of skill distribution in the manufacturing industry may be due to the transfer of corresponding skilled individuals from other industry sectors, particularly the service sector. Therefore, in columns (1) to (2) of Table 4, we first examine the impact of the previous year’s predicted wages on the average education level (edutime) and the proportion of highly educated individuals (bachelor’s degree or above) (highedur) at the prefecture-level. In columns (1) to (2) of Table 4, it can be seen that the predicted wage gap between the two skill groups does not provide explanatory power for changes in the overall educational level of society. This also implies that the changes in skill distribution are not caused by overall progress in social education. Another possible explanation is that changes in wage inequality may affect the age composition of the labor force, thereby observing polarization in skills. Intuitively, if there are fewer (or more) older workers, the wage gap between skill groups induced by foreign trade would incentivize them to postpone retirement (or not participate in the labor force), or the opposite could occur, thereby affecting the composition of intercity labor force skills. To test this potential channel mechanism, columns (3) to (4) of Table 4 examine the impact of the predicted wage gap on the changes in the proportion of workers below 31 years old (Sharep31) and above 45 years old (Shares45). The results show that although the proportion of workers below the age of 31 is positively correlated with Δ2, the proportion of workers above the age of 45 is not significantly influenced by changes in the wage gap. Therefore, changes in the age composition of the labor force are not the key driving factor affecting the baseline test results.

Table 4 Placebo test.

(4) Robustness test

The robustness tests in this study were conducted from three perspectives. The detailed summary tables for the following robustness tests are available upon the request. First, an alternative method was used to calculate the wage gap between different skill groups by simply calculating the average wage gap for each group. This was used to measure Δ1 and Δ2, and the corresponding tests in Tables 3 and 4 were conducted again based on this method. The results remained consistent with the previous analysis.

Second, in the calculation process of the instrumental variable I_cjt (E_cjt) part, besides subtracting the total amount of global imports (exports) from the trade scale after deducting China’s imports (exports) from country c, the trade volume with four countries geographically and culturally close to China (Japan, South Korea, Singapore, Vietnam) was also deducted. This is because these countries have similar social and industrial structures, as well as commercial cycles that may be strongly correlated with China’s trade, which would violate the exogeneity assumption of the instrumental variable. The instrumental variables constructed through this approach yielded consistent estimation results with the previous analysis.

Third, we examined whether the results reported in Table 3 were sensitive to using a 1-year lagged prediction of the wage gap. Estimations were re-conducted using 2 or 3 years of lagged values based on the rationale that workers may have more time to adjust their skill allocation to adapt to changes in wage inequality caused by trade. The direction and significance of the estimation results remained unchanged, but the estimated coefficients were slightly larger. This suggests that if more time is allowed for skill adjustment, this effect would become stronger, confirming the previous hypothesis that skill supply adjustments lag behind changes in wage inequality caused by trade.

• (5) Extended analysis: group tests based on age and industry

Next, this study extends the benchmark results to two dimensions for examination. Firstly, all employees are divided into three groups based on age: young workers (under 31), middle-aged workers (31–45), and older workers (above 45), and the estimation results are reported in Table 5. The study shows that the effects of wage gaps between low and medium skill groups, and between high and medium skill groups on average skills and skill dispersion are consistent within each group, with the impact of income gap between high and medium skill groups being slightly stronger. Among the impacts caused by wage gaps between high and medium skill groups, middle-aged workers are most affected, as the income gap widens, resulting in a more pronounced right shift and divergence in their overall distribution, regardless of both average and dispersion levels of skills.

Table 5 Effects of the wage gap on the distribution of skills in the Labor force (by age groups).

Next, this study also examines whether the impact of wage gap changes on skill distribution through foreign trade depends on the type of exported products. Blanchard and Olney (2017) indicate that trade composition plays a crucial role in influencing incentives for education attainment. The growth of low-skill-intensive exports lowers the average level of education, while the growth of high-skill-intensive exports raises the average level of education. The estimation results in the benchmark regression may mask the heterogeneity mentioned above. This study extends the research by Blanchard and Olney (2017) and focuses on the channel of wage gap changes caused by trade. Estimations are conducted based on industry groups at the inter-city level to analyze the wage gap changes among workers caused by foreign trade shocks in various industries and assess the impact of these predicted changes on the overall skill distribution. To differentiate, the classification method based on the “OECD Taxonomy of Economic Activities Based On R&D Intensity” is used here, dividing industries into high-tech and low-tech categories based on their research and development intensity. Columns 1 and 2 in Table 6 provide the results of predicted wage gap changes, which are estimated from trade shocks within high-skill-intensive industries, while columns 3 and 4 report the estimated coefficients of predicted wage gap changes caused by trade shocks in other industries. The study finds that the skill distribution effects of wage gap changes through the channel of trade are slightly larger in high-skill-intensive industries than in low-skill-intensive industries. As trade shocks in this case still reduce A and increase B, the mean and standard deviation of skills in the local labor market will still be positively affected by the predicted wage gap changes in both types of industries, especially the high-skill-intensive industries. Specifically, for high-skill-intensive industries, a 10% increase in the income gap for the high-middle skill group leads to a 3.2% increase in average skills and a 1.4% increase in skill dispersion. The corresponding impact in industries with lower skill intensity is smaller.

Table 6 Effects of wage disparities on the distribution of labor skills (by industry groups).

Conclusion and suggestion

This paper examines the impact of foreign trade on labor skill polarization from the perspective of international trade, specifically analyzing the effect of trade on wage income disparities among different skill levels of workers, and then determining the distribution of different skill labor through the influence of trade on wage gap changes.

First, trade integration has a negative impact on wage disparities between the medium-to-low skill group based on educational attainment, but has a positive impact on wage disparities between the high and middle skill groups. However, since the income gap for the medium-to-low skill group has a negative effect on skill mean and standard deviation, while the income gap for the high-middle skill group has a positive effect, this ultimately leads to labor skill polarization between cities caused by trade.

Second, the empirical tests emphasize that wage gap changes caused by trade are an important factor in explaining the overall impact of trade on skill distribution, accounting for more than 50% of the total effect. This is consistent with the predictions of the simple theoretical framework in the paper, which explains how changes in wage disparities affect individual decisions regarding skill upgrading. Exogenous changes in wage disparities affect the opportunity cost and return of skill upgrading, which is of practical significance in the current context of China’s relatively expanded educational background and increasingly flexible labor market.

Third, overall, the impact of income disparities between high and middle skill groups is greater. Among the effects caused by wage gap changes between high and middle skill groups, middle-aged workers are most affected, both in terms of skill mean and dispersion. Due to the widening income gap, they exhibit a more pronounced rightward shift and dispersion in overall distribution. Additionally, in the industry-specific analysis, the skill distribution effects of wage gap changes caused by trade in high-skill-intensive industries are slightly larger than those in low-skill-intensive industries.

According to the above conclusions, this paper provides the following policy implications: Firstly, the research provides policymakers with information on how external demand shocks such as trade integration affect skill distribution, especially how changes in wage disparities affect skill distribution. As a country’s skill allocation further affects future economic growth and inequality, understanding distribution changes and considering these changes in policy design is crucial. Secondly, since the impact of income disparities between high and middle skill groups is more prominent in terms of skill polarization, the current expansion of higher education may be more likely to induce skill polarization issues, although it can alleviate social employment pressure to some extent. This issue also deserves attention from decision-makers. Parallel development of higher education and vocational education, cultivating diversified technical talents, should be the target of education. Thirdly, for middle-aged workers, because the impact of wage disparities on skill distribution is greater than that of young and old workers, more support is needed in labor security and skills training. In high-tech industries, the impact of income disparities on skill polarization is also more prominent. Therefore, both employment and education policies may need to differ in specific age and industry sectors regarding implementation efforts and specific measures.

Based on the above research results, this paper puts forward the following policy suggestions, so as to systematically address the issue of skill polarization caused by trade. Firstly, this paper provides policymakers with information on how external demand shocks such as trade integration affect the skill distribution, especially on how the changes in the wage gap influence the skill distribution. Since the skill allocation of a country will further affect future economic growth and inequality, it is crucial to understand the changes in allocation and take these changes into account in policy design. Secondly, in response to the continuous widening of the wage gap between high-skilled and medium-skilled workers, it is necessary to establish a dynamic two-way coordination mechanism between industries and education to optimize the supply of high-skilled labor. For example, joint innovation centers for industry, university and research in key foreign trade-oriented industries can be established. Through the linkage of universities, enterprises and the government, the curriculum content of higher education can be updated in real time. In addition, referring to the German dual system model, relevant national departments can implement enterprise-led practical training semesters in vocational colleges to ensure that the cultivation of high-skilled talents is synchronized with technological iterations (Autor 2014; Li et al. 2017). Thirdly, considering the vulnerability of the middle class and the difficulty of midlife transition, a safety net for medium-skilled workers’ transition should be established to alleviate the need caused by job shrinkage. For example, by establishing special re-training funds for industries, relevant national departments can jointly introduce policies focusing on supporting the manufacturing industries affected by trade shocks and providing skill conversion packages for middle-aged workers. At the same time, relying on the national smart platform for vocational education, a modular micro-credential system can be developed to encourage medium-skilled employees to engage in informal learning (such as in-company training and MOOC courses). Fourthly, the phenomenon of polarization in high-tech industries is more severe, and the impact of income gaps on skill polarization is also more prominent. Therefore, both employment and education policies may need to be differentiated within these specific industry sectors in terms of implementation intensity and specific measure arrangements.