Quantifying Subnational Economic Complexity: Evidence from Romania

Abstract

Over the last two decades, Romania has undergone strong economic growth, catching up to advanced economies and producing one of the best economic performances in the world. Along with these positive changes, industrial product diversification has increased through expanded foreign economic relations, with all of these supporting the complexity of economic activities. Even though there is a world ranking of countries showing the highest level of economic complexity, there is no information about regional contributions to the overall score in Romania. This paper fills this gap by measuring the economic complexity of Romania’s subnational areas (counties) in the last ten years. To calculate the Economic Complexity Index (ECI) at a regional level, 615 economic activities (four-digit classes according to the NACE classification) were taken into consideration, grouped into 68 cluster categories. The results show that significant changes in economic complexity have mainly occurred in less developed areas, the upper- and middle-ranked counties’ positions being relatively stable. Furthermore, we examined the impact of complexity on growth and convergence, finding that the ECI index is a good predictor both for future economic growth as well as for the evolution of income inequalities.

1. Introduction

In recent years, there has been a shift from traditional macro-economic indicators to data-driven approaches in order to measure the complexity of economies and future economic growth [1,2]. One of the central elements for understanding and modeling the complexity of the economy is knowledge-based production [3]. Several studies have used different aspects of intangible capital for measuring knowledge creation and the competitiveness of countries, such as the number of patents, investment flows, or international trade flows, since economic complexity depends on a country’s productive structure [4,5,6,7]. These works have highlighted that the more developed a country, the more likely it is to be characterized by high diversification, thus producing less common goods. The most pioneering work in the field belongs to Hidalgo and Hausmann [3], who developed the Economic Complexity Index by using international trade flows data. They measured the economic diversity of certain countries (the number of exported products with revealed comparative advantage) as well as the ubiquity of the specific products (the number of countries that export that product) based on a linear algebra calculation [3].

The relationship between the economic structure of a state, its economic growth, and income inequalities has long been recognized within the scientific literature [8,9,10]. Unlike traditional approaches to economic growth, usually focusing on labor, capital, and technology (knowledge), the economic complexity method analyzes thousands of economic activities in order to explain the economic trajectories of countries, regions, and cities [11]. The main idea behind economic complexity is to look into productive capabilities (inputs, technologies, ideas) as key drivers of economic development. According to Hidalgo and Hausmann [3], productive capabilities determine the quality and number of products a country can export. As the number of exported products depends on the diversity of a country’s economy, the Economic Complexity Index can also predict future economic growth and income inequality (usually a country exporting more economically complex products has a higher GDP per capita and lower levels of income inequality) [3,11,12,13]. Economic complexity measures not only the economic development of a state, but it can also be seen as a proxy for social development, since producing sophisticated products is highly dependent on interactions on a social and professional level within a state [5,14,15]. Hausmann et al. [16] compared the effects of economic complexity on growth using some of the World Governance Indicators (institutional quality, human capital, and competitiveness) and came to the conclusion that ECI is a better measure for explaining growth than WGI indicators.

As we have observed, this method has primarily been used both in social and economic contexts, including economic growth [1,3,17], population growth of cities [18], the business cycle [19], human development [5,20], income inequality [21,22,23], the labor market and labor productivity [24,25,26], foreign direct investment [27,28], and sustainability [29,30,31,32,33] being applied to different geographical scales. Thus, in recent years there has been a shift towards analyzing economic complexity from global to subnational levels, based on different regional databases: US [7,18,34], Mexico [19,35], China [1], Brazil [36], Australia [37,38], Italy [39], New Zealand [40], UK [41], and Romania [30]. For the applicability of economic complexity and as a literature review, see Hidalgo [11].

This paper goes beyond the existing literature and investigates the relationship between economic complexity, regional economic growth, and income inequality. There is a large consensus among researchers that economic complexity leads to regional growth and development, although there is still a debate on the effects of complexity on income inequalities. So far, most of the studies have concentrated on the fact that a country’s productive structure positively influences the existing income gap. According to Hartmann [5], economic diversity has a major impact on human development, because it stimulates people’s capacity and influences education, health, and infrastructure indicators. Nevertheless, spending on research and development could further boost the innovation capacity of a society. This also explains why in general large urban agglomerations—with a pool of skilled labor, social capital, and decent infrastructure—act as engines of growth and are usually more competitive than small- and medium-sized cities where the available knowledge for the diversification of goods and innovation is narrowed due to the absence of adequate human capital and institutional capacities. Hartmann et al. [21,22] also affirm that higher levels of economic complexity result in decreased income inequalities. The same conclusion is found in the work of Milanovic [42] through analysis of global income inequalities as well as in some recent regional analyses [1,7,35]. Using hierarchical linear modeling in order to analyze 87 developing countries from 1990 to 2017, Le Caous and Huarng [43] also found a negative correlation between economic complexity and income inequality. However, as they concluded, the used model could not be applied to developed countries. Bandeira-Morais et al. [44], by applying a non-linear relationship between the ECI and income inequality for Brazilian states, has confirmed the inverted U-shape relation by proving that countries with a low and a high ECI have lower income inequality than those with a moderate income per capita.

On the contrary, some papers argue that the growth of economic complexity increases inequality as only a few regions enjoy high levels of complexity, leaving many other regions stuck with low levels of complexity and relative incomes [45]. Using the Standardized World Income Inequality Database and comparing the OLS regression and GMM estimator, Lee and Vu [46] revealed that an increase in economic complexity contributes to higher inequality according to the GMM model, which contradicts the cross-country OLS regression results. According to this, income inequality is higher in well-developed countries while an improved economic structure contributes to the mitigation of the income gap. Lee and Wang [47], looking at country risk and applying a finite-mixture model to 43 countries between 1991–2016, revealed that in countries with low economic, financial, and political risk, an increase in the ECI is associated with more equal income distribution. In contrast, an improvement in economic structure does not lead to a decrease in income inequality in countries characterized by high risk. Using panel data and analyzing 88 countries between 2002 and 2017, Chu and Hoang [48] concluded that economic complexity is associated with higher income inequality mainly in countries where the economic openness and educational level are low and government spending is ineffective. Our analysis also shows that the diversification of the economic structure and the increase in economic complexity have a positive impact on future economic growth; however, in the case of Romania, they have a major effect on the increase in existing inequalities. All these approaches illustrate that the effect of economic complexity on income inequality depends on the development level of the analyzed countries, on the applied statistical model, and on the database used for measuring the productive capability of a state.

Although the ECI has numerous critics due its theoretical and mathematical formulations [1,49,50], it is still one of the most reliable and complex indexes for measuring the economic structure and diversification of a country’s economy. The main criticism is towards the Method of Reflection, claiming that this formula underestimates some important nations with high diversification in exports [49,51,52]. Some authors have proposed a non-linear method to weigh the complexity of the products, offering new methods and models for measuring the economic fitness of countries (i.e., competitiveness) and economic diversity [1,49,50,53]. However, most of the authors found a strong similarity between the Economic Complexity Index and the Fitness Index, both having a strong explanatory power [1,34]. Recent studies have also demonstrated that the existing Hidalgo–Hausmann approach cannot be applied to subnational regional economies in developed countries; therefore, they developed a place-specific index, which is better suited for regional and provincial data [39]. The same authors have also highlighted that even in the case of more developed economies, the main arguments behind complexity remain valid on a subnational scale. Davies and Maré [40], measuring the effects of relatedness and complexity of urban employment growth, realized that ECI is effective in large cities but does not contribute to employment growth in smaller cities. This also illustrates the importance of using appropriate measures adapted to the analyzed contexts.

In this paper, we conduct an empirical analysis by applying Hidalgo and Hausmann’s method [4] to determine the economic complexity of Romania’s counties in the 2008–2018 period and to obtain further information about the impact of economic complexity on regional inequality. The main hypothesis of our study is that—following the transition from a planned socialist economy to a market economy—the macroeconomic stabilization has led to an increase in employment and economic complexity in some areas, which in turn has a positive impact on regional economic growth. We assume also that the increase in economic complexity shows significant regional differences, leading to different regional development paths, although in the long run, economic growth is expected to contribute to an increase in regional income inequalities (Figure 1).

This study is novel from several points of view. Firstly, Romania represents a particular case study because it has registered a strong economic growth and convergence in the last decades, reducing the development gap to the EU level measured in GDP/capita (PPS) as a percentage of the EU27 average from 26% in 2000 to 70% in 2019 [54]. The annual GDP growth rate has even exceeded the EU average, reaching 7.9% in 2017. For this reason, in 2019 the World Bank classified Romania as a high-income country (the GDP/capita in 2019 was $12,630) [55]. In this study, we assume that the above-mentioned intense economic development in Romania can be explained by an increase in economic complexity.

Secondly, by combining different statistical sources, we focus the analysis on the county level, assessing the different contributions of the NUTS-3-level regions to the economic diversification of the county. Ferraz et al. [20] made a literature review by analyzing 374 scientific articles between 1988 and 2020 and revealed that one of the research gaps in studying complexity is that relatively few studies have concentrated on analyzing regional data within a state. Therefore, we hope to further contribute to the literature by examining the relation between economic complexity and income inequality in very different socio-spatial contexts: from peripheries and semi-peripheries to core economic areas.

Thirdly, the analysis offers a policy-relevant overview of the most and least complex industries, which could serve as a basic measure for selecting and supporting certain economic activities, a crucial challenge in setting the right smart specialization strategies in peripheral counties.

The paper is structured as follows. In the following section, we present a theoretical background as well as the data and method used. Section 4 compares the calculated ECI for 615 different globally and locally competitive sectors according to the industry-standard classification system used in the European Union at the four-digit level (NACE-4 Rev. 2). This section also presents a multivariate regression analysis in order to explore the effects of complexity on inequality and human capital. Finally, the last subsection presents the conclusions and policy implications.

2. Study Area

Considering the territorial administrative units, Romania is divided into 8 development regions (NUTS-2), 42 counties (NUTS-3), and 3181 local administrative units (LAU) including communes, towns, and cities. In this study, we analyzed economic complexity at the county level. We used this spatial unit for two reasons: firstly, most of the data needed for this analysis is not available at smaller territorial levels, and secondly—in a future study—it allows for a better comparison with other European NUTS-3 level territorial units.

As mentioned in Section 1, Romania has registered an economic growth during the last decade and a strong catch-up to the EU average GDP/capita level (Figure 2). However, the downside of these positive processes was an intense increase in internal regional inequalities [56,57]. Nonetheless, Romania has a twofold economy: an urban, dynamic one, represented mainly by big cities, well integrated into the EU spatial and economic structure, which are places of prosperity and have the role of innovation hubs (such as the capital, Bucharest, Cluj, Timisoara, Iasi). The other side is represented—with a few exceptions—by small towns and rural areas, struggling due to high poverty rates, a lack of infrastructure, and low educational levels [58].

The most important economic sector is services, employing around 49% of the country’s workforce and accounting for 58.2% of the GDP [54]. Tourism has an especially direct contribution to GDP, although the IT sector has become stronger in the last decades due to the highly-qualified workforce and the labor costs which are well below the EU average, accounting for around 6% of the GDP per capita [55]. In the analyzed period, the growth of the industrial sector has also had a major impact on the country’s economic growth, especially the production of vehicle components, textiles, building and construction, and petroleum refining—these being just a few of the main manufacturing activities. Nowadays, industry employs 30% of the active population and contributes around 28% to the country’s GDP [55]. Even though agriculture represents the lowest share of GDP (4.1%), the number of employees is relatively high (21%) [55]. Due to low mechanization, the level of agricultural output remains at a shallow level considering the country’s resources and agricultural potential.

As reported by the Atlas of Economic Complexity, Romania ranks 22nd, becoming somewhat more complex than a decade ago [16]. The main sectors responsible for Romania’s exports are transports (8.3%), parts of motor vehicles (6.4%), cars (4.8%), ICT—Information and Communication Technology (4.3%), and tourism (3.3%). Romania is more economically diverse than expected for its income level, so there are forecasts that the economy will grow moderately in the coming years; growth projections until 2028 predict an annual increase of 3.0% over the next decade [16].

3. Data and Method

3.1. Data Collection

In most countries, subnational trade databases are not officially available; therefore, researchers have used different datasets to calculate economic complexity. Balland and Rigby [7] measured the complexity of US cities by using a patent database while Chávez et al. [35] measured economic complexity and regional growth performance in Mexico by applying data on the employed population. The use of employment data can also be found in the works of Cicerone et al. [39], who analyzed the network connectedness and centrality of exports in certain Italian provinces, of Davies and Maré [40], who analyzed employment growth in New Zealand’s largest cities, and of Fritz and Manduca [34], who analyzed the economic complexity of US metropolitan areas. Reynolds et al. [37] generated a dataset using multiregional input–output export tables for Australian subnational regions while Gao and Zhou [1] calculated the economic complexity of Chinese provinces using firm-level data. Not least, Balsalobre et al. [38] used bilateral trade flows for Spanish provinces.

To calculate economic complexity at county level, we used company data obtained from the National Trade Register Office for the 2008–2018 period [59]. We started our analysis from the year 2008, as this is the first year for which we have such a detailed database. This database contains information about the number of companies, the number of employees, and thet turnover for all four-digit NACE Rev. 2 codes (Nomenclature statistique des Activités Économiques dans la Communauté Européenne—Statistical Classification of Economic Activities in the European Community) [60]. NACE codes are the industry standard classification system used in the European Union according to Regulation No 1893/2006, which is also comparable at a global level. To focus mainly on innovative economies, we grouped the existing 615 NACE codes according to the methodology developed by the authors of the 2020 European Cluster Observatory (European Cluster Panorama and Industrial Change methodology) [60]. The above document suggests the delimitation of 51 traded cluster categories (exporting industries), which compete on a global scale and have high levels of innovation and productive knowledge (See Appendix A 

Table A2. Cluster categories.

Global Industries Clusters
1	Aerospace Vehicles and Defense	27	Leather and Related Products
2	Agricultural Inputs and Services	28	Lighting and Electrical Equipment
3	Apparel	29	Livestock Processing
4	Appliances	30	Marketing, Design, and Publishing
5	Automotive	31	Medical Devices
6	Biopharmaceuticals	32	Metal Mining
7	Business Services	33	Metalworking Technology
8	Coal Mining	34	Music and Sound Recording
9	Communications Equipment and Services	35	Non-metal mining
10	Construction Products and Services	36	Oil and Gas Production and Transportation
11	Distribution and Electronic Commerce	37	Paper and Packaging
12	Downstream Chemical Products	38	Performing Arts
13	Downstream Metal Products	39	Plastics
14	Education and Knowledge Creation	40	Printing Services
15	Electric Power Generation and Transmission	41	Production Technology and Heavy Machinery
16	Environmental Services	42	Recreational and Small Electric Goods
17	Financial Services	43	Textile Manufacturing
18	Fishing and Fishing Products	44	Tobacco
19	Food Processing and Manufacturing	45	Transportation and Logistics
20	Footwear	46	Upstream Chemical Products
21	Forestry	47	Upstream Metal Manufacturing
22	Furniture	48	Video Production and Distribution
23	Hospitality and Tourism	49	Vulcanized and Fired Materials
24	Information Technology and Analytical Instruments	50	Water Transportation
25	Insurance Services	51	Wood Products
26	Jewelry and Precious Metals
Local industries clusters
1	Agriculture, Forestry and Fishing	10	Financial and Insurance Activities
2	Manufacturing	11	Real Estate Activities
3	Electricity, Gas, Steam and Air Conditioning Supply	12	Professional, Scientific and Technical Activities
4	Water Supply; Sewerage, Waste Management and Remediation Activities	13	Administrative and Support Service Activities
5	Construction	14	Public Administration and Defense; Compulsory Social Security
6	Wholesale and Retail Trade; Repair of Motor Vehicles and Motorcycles	15	Education
7	Transportation and Storage	16	Human Health and Social Work Activities
8	Accommodation and Food Service Activities	17	Arts, Entertainment and Recreation
9	Information and Communication

). It must be mentioned that the European Cluster Observatory and the 2019 Panorama report use the notion of “traded industries”, the concept originally introduced in the US cluster-mapping exercise. The term “exporting industries” is used in the 2020 Panorama report, however both refer to the same industries [60].

The cluster categories were defined according to specialization (by calculating the Location Quotient of employment), cluster size (considering the number of employees), employee productivity (by using the average wages), SME performance (by looking into high growth enterprises), and innovation leaders (mainly focusing on global frontier enterprises). The remaining economic activities (17 cluster categories in total) are mainly of local importance and are usually present in most regions, although they still have a high potential for development, being able to generate economic growth that radiates beyond the area of influence of those territories with the support of local actors. The use of grouped economic sectors, i.e., traded and local industries, also appears in the research of Fritz and Manduca [34], analyzing the complexity of US metropolitan areas.

Since in the case of Romania, detailed international trade data at county and settlements level are not available (just for some specific NACE categories for the export of industrial products, but without services), we used the number of employed population, acting as a good proxy for calculating the economic complexity of Romanian counties. We also found a strong correlation (r = 0.562) between the Economic Complexity Index calculated with company data for different occupations and export data for industrial products. All in all, the mentioned 51 global and 17 local cluster categories encompass 615 four-digit different economic activities (sectors) and more than 700,000 companies. We consider that these data are highly appropriate for describing the economic structure of the different regions.

Secondly, for capturing the relation between the Economic Complexity Index and the GDP growth or the human capital (data related to population, schooling, life expectancy at birth, unemployment rate, and income) we also used classical statistical sources: Eurostat (Luxembourg) [54], the National Institute of Statistics (Uchisaiwaicho, Japan) [61], the 2011 Population and Dwellings Census [62], and the Territorial Observatory [63].

The third major source of information is represented by one central government institution providing data on public revenue and public expenses, namely the Ministry of Regional Development and Public Administration [64].

3.2. Methods for Analyzing the Economic Complexity Index

The Economic Complexity Index and Product Complexity Index are good instruments to measure the different economic activities each region specializes in, simultaneously giving insights into the more and less developed areas. In this sense, the ECI also has good predicting power for future economic growth [65]. To calculate the Economic Complexity Index, we applied a three-step method [3,35] and we used the final index to determine the economic complexity of the Romanian counties.

3.2.1. Defining the Revealed Comparative Advantage of Counties

First, we calculated the Revealed Comparative Advantage (RCA) to determine the specialization pattern of each county. For this, we used the mathematical measure known as the Location Quotient or Balassa index [66]. Location Quotients analyze the concentration of industrial employment in a specific region (Equation (1)):

$${\mathrm{LQ}}_{s,a} = \frac{\frac{P_{s,a}}{{\sum }_{a=1}^n{p}_{s,a}}}{{\sum }_{s=1, a=1}^{s=42, a=n}{p}_{s,a}}$$

(1)

where p_s,a is the number of people working in county s in economic activity a; $\sum _a^n=1p_{s,a}$ represents the total number of people employed in county s; $\sum _{s=1}^{42}{p}_{s,a}$ is the total number of people employed in economic activity a throughout the country; and $\sum _{s=1}^{42}\sum _{a=1}^n{p}_{s,a}$ represents the total number of people employed countrywide [35]. Afterwards, this is transformed into a contiguity matrix (m_s,a) where m_s,a = 1 if the LQ is above a certain threshold (in our case the threshold of LQ = 1). This means that a county has a revealed comparative advantage or is specialized in a certain economic activity if the calculated index is equal to or greater than 1 (Equation (2)).

$$m_{s,a}=\{\begin{array}{c}1 if L{Q}_{s,a}\ge LQ\ast \\ 0 otherwise \end{array}$$

(2)

The Location Quotient is one of the most used ways for quantifying the concentration of economic (industry, occupation) or demographic groups (population) in a region compared to the (usually) national average. In this way, certain distinctive characteristics of a region can be identified, i.e., the strengths and weaknesses which give an insight on the tendencies to prosper. Thus, if the employment rate is higher in an industry than the overall national average, it means that the respective industry has a competitive advantage in that regional economy. To conclude, the Location Quotient or Balassa index allows the identification of flourishing industries, also influencing future business decisions [67].

3.2.2. Applying the Method of Reflection (MR) Technique

One of the most important aspects of economic complexity is given by the differences in size between national economies. The different inputs, which determine the mix of outputs, are responsible for a country’s productive capabilities, and hence for its economic development. As already mentioned, productive capabilities can be infrastructure, equipment, people, collective knowledge: in other words, all the inputs, technologies, or ideas [3]. The more diverse a country’s economy, the more developed the underlying productive capabilities. In this way, the ECI measures the productive capabilities of a country indirectly, taking into consideration the exported mix of products. Theoretically, the ECI divides a country’s economics into two dimensions: the diversity and ubiquity of products in the export basket. If diversity refers to the number of products that a country can export with a competitive advantage, the ubiquity represents the number of countries which export a product competitively. These two dimensions can be measured with a cross-country export matrix (countries being in the rows and product categories in the columns); in this way, each cell in the table represents the value of country–product exports. Thus, at the top of the ECI hierarchy we can find countries which export diverse types of products, and these products are not found in many other countries (i.e., they represent a high ubiquity in the export baskets) [68].

Therefore, in the second step, the defined m_s,a matrix (Equation (2)) calculates the counties’ economic diversity and the ubiquity of the respective products. The economic diversity of a county is determined by the number of products with a high RCA [21]. The ubiquity of economic activity shows the number of counties that are specialized in a product, i.e., they have the advantage of exporting the respective product (Equations (3) and (4)).

$$\mathrm{Diversity}: K_{s,0}={\displaystyle \sum _{a=1}^n}{m}_{s,a}$$

(3)

$$\mathrm{Ubiquity}: K_{a,0}={\displaystyle \sum _{s=1}^{42}}{m}_{s,a}$$

(4)

The MR consists of a sequential combination of the measures of diversity and ubiquity over an n number of iterations. Hidalgo and Hausmann [3] suggest that a value of n = 12 iterations is large enough to achieve a convergence process. In our analysis we have used more than 12 iterations until the point where no further information could be extracted, i.e., when the $K_{s, N}$ variable has not changed for three consecutive iterations [35] (Equations (5) and (6)).

$$K_{s, N}=\frac{1}{K_{s,0}}{\displaystyle \sum }_{a=1}^n{m}_{s,a}\ast K_{a,N-1}$$

(5)

$$K_{a, N}=\frac{1}{K_{a,0}}{\displaystyle \sum }_{s=1}^{42}{m}_{s,a}\ast K_{s,N-1}$$

(6)

Thus, the complexity of a county’s economy is given by the variety of exported products with a comparative advantage (high diversity), while for a product to be considered complex, it is essential for it not to be exported on a large scale by many counties (low ubiquity) [23].

3.2.3. Calculation of the Economic Complexity Index

Since the ECI is a relative metric, in the last step we applied a normalization process using the Z-score transformation as follows (Equation (7)):

$$ECI=\frac{K_c-\langle K\rangle }{std\left(K\right)}$$

(7)

where $\langle K\rangle$ and std (K) are the mean and standard deviation of vector Kc.

Similarly, the Product Complexity Index is defined as the ECI from the previous definition, i.e., by simply changing the counties’ index with that of the products [16] (Equation (8)).

$$PCI=\frac{Q_c-\langle Q\rangle }{std\left(Q\right)}$$

(8)

Thus, the more diverse the scale of exported products and the less ubiquitous the goods, the more complex the economic structure of a county. Therefore, the measurement of the ECI using diversity and ubiquity can deeply explain the distribution of productive knowledge among counties.

3.2.4. Applying the Multivariate Regression Analysis

To get a more holistic picture on the effects of complexity on economic growth, we performed a multivariate regression analysis for the year 2018 and then for the period 2008–2018 by using panel data (Equation (9)). In both models, we included human-capital-related factors for examining the relation between economic complexity and income inequality when introducing other variables.

The OLS regression provides a global model of the predicted variable. It is the most commonly used statistical procedure to evaluate the relationship between two or more variables. The linear regression model is described mathematically in the following form:

Y = β₀ + β₁X_i1 + β₂X_i2 + ………β_kX_ik + ε

(9)

where Y represents the dependent variable; x represents the independent variable; β₀ is the so-called intercept of the model; β₁, β₂, β_k are the coefficients of variable X_i; and ε is the residual (error term). Thus, our multivariate regression model can be described with the following equation:

$$GIN{I}_i=\propto +{\beta }_1·EC{I}_i+{\beta }_2·LE{B}_i+{\beta }_3·SGE{R}_i+{\beta }_4·\log {\left(GDPpc\right)}_i+{\beta }_5·UR{B}_i+{\beta }_{6}E{R}_i+{ϵ}_i$$

(10)

The selected human-capital-related variables in this study are based on the well-known Human Development Index, including log GDP per capita (logGDPpc), gross (school) enrolment ratio (GER), and life expectancy at birth (LEB), to which we also added the urbanization rate (URB) and the unemployment rate (UR).

As we had a large database, we also used the panel estimates technique. In the case of panel data, the cross-sectional and time dimensions are present together in the data set: the entire sample contains a time series for each cross-sectional observation. For the panel data model we used the following equation:

Y_it = β₀ + β₁X₁,_it + …… + β_kX_k,_it + y₂E₂ + …+ y_nE_n + δ₂T₂+ …… + δ_tT_t + ε_it

(11)

where Y_it represents the dependent variable, where I = entity and t = time; X represents the independent variable; X_k,it represents the independent variable; β_k is the coefficient for the independent variables; ε is the error term; E_n is the entity n; y₂ is the coefficient for the binary regressors; T_t is time as dummy variable; and δ_t is the coefficient for the binary time regressors [69].

The most important characteristic that distinguishes a panel database from a pooled cross-sectional sample is that we collect data from the same observation units in each period. Due to the additional information generated in this way, the panel data structure is also suitable for answering questions that would be impossible to answer using the combined cross-sectional sample. In this context, we would like to highlight two advantages of the panel data structure. On the one hand, panel data allows us to control for variables, which cannot be observed or measured. On the other hand, it allows controlling for variables that change over time but not across entities [69]. The essence behind the panel data model is heterogeneity and resulting endogeneity, which are important for understanding the use of the panel data concept. The unobserved heterogeneity usually appears when there is an unobserved dependency of other independent variables, while the endogeneity exists when there is a correlation between the independent variable(s) and the error term [69]. There are two important techniques for controlling heterogeneity using panel data: fixed effects (FE) and random effects (RE). The fixed effects model is commonly used for analyzing the impact of variables that change over time. When using FE, we assume that a phenomenon within the analyzed entities may influence or bias the outcome variables and that this must be controlled. In the case of RE, the variation across entities is supposed to be random and uncorrelated with the independent variable included in the analysis [69]. In order to decide which model is appropriate, the Hausman test is usually performed, where the null hypothesis means that the preferred model represents the random effects and the alternative represents the fixed effects. Essentially, it tests whether the unique errors are correlated with the regressors; in other words, the null hypothesis states that there is no correlation between the two [69].

4. Results and Discussion

In this section, we will focus on the general assessment of subnational economic complexity, then link complexity to data on development and inequality to show the catching-up process, starting from the basic idea that in the long run, a region converges to an income group that is in sync with its productive knowledge base.

4.1. Measuring Subnational Economic Complexity (ECI)

Economic complexity is a novel measurement of the knowledge-based economic structure of a country, thus offering a holistic picture of a region’s diversity and industrial sophistication. Usually, the more diverse and less ubiquitous the goods a country produces, the more sophisticated its economy. In this sense, diversity is inversely correlated with ubiquity [70]. Figure 3A presents the territorial distribution of the Economic Complexity Index in Romania at the NUTS-3 level in 2018. As we can observe, a clear spatial pattern cannot be distinguished between the territorial concentration of the highest and lowest EC values. Similar to the GDP per capita, the country is characterized by a mosaic-like structure [71]; however, there is a strong similarity between the development level of a county and the complexity of its economic structure. Thus, a higher economic complexity characterizes the central and western parts of the country while the south-east and eastern areas stand out as having less diverse economies.

Figure 3B shows the relation between the country’s diversity and the ubiquity of economic activities in which the respective counties have comparative advantages. We found a negative correlation between the two variables, strengthening the basic idea that diversified counties turn out less ubiquitous goods and services [35]. More diversified economies, which are specialized in less ubiquitous activities, are usually the characteristic of well-developed counties (lower-right quadrant) while a lack of diversity and ubiquitous economic activities are mainly found in the less developed areas (upper-left quadrant). This is also illustrated by the spatial autocorrelation statistic: while in 2008 the value of Moran’s I was 0.169, in 2013 it reached 0.247, and in 2018 it was already 0.305. This suggests a relatively high similarity among neighboring counties.

Further on, we also tested the difference in the evolution between 2008 and 2018 for three different periods by applying Spearman’s rank correlation. The correlation between 2008 and 2013 is 0.80, between 2013 and 2018 it is 0.93, while between 2008 and 2018 it is 0.81 (Figure 3C). The relatively high difference between the correlations in the periods mentioned above is strongly influenced by the effects of the global economic crisis which negatively affected the companies and the labor market. This is the case for some less developed counties which underwent a consistent drop as a result of closing down some major industrial plants (such as Mehedinti county (MH), which specialized in upstream chemical products, mainly uranium mining, or Tulcea county (TL), which specialized in fishing and fishing products).

Some areas, however, have managed to adapt well to the new economic situation and adjust to the changes in demand. Thus, Gorj county (GJ), occupying the penultimate place in 2008, significantly improved its initial position by merging several companies based on the production and transmission of electricity in 2011. In 2018 the resulting company became one of the best performing firms in the county, specializing in electric power generation and transmission with more than thousand employees. Other examples are Olt (OT) and Dâmbovita (DB) counties, both improving almost ten positions compared to 2008 due to an increase in employment in upstream metal manufacturing and in the production of electrical equipment, respectively (in Dâmbovita we can find one of the country’s oldest companies, Arctic, specialized in the manufacture of electric domestic appliances). However, this significant change—mainly in the case of Olt county—has not led to a diversification in the economic structure, these two counties still being labeled as having some of the less complex economies in Romania.

It must be mentioned that in the case of middle-ranked counties there were just minor relocations, as is the case for the high-ranked counties as well, showing a high stability over time. In this category we can find three of the most complex economies of Romania, namely Bucharest (B), Ilfov (IF), and Cluj (CJ). This is also illustrated by the correlation between the 2008 and 2018 rankings, namely a stable relationship between the highest- and middle-ranked counties, the main changes over time being observed in the case of the lowest positioned areas (Figure 3D).

4.2. Measuring Product Complexity (PCI)

We can get more information about the difference between counties’ profiles by looking at the PCI measures. Figure 4 shows the most and least complex industries based on 2018 data (see also Appendix A Figure A1). It can be stated that the most complex sectors are those requiring highly skilled professionals, information-related and financial sectors, which are usually concentrated in well-developed areas, while the bottom-ranked sectors are mainly related to manufacturing activities. Interestingly, there are no apparent differences between the territorial rankings of the global and local industries, i.e., among the economic activities placed in the top 10 there are also two industries considered local (transportation and storage as well as real estate activities), while at the same time the bottom 10 ranked activities are all part of global industries.

According to our results, one of the most important local sectors in Romania is transportation and storage services. This is also highlighted by the Atlas of Economic Complexity [16], as transportation contributes 7.4% to the country’s total exports. Another important local industry in Romania is strongly related to real estate activities, as part of the massive residential construction projects and a strong response to the housing policy of the communist regime after the transition period. The Global Financial Crisis beginning in Romania in the second part of the year 2008 has not influenced the construction sectors significantly in general; due to the high demand mainly in the big urban centers, we witnessed a strong increase even after this period. All this illustrates that some local economic activities could not only support the development of inner markets but could also favor higher economic growth and competitiveness (Figure 4).

According to the PCI, we also found some emerging industries which improved their position in 2018 compared to 2008. These are primarily group companies belonging to the electric power generation and transmission- (production and transmission of electricity) or to the marketing, design, and publishing cluster category (especially the design and elaboration of web portals).

The economic activities with the least amount of complexity are strongly related to heavy or light industries, which have experienced the biggest drop in the last two decades. This is mainly due to low production capacities and low technological innovation (in the case of heavy industry) or to increasing labor costs because of rising minimum wages (in the case of the light industry). The only exception in this sense is the cluster category related to water transportation, due to the sharp decrease in the number of employees in sectors such as building of ships and floating structures or inland passenger and inland freight water transport, which characterizes mostly Giurgiu (GR) and Mehedinti (MH) counties.

Observably, a country’s productive structure is strongly related to its general level of development. Therefore, in the following subsection, we will explore the nexus between these factors more deeply.

4.3. The Relation between Economic Complexity and Income Inequality

There is a large amount of literature studying the connection between economic growth and income inequality [56,71,72,73,74,75,76,77,78]. The impact of inequalities on economic growth has usually been analyzed with the help of cross-sectional growth regression, using the initial income distribution and long-term economic growth, known as the convergence hypothesis [79,80,81,82]. In this section, we first explore the correlation between economic complexity and GDP per capita; we then build a multivariate regression model for further examining the effects of income inequalities on complexity and human capital. For measuring income inequalities, we use the GINI coefficient.

As Hidalgo and Hausmann [3] have stated, complexity measures are strongly correlated with a country’s level of income, which could further be a predictive variable for future economic growth.

Figure 5 illustrates a strong relationship between the ECI and log GDP per capita: r = 0.498 for the year 2008, r = 0.589 for 2013, and r = 0.621 for the year 2018, which underlines the main hypothesis that counties with a higher level of complexity have a higher rate of economic growth.

To get a more holistic picture of the effects of complexity on economic growth, we performed a multivariate regression analysis for the year 2018 and then for the period 2008–2018 by using panel data (Equations (9)–(11)). In the latter, we also included time and county fixed effects in order to control for unobserved heterogeneity and to check temporal changes within a county. In both models, we included human-capital-related factors for examining the relation between economic complexity and income inequality when introducing other variables. In the first step, we checked the impact of all the related variables, then we controlled each one separately to compare the robustness of the model.

Before examining the results of the regression analysis, we ran several tests in order to prove the accuracy of the model. First, we examined the existence of multicollinearity. In the first model, the Variance Inflation Factor (VIF) varies between 1.264 and 4.436, while in the second model between 1.052 and 3.033. Thus, it can be stated that the independent variables used in the models are not correlated. In order to verify the presence of heteroscedasticity, we performed the Breusch and Pegan [83] test. In both models, the resulted test statistic had a p-value > 0.05, thus we do not reject the null hypothesis and we assume the existence of homoscedasticity. For checking the autocorrelation in the residuals, the Durbin–Watson test was applied. In the case of the first model, the result was 2.459, while in the case of the second, it was 1.706. As both of them are around the value of two we have excluded the existence of autocorrelation in the residuals.

Further on, for testing the possibility of endogeneity in our panel data model, we performed the Hausman test [84]. The null hypothesis is that the covariance between independent variables and alpha (i.e., unobserved heterogeneity) is zero. In this case, RE is preferred over FE. By rejecting the null hypothesis, the FE model became relevant. First, we ran both the RE and FE models, then we performed the Hausman test. Since the p-value was very small (0.00654), we rejected the null hypothesis; therefore, the FE model seemed to be more suitable, because we clearly have endogeneity in our model.

Column I includes all variables considered for explaining the Gini coefficient, while columns II-VII exclude one variable step by step in each model to examine the effects of each specific variable on the full model. All variables explain 57.2% of the variance in income inequality; however, there are some variables with a larger influence on the Gini coefficient. Observably, in all cases, the Economic Complexity Index is inversely correlated with income inequality and—controlling the effects of all the variables separately—has the largest influence on the Gini coefficient.

The GDP, urbanization rate, and gross school enrolment ratio also show an inverse correlation with the Gini coefficient, having a large influence on income inequalities. Previous economic research has highlighted the importance of education on boosting economic development; moreover, people’s accumulated knowledge, productivity, creativity, and skills represent the basic factors for a competitive and complex economy. As all the creative knowledge is usually located in big urban centers, this explains the importance of urbanization in reducing income inequality.

In all generated models, the unemployment rate has no significant correlation and becomes less important in describing the variations in income inequality, while life expectancy at birth becomes significant only in Model 1 when all variables are included, but with an increasing effect (

Table 1. Cross-section regression result: the effect of Gini on ECI and Human Development.

Dependent Variable: Gini
	(I)	(II)	(III)	(IV)	(V)	(VI)	(VII)
ECI	−0.009 *** (0.009)		−0.002 * (0.007)	−0.009 ** (0.009)	−0.009 ** (0.009)	−0.009 * (0.009)	−0.007 * (0.009)
logGDPpc	−0.191 * (0.090)	−0.024 *** (0.067)		−0.022 ** (0.069)	−0.018 * (0.087)	−0.190 * (0.089)	−0.209 * (0.088)
URB	−0.001 ** (0.001)	−0.001 * (0.001)	−0.001 (0.001)		−0.001 * (0.001)	−0.001 (0.001)	−0.001 * (0.001)
GER	−0.001 (0.001)	−0.001 (0.001)	−0.016 (0.001)	−0.001 (0.001)		−0.001 (0.001)	−0.001 (0.001)
LEB	0.005 * (0.005)	0.005 (0.005)	0.006 (0.006)	0.004 (0.005)	0.006 (0.005)		0.003 (0.005)
UR	0.003 (0.003)	0.002 (0.003)	0.004 (0.003)	0.002 (0.003)	0.003 (0.003)	0.002 (0.003)
Observations	42	42	42	42	42	42	42
R2	0.649	0.638	0.691	0.646	0.646	0.639	0.637
Adjusted R2	0.572	0.574	0.521	0.583	0.584	0.575	0.573
Res. Std. Error	0.037	0.037	0.039	0.037	0.037	0.036	0.033
F-Statistic	7.12 *** (df = 6.35)	8.39 *** (df = 5.36)	6.97 *** (df= 5.36)	8.66 *** (df = 5.36)	8.69 *** (df = 5.36)	8.45 *** (df = 5.36)	8.37 *** (df = 5.36)

Notes: * p< 0.1; ** p< 0,05; *** p< 0.001. The number in parentheses represents the standard error.

).

Next, we tested if the temporal changes of the county’s economic complexity had had an influence on income inequality by using time-and county-fixed-effect panel regression. If the cross-section regression results capture the variations in inequality between counties, this model reveals the changes, which occurred between 2008 and 2018. Even though these changes are not so high, they are indeed relevant for explaining the effects of economic complexity on income inequalities.

As we could observe, the ECI still has a negative influence on the Gini coefficient, which suggests that counties that have registered an increase in economic complexity have been able to achieve a decrease in income inequalities. This also means that one standard deviation in economic complexity contributes to a reduction of 0.01 in the case of the Gini coefficient. As has been stated previously, economic complexity contributes to regional economic growth; however, in the case of Romania, the increase in GDP per capita has led to the deepening of existing internal disparities in a relatively short time. The GDP, urbanization rate, and gross school enrolments ratio also show direct correlation with the Gini coefficient, having a large influence on income inequalities. Previous economic research has highlighted the importance of education on boosting economic development; moreover, people’s accumulated knowledge, productivity, creativity, and skills represent the basic factors for a competitive and complex economy. As all the creative knowledge is usually located in big urban centers, this explains the increasing effect of urbanization on income inequality. Moreover, this is also confirmed by the fourth model, which excludes the urbanization rate, meaning that in less developed regions income inequalities are less significant compared to highly urbanized areas. All these results show the strong relationship between income inequality, economic complexity, and human development and validate the assumption that higher economic complexity contributes to higher economic growth, but that the latter is associated with increased income inequality (

Table 2. Fixed-effect panel regression result: the effect of the Gini coefficient on ECI and Human Development.

Dependent Variable: Gini
	(I)	(II)	(III)	(IV)	(V)	(VI)	(VII)
ECI	−0.014 ** (0.006)		−0.014 *** (0.006)	−0.013 * (0.006)	−0.014 * (0.006)	−0.012 ** (0.007)	−0.012 * (0.006)
logGDPpc	0.016 * (0.069)	0.022 * (0.071)		−0.001 * (0.069)	0.017 ** (0.069)	0.206 ** (0.067)	0.029 (0.069)
URB	0.005 ** (0.003)	0.004 ** (0.003)	0.005 * (0.003)		0.005 * (0.003)	0.006 * (0.003)	0.005 * (0.003)
GER	0.007 * (0.001)	0.001 * (0.001)	0.016 * (0.001)	−0.001 * (0.001)		0.001 ** (0.001)	0.001 (0.001)
LEB	0.015 *** (0.003)	0.016 (0.003)	0.004 *** (0.002)	0.016 *** (0.003)	0.015 *** (0.003)		0.017 *** (0.002)
UR	0.004 (0.002)	0.003 (0.002)	0.004 (0.002)	0.004 * (0.002)	0.004 (0.002)	0.008 (0.002)
Observations	126	126	126	126	126	126	126
R2	0.545	0.487	0.542	0.491	0.539	0.418	0.512
Adjusted R2	0.522	0.466	0.523	0.470	0.520	0.394	0.492
Res. Std. Error	0.034	0.035	0.034	0.035	0.034	0.038	0.035
F-Statistic	23.78 *** (df = 6.119)	22.85 *** (df = 5.120)	28.48 *** (df = 5.120)	23.21 *** (df = 5.120)	28.13 *** (df = 5.120)	17.27 *** (df = 5.120)	25.27 *** (df = 5.120)

Notes: * p < 0.1; ** p < 0,05; *** p < 0.001. The number in parentheses represents the standard error.

).

These results are also in line with the main theoretical finding, which has proved countrywide—based on Standardized World Income Inequality Database—that an improved economic structure contributes to the mitigation of the income gap mainly in low-developed countries; however, in advanced economies, economic complexity is associated with higher income inequality. Even though the Romanian economy has registered great changes and increased economic openness over the last two decades, economic growth was stronger in more developed counties, which was further strengthened by the concentration of human capital and the positive effects of economic agglomerations. This has led in a very short time to the appearance of a twofold economy: one where the economic growth and diversification of economic structures was much more pronounced and has deepened the existing income inequalities even more, the other mainly dominated by primary and secondary economic structures with high ubiquity. As the analysis confirms, one of the greatest changes in economic complexity has occurred in this latter group of counties, as any economic investment has contributed to a decrease in the unemployment rate and a higher occupation rate. Although the diversification of economic structures has led to growth in the economy, this has also contributed to increased income inequalities at a subnational level. Our research is also in line with mainstream economic findings, stating that ECI could predict future economic growth [3,5,11,38]. This result became robust in many studies elaborated in the last years by using both international and subnational data [1,7,18,34,35,36,37,38,39,40,41]. On the other hand, it has also been shown that one standard deviation in the ECI contributes around 3–4% to future economic growth. For example, using panel data in Chinese cities, it was found that one standard deviation in economic complexity contributed to an increase of 0.7% in GDP per capita [85]. Using employment data, Chavez et al. [35] showed an increase of 0.4% in yearly per capita economic growth. Our results are somewhat modest, but we have also demonstrated that one standard deviation in economic complexity decreases income inequality by 0.01%.

Most of the previous studies have concentrated on analyzing the impact of economic complexity on development and income inequality between different countries or worldwide, neglecting the heterogeneity across regions. Even though causality cannot be established between economic complexity and economic growth, our work contributes to the literature by exploring the complexity of economic structure within an emerging CEE country.

Although the analysis provides a complex picture of economic complexity at the subnational level, there are also some limitations.

One of the main limitations of our study is related to the absence of detailed international trade data at county and settlement level, which does not allow for making a comparison with countries that have a similar economic structure. The second limitation is related to the same aspect, namely the relatively short time of the analyzed period (2008–2018). Extending the research for a longer period could explore the changes which have occurred in the economic complexity of each region and the impact of regional economic diversification on growth and income inequalities. It is hard to overcome these limitations due to the unavailability of some variables before 1990.

The third limitation is closely connected to the use of territorial units. As most researchers consider that nowadays cities and metropolitan areas represent the new nations, a detailed analysis of economic complexity at the level of smaller administrative units could explore those local key sectors which have a comparative advantage and could enhance local development. In this sense, targeted place-based policies could optimize the sectoral structures of lagging regions, thus enhancing their integration into a more global economy. Therefore, in the near future, we plan to extend our study even to smaller territorial units.

The fourth limitation refers to the use of four-digit cluster categories for analyzing economic complexity. Extending our research to at least six-digit NACE codes, we could have a more detailed picture about the complexity of different occupations at the county level. We will overcome this limitation in the planned next study.

5. Conclusions

In this paper, we have proposed a measurement of Romania’s economic complexity between 2008 and 2018. To our knowledge, this is the first attempt in Romania to calculate economic complexity by using employment-related company data, which not only complements the existing research in this field, but also gives policy support for enhancing knowledge-based economic activities. As shown earlier, Romania has several key industrial sectors with a comparative advantage; therefore policymakers should focus not only on embracing these sectors but on promoting others as well. For example, behind the IT sector, which is considered the most important driver of growth, contributing 6–7% to the national GDP [54], or the automotive industry which has rapidly grown in the last decades, placing Romania as the fourth biggest automotive manufacturer in Central and Eastern Europe [54], there are other key sectors with a high economic potential. For example, the agriculture and tourism sectors do not stand out as the most complex; however, their contribution to the GDP is around 4–5% at the EU level [54]. Thus, policymakers should focus not only on the diversification of products but also on the types of industries and occupations which could boost the economic prosperity of regions.

The Economic Complexity Index is also becoming more important in the policy programs of the European Commission, mainly in smart specialization [86] and in innovation strategies [87].

At the same time, there is a need for regional development policies to be harmonized between different administrative levels: national, regional, county, and municipal. This is important for boosting the spillover effects; otherwise, any efforts of a region to increase its economic, social, or demographic conditions may not be successful if the adjacent administrative units do not take similar actions [88]. Developing specific productive capabilities based on endogenous resources could also enhance regional development and economic complexity and would have a positive impact on income distribution as well. Since one-size-fits-all strategies cannot be adapted for each region as the economic structure differs, policies should take into consideration those economic activities in which the respective regions have comparative advantages. In this case, focusing much more on bottom-up planning approaches while considering regional diversity and region-specific characteristics could not only facilitate regional development but would also contribute to meeting the Sustainable Development Goals.

As Hidalgo and Haussmann [3] have stated, economic complexity is the best predictor of development potential and the best measure of future economic growth.

Multiple regression analyses have revealed that economic complexity has a major impact on reducing income inequalities, while human-capital-related factors such as urbanization and education dos not have a large capacity to explain income inequality. However, our research has again highlighted the basic idea that well-developed regions have a higher level of economic complexity and lower income inequality; therefore, there is a strong need for the current economic policymakers to elaborate well-founded development strategies for the catching up of lagging regions but also to fulfill the SDG-10.

As the ECI has revealed, the country also has some competitive industries and hidden champions; therefore, by identifying the most sophisticated industries with a comparative advantage, policies can also focus on some areas where domestic production is more efficient.

Our analysis has proven that lower levels of income inequality characterize counties producing complex products and that further diversification of goods and services can boost economic growth and deepen existing inequalities in relatively short periods.