Periurban Tides and Archipelagos: Labour Flows and Aggregate Demand in Romania’s FDI-Driven Spatial Economy

Abstract

This data repository contains data at municipality data in Romania about population and population change, wages, agregate turnover, sectorial employment, and e,ployees by types of capital (FDI-capital, domestic and public). The data explores the spatial dynamics of the labor market at the subnational level, providing insights into wage moderation and repression in export-led growth regimes in Central and Eastern Europe. The data are used to investigate the spatial concentration of specialized economies within cities and their periurban areas, where the regional labor force is leveraged to moderate wage increases and attract populations in economies heavily reliant on FDI, in Romania. The data and models are used to show that this leads to the formation of ‘enclave economies’, characterized by localized labor regimes shaped by territorial zoning strategies that regulate labor migration and economic zoning of capital. Employing spatial regression with SARAR-SUR, we model population change and concentration to identify different labor regimes in regional enclaves, examining the impact of wages, sectorial employment, and types of capital. Our findings demonstrate that the sub-national distribution of FDI-led growth in Romania primarily revolves around labor-intensive activities and low capitalization costs, rather than urbanization. Additionally, we observe that changes in employment within public services are not significantly associated with population changes, suggesting that the state does not play a major role as a competitor in the labor markets. Furthermore, our analysis captures the specific labor requirements of multinational firms operating in business services and manufacturing, highlighting the negative impact of foreign companies in the business service sector on the population of core cities.

Data

Dependent Variable

- The dependent variable in our analysis is population change, which we measure using the population ratio of 2012 to 2011 on a natural logarithmic scale due to its mathematical similarity to percent growth. Dual citizens of Romanian descent from Moldova and Ukraine are drawn to the eastern region, particularly border towns in Suceava, Botoșani, Iași, and Vaslui counties. These immigrants often use their Romanian citizenship to migrate within the EU. To account for this regional trend, we substracted from the population the number of emigrants from each locality over the past decade (both in 2011 and 2021)

Independent variables

- We model population change at locality level using wages as a pull factor, capital type and economic sector. To assess the impact of personal income on employment, we have utilized personal income tax data to estimate aggregated wages at the local level as provided on the data portal of the Romanian's Ministry of Regional Development and Public Administration.

- The Romanian National Institute of Statistics categorizes individuals as employed or working to account for those not receiving wages, including self-employed and contributing family workers. Agricultural workers make up most of the un-waged working population. To measure waged relations within the labor pool, we used the ratio of waged employees to the active age population (16-65 years), which serves as a measure of the size of the labor market at the local level. 

- We obtained employment data from the National Institute of Statistics, which provided aggregated balance sheets of all companies at the subsidiary level for 2011 and 2021. 

- We used a dummy variable to distinguish between foreign and domestic companies, and we separately aggregated the number of employees working in local and foreign companies. For the purposes of this study, a company was deemed foreign if it was incorporated in Romania and had 50% or more of its equity shares or share capital owned by a natural or legal person residing outside of Romania.

- NACE codes related to manufacturing were used to isolate the companies of interest. To differentiate business services from other service activities, NACE codes related to activities such as information and communication, financial and insurance activities, real estate activities, professional, scientific, and technical activities, and administrative and support service activities were used to filter companies. The aim is to capture the growth of outsourcing in this sector while excluding other service activities such as social services and commerce and logistics.

- The first layer of municipalities surrounding the 260 cities in Romania are referred to as periurban area, as they represent the transitional zone between urban and rural environments (Dadashpoor and Ahani, 2019; Stahl, 1969). Out of the 319 cities in Romania, 59 towns are located within the periurban areas adjacent to larger cities in terms of population. This classification creates three typologies of administrative territorial units (3180): core cities (260), periurban localities (1330), and villages (1590). These locality types (core city, periurban localities, and villages) were transformed into dummy variables and utilized as interaction variables. 

- In Romania, periurban areas are defined by Law no. 246/2022, which focuses on metropolitan areas and involves modifications and additions to certain normative acts. For municipalities, the periurban area includes the first two layers of municipalities surrounding the core city, while for cities, it encompasses only the first layer. However, it's important to note that the majority of population growth occurred in the first layer of municipalities. Due to this, we considered the periurban area for all 260 cities to be equivalent to the first layer of municipalities.

- An alternative approach for analysis would have been to use the functional urban area, as defined by Eurostat, which covers the localities within the commuting range of the central city. Nevertheless, this scale is unsuitable for examining population change, as it shows only a negligible percentage change (0.1%) between 2003 and 2020. In contrast, periurban areas exhibited a much more significant population change of 5% during the same period. Consequently, we opted to present our analysis using the concept of periurban areas.

Model specification

- We employed two specifications for the independent variables: a cross-sectional specification for 2021 and a first-differencing strategy that measures the difference between municipality-level values in 2011 and 2021. We used both specifications to assess the effect of employment composition across municipalities on population dynamics. The cross-sectional specification assumes that larger employment markets with more employees out of the active age population act as population magnets. However, it is influenced by idiosyncratic factors specific to each municipality. The first-differencing strategy controls for individual-level effects and plays a similar role to a fixed effect for two discontinuous time points, accounting for time-variant factors . It assumes that an increase in employment opportunities at the municipality level generates an overall population increase, irrespective of the size of the labor market out of the active age population.

- Despite utilizing all three categories of the locality type simultaneously as a dummy system, multicollinearity did not pose a problem. The interaction term divided the independent variable into spatial components, forming spatial regimes, as described by Anselin and Rey (2014). Furthermore, when the locality type was used as an interaction effect in all models, the continuous variable only covered a portion of the population, specifically those within an economic sector. It did not account for the entire population of employees.

Model selection

- We employed a three-stage least squares approach using Seemingly Unrelated Regression (SUR) models. We incorporated Spatial Autoregressive terms and Spatial Autoregressive Disturbances (SARAR). R package spsur (Angulo et al., 2021) to estimate the model, and a row-standardized queen contiguity spatial weights matrix was computed from the geometries of Romanian localities to perform the regressions.

Files

- data cross.csv: The data for the cross-sectional specification for 2021

- data diff.csv: The data for the first-differencing strategy that measures the difference between municipality-level values in 2011 and 2021

- model.csv: The data which contain the analysis for preparatory analysis and model selection

- model A.R: The code in R with the preparatory analysis and model selection

- model B.R: The code in R with the analysis with the two specifications for the independent variables

- Vizualization.twbx: The Vizualization in Tableau of the data and the predicted values of the different models

- codebook data cross.csv

- codebook data diff.csv

- codebook model.csv