Raising the bar (24)

ABSTRACT

This editorial summarizes the papers in issue 18(2) (2023). The first paper extends the Solow–Swan growth model with spatial dependence, pollution and time delay. The second paper investigates the (mis)match between relative factor costs and the output elasticities of production factors due to innovations in the European Union’s Smart Specialisation Strategy (S3). The third paper studies whether and in what way price regulation of gasoline affects competition and collusion between gasoline stations. The fourth paper provides empirical evidence that public library programmes encourage labour force participation in underdeveloped regions. The fifth paper proposes a general nesting spatial stochastic frontier model and a maximum likelihood estimation procedure to determine inefficiencies across units of observations. The sixth paper proposes a dynamic spatial autoregressive model in which the overall spatial weight matrix is composed of a convex combination of multiple underlying spatial weight matrices and the coefficients of all regressors are heterogeneous.

KEYWORDS:

• growth
• economic activity
• innovation
• pricing
• frontier

JEL:

• C21
• L11
• O3
• O4
• R11

Spatial Economic Analysis is a pioneering journal dedicated to the development of theory and methods in spatial economics. This issue contains six different kinds of papers contributing to the journal’s mission. The first paper presents a purely economic–theoretical model; the next three papers present an empirical application based on an economic–theoretical model (in either mathematical or verbal form); and the final two papers introduce new spatial econometric models and estimators illustrated with empirical applications.

The first paper, by Segura et al. (2022, in this issue), introduces a time lag and pollution effects to a spatial Solow–Swan economic growth model (Camacho & Zou, 2004). The time lag is introduced because it takes time for economic agents to act on information about the current market situation. The authors’ economic–theoretical model generalizes earlier studies by Solow (1956), Camacho and Zou (2004) and Matsumoto and Szidarovszky (2011). Next, the authors introduce conditions under which it is possible to guarantee the existence of a delay-independent constant globally asymptotically stable (GAS) steady state. For this purpose, they first apply the method proposed by Yi and Zou (2010), which reduces the problem to the analysis of a simpler first-order derivative equation. They then investigate the conditions for a GAS steady state under different scenarios: without pollution (considering both Cobb–Douglas and constant elasticity of substitution production functions) as well as with pollution (for a Cobb–Douglas production function associated with four different functional forms of pollution effects). Finally, the authors obtain a condition for a GAS steady state in the absence of spatial dependence, which has not been studied previously.

The second paper, by Antonelli et al. (2022, in this issue), links the Smart Specialisation Strategy (S3) of the European Union with technological congruence, in simple terms defined as the (mis)match between relative factor costs and the output elasticities of production factors. For detailed mathematical equations and explanations, we refer the reader to section 2 of this paper. The most important extension is that the authors consider the production elasticities $\alpha$ and $1-\alpha$ of capital and labour in the Cobb–Douglas production function to be time dependent. In Solow (1957), innovations do not affect the capital-to-labour ratio as technological progress is neutral. However, if innovations can affect production elasticities, this property no longer holds, possibly causing technological congruence due to innovation strategies. These innovation strategies aim to introduce new process technologies that intensify the exploitation of production factors that are locally abundant and hence relatively less expensive. Consequently, for firms competing in global product markets, heterogeneous market conditions are advantageous from a competitive viewpoint if they operate in regions with relative abundant factor endowments that match the intensity of new process technologies. To investigate this empirically, the authors estimate both a non-spatial panel data model and a spatial Durbin model with region and time fixed-effects, using an unbalanced panel of 278 regions across 28 European countries with starting years ranging from 1980 to 1991, dependent on the region. They find that technological congruence has significant effects on total factor productivity. In the last two sections of their paper, the authors extensively discuss the policy implications of their findings for S3. Readers interested in S3 and government action to support and orient technological change (Aghion et al., 2011; Foray, 2014, 2018; McCann & Ortega-Argilés, 2015) should read this part of the paper in detail as the implications are too diverse to summarize here in one or two sentences.

The third paper, by Yu et al. (2022, in this issue), adopts a spatial autoregressive (SAR) model to explain gasoline prices in Inner Mongolia using data from 712 gasoline stations between May and December 2018. The purpose of this study is to investigate whether and in what way gasoline prices affect each other when price regulation in China, as in many other countries, prohibits stations from charging more than a price ceiling. For this purpose, the authors introduce dummies for the two state-owned enterprises (PetroChina and Sinopec) and compare their coefficients with the intercept reflecting other enterprises. They also add and compare the coefficients of interaction effects of these dummies and the intercept with two additional variables: (1) a dummy labelled ‘stand-alone’ for stations operating in an area without any other stations within a radius of 3, 5 or 7 km; and (2) the ratio between the sales of stations of the same brand as the focal station and the total sales of all gas stations, both again within a radius of 3, 5 and 7 km. Using this set-up, the authors make a distinction between price-setting behaviour that points to competition or to collusion. They find that the SAR coefficient is positive and significant and that stand-alone stations collect a price premium due to local monopoly power if the price does not exceed the price ceiling. They also find that there is no competition between PetroChina’s stations and little competition among Sinopec’s, unlike the stations of other companies.

The fourth paper, by Neto (2022, in this issue), studies the effect of public library programmes on unemployment and labour force participation in Appalachia, a rural and economically underdeveloped region in the United States comprised of 420 counties across 13 states. Its labour force participation rate is below, and its unemployment rate is above, the national average. Interestingly, when exploring the mutual relationship between the presence of a public library system in a county with these labour market outcomes, the author finds that its presence has a downward effect on the unemployment rate and an upward effect on the labour force participation rate when all US counties are considered, but that the upward effect on the labour force participation rate disappears when only Appalachian counties are considered. In view of this, the author investigates whether adult and child library programmes can serve as an alternative to other active labour market programmes (ALMP), which generally appear to be ineffective. For this purpose, the author uses data extracted from the Public Library System (PLS), namely either the number of adult and child programmes or the participation in these programmes. Moreover, since these explanatory variables are likely to be endogenous to both labour market outcomes, the author instruments each of them with the number of computers and the number of librarians without a master’s degree, which are shown to be valid (strong and exogenous). The author estimates several econometric models. Standard econometric models without any spatial relationships between counties show no direct effect of the programmes, similar to previous studies on ALMPs. This also holds when either switching to spatial econometric models – the spatial lag of X (SLX) model and the spatial Durbin error model (SDEM) – or instrumenting the number or the participation in adult and children’s programmes. However, when doing both, he finds significant empirical evidence in favour of spillover effects of the adult programmes on labour force participation. Although public library programmes may not help people find jobs, the author explains that these programmes reduce the cost of joining the labour market. Adult programmes focus on job services and skills training and are designed to help adults find or keep their jobs (Hunt, 2017). This holds especially for Appalachia, where people have less access to formal training (Pollard & Jacobsen, 2017) and to the internet at home (Stenberg et al., 2009).

The fifth paper, by Galli (2022, in this issue), proposes a spatial Durbin stochastic frontier production function for time-series cross-section data and spatial dependence in both the standard error term and the inefficiency error term. The inclusion of spatial lags in the dependent variable (output) and the independent variables (inputs) follows Glass et al. (2016) and is meant to account for flexible global and local spillover effects. The introduction of spatial dependence in both the standard error term and the inefficiency error term follows Orea and Alvarez (2019) and is meant to avoid biases in the inefficiency error term and the ranking of inefficiencies across the units of observation. In addition to this, the inefficiency error term is taken to depend on a set of exogenous variables. Although not indicated, considering spatial lags in the dependent variable, the independent variables and both error terms, also makes the proposed model a general nesting spatial stochastic frontier model. To estimate the parameters of this model, the author develops a maximum likelihood estimation procedure. We use the terminology ‘time-series cross-section data’ here rather than ‘panel data’ (as used in this paper) since controls for cross-sectional or time-specific effects, either fixed or random, are not included. In addition to Monte Carlo simulation results, the author also provides an empirical illustration. They consider a Cobb–Douglas production function with five production factors for the Italian agricultural sector at the NUTS-3 level over the period 2008–18, and eight exogenous variables in the inefficiency error term. The spatial weight matrix is the same for all spatial lags and specified as a row-normalized second-order binary contiguity matrix. In sum, this paper offers a major step forward within this area of literature.

The sixth paper, by Bao and Zhou (2022, in this issue), combines two advanced spatial econometric models into an even more advanced spatial econometric model. The first model is taken from Aquaro et al. (2021) and represents a dynamic SAR model with heterogeneous coefficients for each unit of observation. This model requires a large number of observations in the time domain otherwise its coefficients cannot be estimated consistently. The second model is taken from Debarsy and LeSage (2022), who specify the spatial lag in the dependent variable as a convex combination of different types of spatial weight matrices. Multiple weight matrices are used to avoid ad-hoc choices of a single matrix, for which empirical studies based on spatial econometric models are often criticized. The resulting model takes the form of a dynamic SAR model in which the overall spatial weight matrix is composed of a convex combination of multiple underlying spatial weight matrices and the coefficients of all regressors are heterogeneous. Recent empirical studies applying this method but with homogeneous coefficients include Cai et al. (2022) and Moallemi et al. (2022). To estimate the parameters of the model, the authors develop a Bayesian estimation procedure. In addition to Monte Carlo simulation results, they provide an empirical illustration. They explain real house price growth rates in 338 metropolitan statistical areas (MSAs) in the US as a function of population and real per capita disposable income growth rates. The data are taken from the period 1975.Q2–2019.Q4, so T is large. Furthermore, two geographical and one non-geographical spatial weight matrices are considered. By comparing the results of their model with baseline models, the authors find that many more MSAs have positive net spatial parameter estimates and, relatedly, that more MSAs produce positive and significant population and income growth spillover effects.