Premature Deindustrialisation and Income Inequality Dynamics: Evidence from Middle-Income Economies

Abstract The structural transformation path in most developing economies follows an employment shift towards service activities, skipping an industrialisation phase. In this paper, we explore how this premature deindustrialisation trend affects the inclusive growth trajectory of middle-income economies. Considering the trends in manufacturing employment and value-added share, we identify premature deindustrialisation phases in economies. We apply panel fixed-effects and bootstrap-corrected dynamic fixed-effects models to empirically examine the relationship between premature deindustrialisation and income inequality. Our findings suggest that income inequality rises with premature deindustrialisation if the displaced workers are absorbed into market services (especially with employment movement towards non-business market services such as trade, transport, hotels, and accommodation). In contrast, if non-market services (such as education and health) or business services (such as banking and financial services) are the dominant employment provider, it helps to reduce income inequality even in the presence of premature deindustrialisation.


Introduction
Structural transformation enhances economic growth when the labour transition occurs from low-productivity to high-productivity sectors. 1 During the process of structural transformation, economies often face a trade-off between economic growth and income inequality (Kuznets, 1955).However, recent literature on structural transformation highlights that labour transition to non-agriculture sectors increases income inequality only if the dominant employment provider is the service sector and not the manufacturing sector (Baymul & Sen, 2020;Jaumotte, Lall, & Papageorgiou, 2013;Sarma, Paul, & Wan, 2017;Sumner & London, 2017).This finding generates concern for middle-income economies, as many are experiencing a labour transition towards the service sector and a fall in manufacturing employment. 2 This shift in labour towards services before experiencing industrialisation is known as 'premature deindustrialisation' (Dasgupta & Singh, 2006;Rodrik, 2016;UNCTAD, 2003). 3In this study, we analyse whether economies can achieve inclusive structural transformation in the presence of premature deindustrialisation.
Traditionally, manufacturing expansion is regarded as key to reducing income inequality.The expansion of the manufacturing sector boosted economic growth, reducing the betweencountry inequality.Simultaneously, manufacturing also has more potential than any other sector to expand employment opportunities and, thus, decrease the within-country component of global inequality (Sumner, 2021).The structural transformation path in developing economies mostly followed an employment shift towards service activities, skipping an industrialisation phase.Alisjahbana, Kim, Sen, Sumner, and Yusuf (2022) note that service activities played a significant role in the structural transformation of all regions, whereas manufacturing played a less central role.The authors note that the expansion of business services (which includes finance, renting and other business activities) is substantial in Sub-Saharan Africa and India, and service activities in Latin America accounted for around one-half of total employment.
Consequently, the deindustrialisation and tertiarisation pattern in developing economies has implications on the level of inequality in these economies.For instance, Bhorat, Lilenstein, Osthuizen, Steenkamp, and Thornton (2022) specify that South Africa underwent a secular deindustrialisation trend, which is associated with an increase in income inequality.Similarly, Osei, Atta-Ankomah, and Lambon-Quayefio (2022) report that structural transformation in Ghana (decline in agriculture share of employment and released labour absorbed into lowproductivity service activities) can be characterised as a non-inclusive process.Sumner (2021) documents that stalled industrialisation and premature deindustrialisation created a new 'rust belt' which is threatening to raise the inequality levels in middle-income developing economies.This rise is due to the fact that deindustrialisation generates fewer better-paid manufacturing jobs, and workers move to low-wage, informal service sector activities.
To identify the premature deindustrialisation episodes in middle-income countries, we adopt a set of five conditions following Rekha and Babu (2022).Among the five conditions, the first and second filter out cases of deindustrialisation by considering the trends in both employment and value-added share of the manufacturing sector.The last three conditions distinguish 'premature' cases from general cases of deindustrialisation.According to Rodrik (2016), the 'premature' aspect of deindustrialisation occurs in two senses.First, developing economies experience deindustrialisation at considerably lower levels of income compared with advanced economies.Second, the process of deindustrialisation in the initial stages of development is premature, as the economy might thereby lose a channel via which it could have gained rapid economic growth through the manufacturing sector.Similarly, Felipe, Mehta, and Rhee (2018) specify that the inability to meet a historically derived threshold for the manufacturing share (in employment) gives early deindustrialisation its 'premature' character.Following these arguments, the last three conditions capture the 'premature' factor of deindustrialisation in terms of threshold levels of income, manufacturing employment share, and manufacturing value-added share.
We apply panel fixed-effects and bootstrap-corrected dynamic fixed-effects techniques to analyze the relationship between premature deindustrialisation and inequality in middle-income countries.For income inequality data, we depend on the World Income Inequality Database (WIID; UNU-WIDER, 2022).The International Labour Organization (ILO; ILOSTAT, 2020) and the United Nations Statistic Division (UNSD 2020) National Aggregates database provide the sectoral employment and value-added data, respectively.We also use the World Bank's (2020) World Development Indicator (WDI) database for the empirical analysis.For the analysis, we group service activities into market services (International Standard Industry Classification, Revision 4 (ISIC Rev. 4) G-M) and non-market services (ISIC Rev. 4 N-S).Non-market services include activities such as public administration, defence, education, health, and other service activities.We further classify market services into non-business market services (ISIC Rev. 4 G-I) and business market services (ISIC Rev. 4 J-M). 4Non-business market services include more informal and low-productivity activities such as trade, transport, food, and accommodation.Business market services consist of more formal and high-productivity activities such as banking, finance, information and communication, and other professional service activities.
We also consider the recently released Economic Transformation Database (2021) as an alternative data source for sectoral employment and value-added data.The Economic Transformation Database (ETD) provides a comprehensive, harmonised, longitudinal dataset across 51 countries covering employment and value-added (real and nominal) data (Kruse, Mensah, Sen, & de Vries, 2022). 5 Our empirical findings suggest that if premature deindustrialisation leads to labour absorption in non-business market services, it increases overall income inequality.In contrast, if the employment increase is in business market services or non-market services, it reduces income inequality in the economy.Hence, our study finds that the premature deindustrialisation trend hinders the inclusive growth trajectory of economies when the labour shift is towards low-productivity non-business market services.
Although much empirical work focuses on the relationship between economic growth and income inequality, except for a few studies (such as Baymul & Sen, 2020;Alisjahbana, Sen, Sumner, & Yusuf, 2022), limited research has examined how this relationship alters with changes in structural transformation in developing economies.Our study adds to this literature by providing a detailed analysis of premature deindustrialisation trends and inequality linkages in middle-income economies.Furthermore, this study offers an empirical assessment of how various service sub-sectors influence the income inequality level in these developing economies.The results show that when premature deindustrialisation is associated with a rise in non-business service activities, it adversely affects the inclusive growth trajectory of economies.The paper also employs a novel method (proposed by Rekha & Babu, 2022) that can capture the premature deindustrialisation phase in the economy in terms of both employment and valueadded share of manufacturing.Finally, the paper also draws its results from the recently released Economic Transformation Database (2021), which provides more comprehensive and consistent data on sectoral employment and value-added shares, especially for business and financial services.
The rest of the paper consists of the following sections.Section 2 explores the theoretical background of structural transformation and inequality relationships.Section 3 explains the method to detect premature deindustrialisation.Section 4 provides some stylised facts about the relationship between structural transformation and inequality in middle-income countries.Section 5 describes the empirical strategy and data.Section 6 documents the empirical findings.Section 7 provides a discussion and concludes the paper.

The relationship between structural transformation and income inequality: an overview
Economic development through structural transformation and its implication on income inequality is mainly rooted in the influential works of Lewis (1954) and Kuznets (1955Kuznets ( , 1956)).The theories postulated by these authors imply a dualistic economy model with labour productivity varying between sectors.The sectors in the dual economy model can be regarded asagriculture and non-agriculture sector, traditional and modern sector, or rural and urban sector.The transition of labour between these sectors (i.e., structural transformation) has implications for both economic development and income inequality.
For Lewis (1954Lewis ( ,1976Lewis ( , 1979)), income inequality would rise during economic development, at least up to the exhaustion of surplus labour in the traditional sector.He argued that for economic development, the capital share of income should rise so that it can be used for reinvestment, which in turn results in a declining/steady share of labour income.Consequently, income Premature deindustrialisation and income 1887 inequality may rise in the economy as it holds an inverse relationship with the labour share of income.He also noted that spatial variation in economic development in the initial stages of structural transformation could also lead to a rise in income inequality.
To represent the structural transformation-inequality trade-off, Kuznets (1955) proposed an inverted U-shaped relationship between inequality and economic growth.This relationship implies that income inequality worsens in the early phases as the economy grows and then improves in the later phases.The within-sector inequality is low in the agricultural sector and high in the urban industrial sector.In the initial phase of development, inequality increases when the rural population shifts from agriculture to industries in urban areas.This pattern is observed when growth is more likely to reward those with high skills or higher access to capital; hence, economic growth initially exhibits a pro-rich nature.However, the growth gradually becomes pro-poor over time as low-skilled workers move to higher productivity sectors and income levels.
By decomposing inequality into the between-sector and within-sector components, Anand and Kanbur (1993) provide an insight into the Kuznets process by relating widening inequality with structural transformation.The between-sector inequality depends on the share of workers in each sector, whereas the within-sector component of inequality is defined as the difference between the overall inequality and the between-sector component of inequality.The framework assumes there are only two sectors -an agricultural sector and a non-agricultural sector.In the initial phase of structural transformation, the between-sector inequality rises as the share of workers in the non-agriculture sector increases since the share of workers in the low-income agricultural sector is larger than the share of workers in the high-income non-agriculture sector.In the later phase of structural transformation, a larger proportion of the working population will shift to the high-income sector, which will cause the between-sector inequality to fall.Hence, the between-sector inequality follows an inverted U-shaped path during the process of structural transformation.
On the other hand, the behaviour of the within-sector component with respect to changes in the share of workers depends on the assumptions made regarding the within-group inequality of each sector.Following Kuznets, if one assumes that the non-agricultural sector has higher withinsector inequality than the agricultural sector, the within-sector component of inequality rises with an increase in the sector's labour share.This results in an unambiguous increase in inequality.However, after a certain level of rise in labour shares, the between-sector component dominates the within-sector component, and inequality starts to decline.This process generates the wellknown inverted U-shaped relationship between structural transformation and inequality.
Further, Baymul and Sen (2020) document that Kuznets' argument of an increase in withinsector inequality is more applicable to the services than the manufacturing sector.The presence of a large informal sector where unorganised workers are not able to access better wages and working conditions contributes to the rise in within-sector inequality in the services sector.The coexistence of formal services (such as banking and finance) and informal services (such as trade, transport, hotels and restaurants) contributes to the difference in the behaviour of within-sector inequality when the labour movement is towards the services sector.With further empirical evidence, the authors document that the movement of workers to the manufacturing sector unambiguously decreases income inequality, irrespective of the stage of structural transformation that a country is in.On the other hand, the movement of workers into the services sector has a positive impact on inequality at the early stage of structural transformation and a negative impact at a later stage.The authors conclude that Kuznets' postulate may apply more to service-driven structural transformation than to manufacturing-driven structural transformation.
In a similar vein, Sumner (2021), in his book 'Deindustrialization, distribution and Development' revisits Kuznets's seminal work and explores how deindustrialisation influences economies inequality levels.The author finds that the fragmentation of global production in a Global value chain (GVC) system is driving stalled industrialisation and premature deindustrialisation in middle-income developing countries.The deindustrialisation-led growth has exacerbated inequality within the country as workers moved from the GVC-integrated modern manufacturing sector to low-wage and low-productivity services which are predominantly informal activities.The process of tertiarisation, especially the modern financial, rent and business services, is associated with a rise in income inequality.Thus, the kind of structural transformation pattern the GVC system has engendered is likely to be associated with upward pressure on economies' income inequality.

Premature deindustrialisation: identification method
In this section, we seek to identify economies that are experiencing a premature deindustrialisation phase following Rekha and Babu (2022).Generally, the fall in either manufacturing employment share (in total employment) or manufacturing value-added share (in total output) represents the measure of deindustrialisation (for example, see Palma, 2014;Rowthorn & Ramaswamy, 1999;Tregenna, 2009).Tregenna (2013) suggests that an economy might experience a fall in manufacturing employment due to the rise in labour productivity (caused by an increase in skills and technology or labour-displacing capital intensification).Hence, a falling employment share of manufacturing, along with a similar trend in value-added share, is necessary to identify a deindustrialisation phase.For instance, China has experienced a steep fall in manufacturing employment recently, whereas its value-added share of manufacturing is still on the rise (see Appendix A3).In this case, if we consider only the fall in manufacturing employment share, we might classify China as among those economies facing premature deindustrialisation.However, Rodrik (2016) notes that deindustrialisation is more evident in terms of manufacturing employment share than in terms of value-added share (due to reliance on valueadded measures at current prices rather than constant prices).Similarly, Felipe et al. (2018) find that output shares are poor performers relative to employment shares in tracing peak industrialisation periods in economies.Thus, there exist no consensus in the literature regarding which is a better gauge for capturing deindustrialisation trend.
Similarly, there is ambiguity in defining a deindustrialisation phase as premature or not.The description of deindustrialisation as premature implies that the economy is at lower income levels and is yet to exploit the possibilities of manufacturing-driven economic growth (Felipe et al., 2018;Rodrik, 2016).Hence, to identify the 'premature' element in economies, both threshold levels of income and share of manufacturing (in employment and value-added) need to be considered.
The method suggested by Rekha and Babu (2022) addresses these concerns, and the authors propose a set of five conditions to identify premature deindustrialisation phase, as described below: Eman t 0:18 (4) where time period t ¼ 1, 2, :::, T and T is the most recent period for which the data are available.y t is the GDP per capita (in 2015 constant US dollars) in the year t: Eman tÀn, t (or

Premature deindustrialisation and income 1889
VAman tÀn, t ) represents the average share of manufacturing employment (or value-added) between years t À n and t: Similarly, Eman t, tþn (or VAman t, tþn ) indicates an average manufacturing employment (or value-added) share between year t and t þ n: VAman t and Eman t are the share of manufacturing value-added and employment in year t, respectively.This study considers n as seven years, following Rekha and Babu (2022), Eichengreen, Park, andShin (2012), Hausmann, Pritchett, andRodrik (2005) and Felipe et al. (2018).
The first condition implies that if the seven-year average manufacturing employment share in the current period (t to t þ n) is lower than in the previous period (t À n to t), the economy is experiencing a deindustrialisation phase in period t to t þ n: The second condition implies that if the seven-year average value-added share in the current period (t to t þ n) is lower than in the previous period (t À n to t), the economy is experiencing a deindustrialisation phase in period t to t þ n: The two conditions trace the deindustrialisation trend in economies in terms of both employment and value-added share of manufacturing, respectively.
The third condition states that y t (GDP per capita in 2015 constant US$) is less than $11,750.According to Felipe, Kumar, and Galope (2017), if an economy achieves an income higher than this threshold, then that economy has reached the high-income category.Hence, if deindustrialisation occurs below this income level, it signals the 'premature' element in the shrinking manufacturing share (Rekha & Babu, 2022).The fourth and fifth conditions specify that the economy in the year t should not have employment and value-added shares of manufacturing greater than 18 per cent (Rekha & Babu, 2022).This threshold implies that the manufacturing sector is not the dominant sector in the economy.
If all five conditions are satisfied, we consider an economy to be experiencing premature deindustrialisation.This method identifies that 32 economies have faced a premature deindustrialisation phase at various periods (see Table 1).Algeria, Argentina, Brazil, Jamaica, Kenya, Mexico, Nigeria, Pakistan, Peru and South Africa are some of the economies from various regions that face premature deindustrialisation.In contrast, economies such as Belarus, Bangladesh, Indonesia, Malaysia, Namibia, Sri Lanka and Vietnam are the ones that are not under the premature deindustrialisation phase.

Structural transformation and income inequality: trends and patterns
In this section, we explore trends in income inequality and how inequality is associated with the premature deindustrialisation pattern in middle-income economies.Figure 1 exhibits inequality trends in both PDI and non-PDI countries from 1992 to 2017.The trend in net Gini value depicts that both groups of economies experienced higher income inequality.Nevertheless, the PDI countries depict a persistently higher Gini value than non-PDI economies throughout the period.
From Figure 2, we observe steady growth in the employment share of services in middleincome economies.Nevertheless, the service employment share in PDI economies is much higher than in non-PDI economies.Thus, the employment trend clearly depicts the dominance of the service sector in PDI economies, with the sector providing around 60 per cent of total employment in recent periods.
Figures 3-5 explores the association between the employment share of the services sector (and its sub-sectors) and net income per capita Gini (Gini index based on income after tax and transfers) in PDI-countries (panel a) and non-PDI countries (panel b). Figure 3 depicts the association between income inequality and services employment share (out of total employment).Figures 4 and 5 provide the relationship of inequality with market services and non-market services employment share, respectively.
Figure 3 reveals that income inequality follows an inverted-U relationship with service employment share in PDI countries.In contrast, non-PDI countries have a U-shaped relationship with income inequality.This pattern implies that non-PDI countries experience a fall in inequality, whereas PDI countries are experiencing a rise in income inequality with a rise in Notes: Column 1 represents countries under premature deindustrialisation when all the five conditions are considered.Column II provides the list of countries that are not facing premature deindustrialisation when we follow all the five conditions.In coloumn II, Ã denotes countries included in the premature deindustrialisation list when only the manufacturing employment share trends are considered and not the value-added share.Source: Authors' construction based on ILO and UNSD databases.Premature deindustrialisation and income 1891 service employment share.Analyzing the inequality relationship with market service (Figure 4), we observe that in both PDI and non-PDI countries, a rise in the employment share of market services raises the inequality level.Nevertheless, the increase in inequality is more severe in PDI economies than in non-PDI economies.An increase in non-market service employment share (Figure 5) initially reduces inequality and then raises it in non-PDI countries.In economies under PDI, non-market service employment share exhibits a negative relationship with income inequality.

Empirical strategy
This paper primarily examines the implications of an early deindustrialisation trend, coupled with a service-driven labour transition, for the income inequality levels of middle-income countries.To capture this relationship, we estimate the marginal effect on net Gini (net income per  1892 R. Ravindran and M. Suresh Babu capita Gini) when service sector employment increases in the presence of premature deindustrialisation (Equation 6).
where Service and Agriculture represent the employment share of these sectors in total employment.PDI denotes the dummy variable, which takes the value 1 for those years for which we detect a premature deindustrialisation pattern in the economy.The main explanatory variables are the dummy (PDI) interaction terms with employment shares of services.To estimate the differential effects of the service sub-sectors on income inequality, in the empirical estimation, we  Premature deindustrialisation and income 1893 replace Service in Equation 6 with these sub-sectors.In this case, the main variables of interest are the interaction terms of premature deindustrialisation (PDI) and the employment share of service sub-sectors-namely, business market services, non-business market services, and nonmarket services.We use panel fixed-effects (FE) and bootstrap-corrected dynamic fixed-effects (BCFE) models to analyze the influence of deindustrialisation on income inequality.
Gini is the dependent variable to capture the income inequality level in an economy.Following Baymul and Sen (2020), we consider the net Gini coefficient as the dependent variable (based on income after taxes and transfers).The variable X represents the set of control variables, which include GDP per capita (GDPpc), trade openness (TROP), government capital expenditure (GOV), population growth rate (PPL), and educational attainment (EDN).Baymul and Sen (2020) note that per capita income affects inequality by providing more resources for redistribution.Goldberg and Pavcnik (2007) observe that trade may lower inequality by increasing the demand and wages for abundant low-skilled workers.In contrast, if trade brings technological progress that is skill biased, it can worsen income inequality levels (Feenstra & Hanson, 2003).Further, government spending is positively associated with a reduction in income inequality (Dabla-Norris, Kochhar, Suphaphiphat, Ricka, & Tsounta, 2015).Rougoor and Van Marrewijk (2015) find that difference in population growth determines inequality levels across countries.The level of educational attainment indicates human capital, and educational expansion is a major factor in reducing income inequality (Lee & Lee, 2018).

Data
The paper makes use of annual data spanning the period 1992-2017 across 54 middle-income countries.The availability of sectoral employment data constrains the selection of the sample period and countries.The sample includes middle-income countries from different regions of the world, including Asia, Latin America and the Caribbean (LAC), Sub-Saharan Africa (SSA), and Europe.In most of these economies, the services sector holds the largest share of employment, whereas industry has the largest share in terms of value-added (see A4 in the Appendix).
This study depends upon five primary sources for data assimilation-namely, WIID, ILO, UNSD, ETD and WDI.To capture income inequality, we consider net Gini (net income per capita Gini coefficient) from WIID, Revision 4. The data for sectoral employment shares are from the database of the ILO.The ILO employment dataset consists of sub-sectoral employment data based on the fourth revision of the ISIC of all economic activities (ISIC Rev. 4).However, these data are available mostly from 1992 to 2017, which acts as a binding constraint for our sample selection.Data on manufacturing value-added share and GDP per capita are from UNSD (2020).We also depend on ETD as an alternative source for employment and value added data, as the database provides structured and consistent data at sub-sectoral level.The data source for control variables-trade openness, government final consumption expenditure, the annual growth rate of the population, and educational attainment-is World Bank (2020).
We employ an unbalanced panel dataset for empirical estimation.Table 2 reports the key summary statistics of variables used in the empirical analysis.The minimum value for net Gini is 23 per cent, and the maximum is 62.4 per cent, while the mean is 41.52 per cent.These values indicate the presence of a high level of income inequality among the countries in our sample.The minimum agriculture employment share is below 1 per cent, and the maximum is at 85 per cent, respectively.In contrast, the maximum employment share of the services sector is around 85 per cent, and the minimum share is 12 per cent.The difference in the range of employment share in these sectors signals that the services sector is the dominant employment provider in the sample economies (see also Appendix A4).For instance, while the mean share of services employment is around 49 per cent, that of agriculture is only about 31 per cent and that of manufacturing 11 per cent.Within the services sector, if we compare the mean values, market services have a higher share of employment compared with non-market services.Within the market services, non-business activities have an average contribution of 22 per cent of total employment, whereas the average share of business market services is only around 5 per cent.Thus, non-business market services such as trade, transport, food, and accommodation are the dominant employment providers in our sample relative to other service activities.

Baseline framework
This section presents the regression results examining the implications of premature deindustrialisation for income inequality.In Table 3, Column I reports the panel FE estimation results of Equation 6. Column II presents regression results when we substitute employment in services with its components-employment share of market services and non-market services-in Equation 6. Column III is similar to Column II, except for the fact that in Column III we replace market services with its components-employment share of business market services and non-business market services.The Hausman specification test supports the FE model over the random-effects model; hence, in Table 3, we report panel FE estimation results in Columns I-III.
The findings in Table 3 indicate that in middle-income countries, a services-driven structural transformation initially results in an increase in inequality and then in a decrease.The linear and quadratic terms of service employment share (in Column I) are significant at the 1 per cent level and depict an inverted-U-shaped relationship with net Gini.In Column II, the employment share of market services significantly (at the 1 per cent level) affects income inequality, with the linear and square terms depicting positive and negative relationships.Similarly, the employment share of both non-business market services and business market services (the components of market services) show an inverted-U-shaped relationship with inequality (Column III).Premature deindustrialisation and income 1895 The linear and quadratic coefficients of business market services employment share are significant at the 5 per cent level, and those of non-business market services are significant at the 1 per cent level.However, non-market services' employment share has no significant influence on income inequality (in either Column II or Column III).Thus, in middle-income economies, the services sector has an inverted-U-shaped relationship with income inequality, and this relationship results primarily from an increase in the employment share of market services (both business and non-business market services) rather than non-market services.Concerning agricultural employment, we find that its coefficient is positive and significantly (at the 1 per cent level) affects income inequality.This result indicates that a labour shift away from agriculture through structural transformation is beneficial for economies in reducing income inequality.
We now consider the influence of premature deindustrialisation on inequality (Table 1), which is the primary focus of this study.With an increase in services employment in premature Notes: All explanatory variables except for the dummy variable, PDI, are in their one-period lag terms.
deindustrialisation cases, inequality reduces in the initial stage and increases at higher levels of services employment share (Column I).Though this finding contradicts our earlier results (without considering premature deindustrialisation), the sub-sectoral analysis implies that this contradiction primarily arises from the influence of non-market services.When economies experience premature deindustrialisation, the employment shares of market services (Column II) and its components-both business and non-business services (Column III)-do not significantly affect the income inequality level.However, in premature deindustrialisation episodes, the coefficient of non-market services is negative and significant at the 1 per cent level (in both Columns II and III).The coefficient of agriculture employment in deindustrialisation cases is positive and significant at the 10 per cent level in Column I and not significant in Columns II and III).Concerning the control variables in Columns I to III, GDP per capita has a positive and significant effect, while population growth rate and education have a negative and significant effect on net Gini.Both trade openness and government expenditure have no discernible impact on the level of inequality.

Bootstrap corrected dynamic fixed effects estimation framework
The literature on income inequality often reports that inequality has dynamic effects, i.e. the level of past income inequality affects the current state of inequality (Calder on & Chong, 2001;Chong, 2004).We capture this dynamic relationship by adding lagged dependent variables to the individual-effect panel model specification.To consider this dynamic effect of inequality, we are using iterative bootstrap-based bias correction for the FE estimator in dynamic panels, based on Everaert and Pozzi (2007).
From BCFE estimation (Columns I and II in Table 4), we can confirm that with premature deindustrialisation, an increase in non-market services reduces income inequality initially and increases it at a higher level of employment share.The coefficients of both the linear and quadratic terms of non-market services are significant at the 1 per cent level.In contrast, market services have a hump-shaped relationship with inequality under premature deindustrialisation (Table 4, Column I).The linear coefficient of market services is positive and significant at the 5 per cent level, and the coefficient of the square term is negative with a 5 per cent significance level.When we consider the components of market services (Table 4 Column II), the coefficient of business services' employment share has no significant effect on net Gini.However, as the employment share of non-business market services rises, the level of income inequality follows an inverted-U-shaped relationship.The coefficients of the linear and quadratic terms of nonbusiness market services are positive and negative (at the 5 per cent level of significance), respectively.
6.3.Robustness check 6.3.1.Estimation using alternative definition for premature deindustrialisation.We also check whether our results (as in Tables 3 and 4) hold when we use the alternative definition for premature deindustrialisation.In this alternative definition, only the trend in manufacturing employment share (and not value-added share, i.e. excluding condition 2 as described in Section 2) is considered for identifying premature deindustrialisation phases. 66Then, with this alternative definition of PDI, we apply both the FE and BCFE models to estimate the inequality-structural transformation relationship (Tables 5 and 6 provides the empirical results of FE and BCFE estimation, respectively).In this method, the results indicate that premature deindustrialisation significantly (at a 1 per cent level) and positively affects income inequality (Table 5, Columns I-III).This result emphasises that a premature deindustrialisation pattern can increase income inequality in middle-income economies.

Premature deindustrialisation and income 1897
The results also show that an increase in non-market services decreases income inequality (Tables 5 and 6).The findings are consistent with the earlier results, with the employment share of non-market services having a U-shaped relationship and that of market services having a positive linear relationship with income inequality (Tables 3 and 4).To sum up, the empirical findings suggest that income inequality rises with premature deindustrialisation if the displaced workers are absorbed into market services (especially with employment increase in non-business market services such as trade, transport, hotels, and accommodation activities).In contrast to this relationship, if the employment share of high-productivity non-market services increases, this helps in reducing income inequality when an economy faces premature deindustrialisation. 6.3.2.Analysis using economic transformation database (ETD).We also estimate PDI-income inequality relationship by considering the recently released alternative data source namely, Economic Transformation Database (ETD).The database provides sectoral employment and value-added details for 51 economies across regions.Of our 53 countries samples, 29 countries' data are available in the ETD database.Using the identification conditions for premature deindustrialisation (as specified in section 2), 16 countries are identified as facing premature deindustrialisation and 13 are identified as non-PDI economies.Appendix A5 provides characteristics of the ETD database, PDI classification (Table A5.1) and sectoral grouping of activities (Table A5.2) based on the ETD.
The relationship between inequality and PDI are estimated using both FE and BCFE models (Table 7).Finding from the ETD database shows that labour movement towards business market services reduce income inequality level in PDI countries.As per the ETD classification, business market services include activities under business services, financial services, and real estate activities.
To further check the robustness of the results, we also consider all 51 countries' data provided in ETD, out of which 26 are identified as facing PDI and 25 as non-PDI countries.(see Appendix Table A5.3).The results are consistent with the findings presented in Table 7 that the business market services have a negative and significant impact on the level of net Gini value given that the economy is under PDI phase.

Discussion and conclusion
Structural transformation (labour transition from one sector to another) is one of the central channels for economic development.Nevertheless, structural transformation often brings with it a trade-off between economic growth and income inequality, known as the 'developer's dilemma' or 'Kuznetsian tension' (Alisjahbana et al., 2022;Sumner & London, 2017).The recent study by Baymul and Sen (2020) shows that this dilemma (i.e. economic growth results in higher income inequality) occurs when the labour movement is services-driven rather than when it is manufacturing-driven.These findings signal that developing economies need to be more vigilant in checking inequality trends, as structural transformation is mostly service driven in these economies.The present study adds more evidence to this body of literature by analyzing the inequality-structural transformation nexus in middle-income economies.Notes: To identify premature deindustrialisation, we exclude condition 2 and include condition 1, 3, 4, and 5 as described in Section 2. Thus, we consider only trends in manufacturing employment share in this table to define the variable PDI.All explanatory variables except for the dummy variable, PDI, are in their one-period lag terms.Standard errors are in parentheses.ÃÃÃ , ÃÃ , and Ã imply p < 0.01, p < 0.05, and p < 0.1, respectively.Source: Authors' construction based on own results.

Premature deindustrialisation and income 1899
After some initial industrialisation experience, most of the developing economies are turning towards the services sector, leading to premature deindustrialisation.Applying the identification method to detect premature deindustrialisation as proposed by Rekha and Babu (2022), we observe among the 54 middle-income countries, 32 economies face premature deindustrialisation.We observe that middle-income economies, in general, experience an employment shift towards the service sector; however, this trend is more prominent in PDI countries relative to non-PDI countries.Further, the inequality trends depict that the net Gini index is relatively higher in PDI economies than in non-PDI economies.
The empirical results suggest that for middle-income countries in general, the inequality level rises in the initial stage of structural transformation with an increase in services sector employment.Baymul and Sen (2020) document a similar relationship between service employment share and income inequality.The authors find that 'Kuznetsian tension' exists when the employment shift is towards services, and the movement of workers towards manufacturing decreases income inequality.Adding to the literature, our results indicate that in the presence of premature deindustrialisation, this relationship depends on which services sub-sector acts as the dominant player in labour absorption.If the displaced workers are absorbed into non-business market services, such as trade, transport, food and accommodation, there is upward pressure on income inequality.In contrast, if non-market services, such as education, health and public administration, are the dominant employment provider, it helps to reduce income inequality in economies facing premature deindustrialisation.The ETD based analysis implies that in an economy under PDI phase labour movement towards business market services (business and financial service activities) can reduce level of income inequality.Thus, with a PDI trend, 'Kuznetsian' tension will be prevalent with an increase in non-business service employment, and the tension can be arrested with an increase in non-market service activities or business market services.
To tackle the issues of widening income-inequality developing economy primarily depends on social protection policies.Alisjahbana et al. (2022) note that any upward pressure on inequality arising out of economic development requires counter-intuitive public policies.The authors further state that Kuznets, too, argued for such policy efforts; however, excessive focus on the inverted-U curve undermined the significance of public policies (Kanbur, 2019).Analyzing the Indonesian experience, Kim, Mungsunti, Sumner, and Yusuf (2022) note that to achieve structural transformation and inclusive growth government needs a strategy to create more formal jobs through the expansion of high-productive activities.In a similar vein, Ray and Kar (2022) observe that India can achieve inclusive growth by promoting higher manufacturing growth (as the sectors absorb low-skilled workers) and increasing anti-poverty policies.Brazil experienced a decline in inequality at the beginning of this century due to the increase in formal jobs and increase in education (Firpo, Pieri, & Nogueira, 2022).Bhorat et al. (2022) state that for South Africa, upgrading industrialisation played a critical role in keeping inequality levels stable, and the period of deindustrialisation and tertiarisation witnessed a widening inequality.Furthermore, the authors identify that the challenge faced by South Africa is that there is a shortage of highly educated, high-skilled workers coupled with insufficient demand for lowskilled workers.Furthermore, Newfarmer and Page (2018) suggest that, in the case of Africa, structural transformation relies on a new set of activities known as 'industries without smokestacks' (high-value agriculture, horticulture, tourism, business services, and other tradeable services).The advantage is that, as in the manufacturing sector, these activities have both high Premature deindustrialisation and income 1901 productivity and high employment generation capacity.These studies underscore the relevance of formulating policies to upgrade industrialisation, providing formal jobs in other productive sectors, expanding education, generating a skilled workforce, and adopting anti-poverty measures.
To sum up, our results highlight that premature deindustrialisation coupled with a labour shift towards non-business service activities prove detrimental to inclusive economic development.Further, our study finds that the movement of the labour force towards non-market services or business market services is desirable to reduce income inequality in economies under the premature deindustrialisation phase.

Notes
1. Structural transformation implies the movement of labour and other resources from one sector to another during the process of economic development (McMillan, Rodrik, & Verduzco-Gallo, 2014).In the traditional pattern of structural transformation, labour movement occurs from agriculture to industry in the initial phase of economic growth.Later, as the economy reaches higher stages of development, the shift in labour occurs from industry to the service sector.Generally, advanced economies experience this pattern of structural transformation.2. The World Bank country classification (the fiscal year 2021) specifies that middle-income countries are those with a GNI (gross national income) per capita of between US$1,036 and $12,535 in 2019.High-income countries are those with a GNI per capita of $12,536 or more.Low-income countries are those with a GNI per capita of $1,035 or less.3. Deindustrialisation generally implies a fall in the manufacturing employment (or value-added) share in total employment (or total output) after reaching a peak.Hence, it follows a hump-shaped relationship with income level (Herrendorf, Rogerson, & Valentinyi, 2014).4. Appendix Tables A1 and A2 provide the aggregation of activities into sectors based on the ISIC of all economic activities, Revision 4 (Rev.4). 5.The grouping of service sector activities based on ETD classification are given in Appendix Table A5.2. 6.In this alternative definition of premature deindustrialisation, we use conditions 1, 3, 4, and 5 as described in Section 2.

Figure 2 .
Figure 2. Service sector employment share in PDI and non-PDI countries (1992-2017).Source: Author's construction based on ILO and WIID database.

Figure 3 .
Figure 3. Relationship between income inequality and service employment share (in total employment).(a) PDI countries.(b) Non-PDI countries.Source: Author's construction based on ILO and WIID database.

Figure 5 .
Figure 5. Relationship between income inequality and non-market service employment share (in total employment).(a) PDI countries.(b) Non-PDI countries.Source: Author's construction based on ILO and WIID database.

Figure 4 .
Figure 4. Relationship between income inequality and market-service employment share (in total employment).(a) PDI countries.(b) Non-PDI countries.Source: Author's construction based on ILO and WIID database.

Table 2 .
Summary statistics.Net Gini is the net income per capita Gini after taxes and transfers.EMAN, ESRV, EAGR, Emkt-srv, Enonmkt-srv, Ebus-srv, and Enonbus-srv are the employment shares of manufacturing, services, agriculture, market services, non-market services, business market services, and non-business market services in total employment, respectively.GDPpc represents the variable GDP per capita in log form.TROP denotes the sum of exports and imports of goods and services as a share of the GDP (in percentage) .GOV is the annual percentage growth rate of government final consumption expenditure.PPL is the annual population growth rate.Education is the percentage of the population that has at least com- pleted upper secondary education.Source: Authors' construction based on ILOSTAT (2020); UNSD (2020); UNU-WIDER (2022); World Bank (2020).

Table 3 .
Relationship between premature deindustrialisation and income inequality: Panel fixed effects results.

Table 4 .
Relationship between premature deindustrialisation and income inequality: Panel bootstrap corrected fixed effects results.This table reports the results of bootstrap correct dynamic fixed effects estimation.All explanatory variables except for the dummy variable, PDI, are in their one-period lag terms.ÃÃÃ , ÃÃ , and Ã imply p < 0.01, p < 0.05, and p < 0.1, respectively.Standard errors are in parentheses.The employment shares of services and its sub-sectors in total employment are in both linear and quadratic terms.Source: Authors' construction based on own results.

Table 5 .
Panel fixed effects regression results with alternative identification of premature deindustrialisation.

Table 6 .
BCFE regression results with alternative identification of premature deindustrialisation.
Notes: In this table the dummy variable PDI is constructed considering all conditions in section 2, except for condition 2. All explanatory variables except for the dummy variable, PDI, are in their one-period lag terms.ÃÃÃ , ÃÃ , and Ã imply p < 0.01, p < 0.05, and p < 0.1, respectively.Standard errors are in parentheses.The employment shares of services and its sub-sectors in total employment are in both linear and quadratic terms.Source: Authors' construction based on own results.

Table 7 .
Relationship between PDI and income inequality: ETD based analysis.This table reports the results of analysing structural transformation and inequality relationship using Economic Transformation Database.Column I and II reports FE and BCFE estimation results, respectively, for 29 middle-income economies that are common in both ILO/UNSD and ETD data.All explanatory variables except for the dummy variable, PDI, are in their one-period lag terms.ÃÃÃ , ÃÃ , and Ã imply p < 0.01, p < 0.05, and p < 0.1, respectively.Standard errors are in parentheses.The employment shares of services and its sub-sectors in total employment are in both linear and quadratic terms.Source: Authors' construction based on own results.