Does technological heterogeneity promote regional convergence? Implications for regional policy and entrepreneurship

Abstract

Regional convergence depends on the ability to adopt advanced technology. This ability can be reflected on the resources devoted to science and technology. An empirical examination of this proposition suggests that technology adoption has a significant effect in regional convergence in Europe. The empirical analysis is also shown to have important implications for the direction of regional policy in Europe.

Key Words:

• technological catch-up
• EU regions

Introduction

The debate on regional convergence has bred, and continues to do so, dozens of empirical studies (e.g. Button & Pentecost, 1995; Hierro & Maza, 2010; Neven & Gouyette, 1995; Rodríguez-Pose, 1999a). Although in this fast growing literature, technological progress has been acknowledged to be of paramount importance; nevertheless, the impact of the adoption of technology has received less attention. Indeed, Bernard and Jones (1996) claim that empirical studies on convergence have over-emphasised the role of capital accumulation in generating convergence at the expense of the adoption of technology. As acknowledged by Abramovitz (1986), technological progress is driven not only by indigenous innovation but also by the process of technology absorption, and thus the ability of a region to ‘catch-up’ might substantially depend on its capacity to imitate and adopt innovations developed in more technologically advanced regions. The aim of this paper is to open a discussion on an empirical model that captures some implications of technology creation and adoption in the process of regional convergence.

Divided into five sections, the theoretical framework upon which the empirical analysis will be conducted is articulated in the second section. Data related issues are overviewed in the third section, and the models are submitted to the usual econometric test yielding the main findings in the fourth section. In the concluding section we offer a possible explanation for the results we obtain and suggest what might afford an interesting policy conclusion.

Technology adoption and regional convergence

A useful starting point is the neo-classical theory, since the assumptions of this theory actually carry implications for the regional convergence/divergence debate. In the standard neoclassical model, a factor that promotes regional convergence is the diffusion of technology. Under the assumption of perfect competition it may be argued that technology has such characteristics and is, as Borts and Stein (1964) argue, ‘available to all’ (p. 8). A process of technology diffusion1 is not a simple and automatic process. Instead, it requires that lagging economies (countries or regions) should have the appropriate conditions to adopt or absorb the technological innovations.2

An implicit assumption of the neoclassical model is that all regions are able to absorb technology to the same degree so that the higher the technological gap, the higher the effect on growth, ceteris paribus 3 . However, it may be argued that large gaps do not necessarily promote convergence in this way. It is quite possible that a significant technological gap is associated with unfavourable conditions for the adoption of new technology. Assume that the rate of technology adoption (ξ) is endogenously determined as a non-linear function of the technological gap4 (b_lfi): with ρ,π>0. This implies that the rate of adoption is not constant but varies across regions, according to the size of the gap. Thus, for a given value of ρ, a high technological gap implies a low capacity to absorb technology. The parameter ρ can be interpreted as a constant underlying rate of diffusion, which would apply to all regions if there were no resource constraints upon technological adoption. However, the existence of such constraints causes the actual rate to diverge from ρ. In other words, the higher the technological gap, the slower the rate of technological adoption (ξ_i). Consider an economy divided into three regions, a technological ‘leader’ (l) and two followers, i=1, 2.5 Assume that , which implies that ξ₁−ξ₂<0. If (δξ_1,2)_t→∞, then , as t→∞ and the two regions move towards different directions (Fig. 1). This can be shown using a Cobb-Douglas production function.6 The growth rate of output per-worker (y) is expressed as and log(y_i/y*)=α log(k_i/k*). Convergence towards steady-state equilibrium, approximated by output per-worker of the leading region, can be expressed as, where Z_i=(1−α)(n+ξ_i+δ) measures the rate of convergence towards the leading region. Given that ξ₁−ξ₂<0, then Z₁−Z₂<0. Movements towards overall convergence occur only as regions become similar in terms of their adoptive abilities.

Fig. 1. Technological divergence.

The general framework, discussed in this section will be tested empirically in an extensive regional context, viz. the NUTS-2 regions of Europe.7 Prior to this, however, the next section reviews the most commonly used ways to approach the issue of convergence empirically together with an extended discussion of the appropriate measurement of the main variables of the model.

The empirical context

The empirical literature on regional convergence makes extensive use of two alternative tests for convergence, namely absolute and conditional convergence, described by Equations 1 and 2, respectively: 1 2 where y_i represents per-capita output of the i ^th economy (in logarithm form), g_i=(y_i,T−y_i,0) is the growth rate over the time interval (0,T), and ɛ_i is the error term. Absolute convergence occurs if b₁<0 8 while the speed at which economies approach the steady-state value of output per-worker over the given time period; that is, the average rate of convergence9 is calculated as β=ln(b ₁ +1)/− T . If b<0 then β>0. Obviously, a higher β corresponds to more rapid convergence.10

Conditional convergence requires that b₁<0 and .11 If different regions have different technological parameters, then convergence is conditional on these parameters, giving rise to different steady-states. It follows, therefore, that a test for conditional convergence is more suitable to accommodate an empirical application of the model developed in the second section mentioned above, and it becomes of critical importance to choose an appropriate variable to approximate the process of technology adoption. At the heart of this concept is a constellation of technologically dynamic and innovative sectors. For the purpose of this paper, a region's adoption capacity is measured as the percentage of total employment in technologically dynamic sectors (Ξ_i).12 More formally: 3 where refers to personnel employed in high-tech manufacturing and knowledge-intensive high-technology sectors13 (v=1, …, k), while is the employment in all the sectors (j=1, …, m) of a regional economy i.

Before we proceed further, it is important to provide some descriptive statistics on the adoptive variable using a transition matrix Table 1.

Table 1. Transition matrix, employment in technological advanced sectors

	n [1995]	Ξi, 2006						n [2006]
Ξi, 1995	51	[0–0.5)	0.1493	0.0299	0.0075	0.0000	0.0037	49
	34	[0.5–0.75)	0.0261	0.0597	0.0299	0.0112	0.0000	34
	51	[0.75–1)	0.0075	0.0187	0.0970	0.0522	0.0149	51
	69	[1–1.3)	0.0000	0.0187	0.0560	0.1194	0.0634	68
	63	[1.3–	0.0000	0.0000	0.0000	0.0000	0.0000	66
	268		[0–0.5)	[0.5–0.75)	[0.75–1)	[1–1.3)	[1.3–	268

Over 40% of the EU-27 NUTS-2 regions have remained to the same range of distribution. About 21% of the technologically lagging regions (less than 75% of the EU-27 average employment in the technologically advanced sectors) have not changed their low position while over 22% of the technologically advanced regions retain in the same range of distribution. Fewer upward movements took place, suggesting that difference in technological levels across regions remained virtually unchanged during the period 1995 to 2006.

Equation 3, represents the level of technological development, but also indicates a capacity for technology adoption since these are taken to apply high technology. However, the potential for such technology diffusion increases as the technological gap increases, defined as the distance between a region's technological level and that of the most advanced technological region with the highest percentage of employment in high-tech manufacturing and knowledge-intensive high-technology services.14 Consequently, in this context a variable that approximates the technological gap for region i at time t can be defined as follows: 4

The gap with the leading region has remained the same range (less than 50%) for 76% of the EU-27 regions (Table 2). For the period under examination, only 10% of the EU-27 regions were able to reduce their technological gaps. These regions had a gap in 1995 within the range between 50 and 75% while in 2006 they moved to a gap less than 50%. Only 2% of the EU-27 regions were able to reduce their gap with the leader; that is, regions that were above 75% in 1995 and moved to the range between 50 and 75% in 2006. It might be argued, therefore, that the process of technological diffusion across the regions of the EU-27 is a slow one. For example, the extremely technologically lagged regions (i.e. those with a technological gap above 75%) retained this relatively low position throughout the period 1995 and 2006. Clearly, a catch-up with the technologically advanced regions is difficult. A similar situation appears estimating the density functions15 for the initial and the terminal years of the sample (Figs. 2 and 3, respectively), which show that the probability mass is concentrated around a range between 30 and 40% of the gap with the leading region.

Fig. 2. Density function, ‘Technological Gap’, EU-27 NUTS-2 Regions, 1995.

Fig. 3. Density function, ‘Technological Gap’, EU-27 NUTS-2 Regions, 2006.

Table 2. Transition matrix, technological gaps

	n [1995]	Technological gap, 2006				n [2006]
Technological gap, 1995	214	[0–0.5)	0.7603	0.0412	0.0000	232
	46	[0.5–0.75)	0.1086	0.0637	0.0000	33
	7	[0.75–1)	0.0000	0.0187	0.0075	2
	267		[0–0.5)	[0.5–0.75)	[0.75–1)	267

Embodied in the TG_i variable is the idea of both a gap and the capacity to adopt technological innovations. The further away a region's technology is from that of the most advanced region, the faster will be its rate of technological progress. The logic behind this hypothesis is that technology transfer will be relatively cheap for lagging regions when compared to regions that are already employing the most modern technologies and that cannot, therefore, simply imitate existing production techniques in order to promote further growth. Lagging regions can, therefore, experience faster growth provided, of course, that they possess the necessary conditions to facilitate the adoption of technology from the more technically advanced regions. According to this model, the potential for technology adoption is positively related to the technological gap, i.e. the higher the technological gap, the higher the potential for technology adoption and the faster the rate of convergence. The presence of a technological gap alone is not sufficient to promote significant technology diffusion. There has to be an appropriate level of capability to adopt technology. Thus, the bigger the gap, the greater the potential for technology adoption but the lower the capacity to actually achieve this. Therefore, it is possible to express a model of ‘technologically conditioned’ convergence as follows: 5 In Equation 5 the variable TG_i is expressed in the initial time. There are two primary reasons for such an approach. The first is related to the fact that future growth is affected by current efforts to enhance technology. Therefore, including the TG_i variable at the initial time captures these long-run effects of technology on regional growth over a specific time period. A second reason for using initial values is that it tests the hypothesis that initial conditions ‘lock’ regions into a high or low position, for example, how high or low levels of technology affect the pattern of regional growth and convergence. In addition, including the TG_i variable in initial time reflects the argument that a low (high) initial technological gap can be conceived as favourable (unfavourable) conditions. From an econometric point of view, inclusion of the technological variable measured at the initial time helps to avoid the problem of endogeneity (Pigliaru, 2003). Equation 5, thus, incorporates the potential impact of both internally generated technological change and technology adoption upon a region's growth. The TG_i variable reflects two distinct features, namely, the level of ‘technological distance’ from the leading region and the degree to which existing conditions in a region allow adoption of technology. A high initial technological gap combined with a high rate of growth may indicate, ceteris paribus, that less advanced regions are able to adopt technology, which is transformed into high growth rates and, subsequently, convergence with the technological regions. It may be argued, therefore, that the condition b₂>0 promotes convergence. On the other hand, a high initial value for TG_i may indicate that although there is significant potential for technology adoption, initial conditions are not appropriate to technology adoption and, therefore, there are no significant impacts on growth. In other words, if the latter effect dominates, then b₂<0 and convergence between technologically lagging and technologically advanced regions is severely constrained. Having outlined the empirical context, the next step forward is to begin to investigate more systematically the pattern of regional convergence in Europe.

Empirical application

The potential for absolute convergence is indicated by a cross-section test, based on estimation of Equation 1 for the NUTS-2 regions of the EU, over the period 1995 to 2006 using data for Gross-Value-Added (GVA) per-worker. Furthermore, the conventional test of regional absolute convergence is modified to include the hypothesis of ‘technologically conditioned’ convergence. The two specifications were estimated using the method of Ordinary Least Squares (hereafter OLS).16 The results are set out in Table 3.

Table 3. Regional convergence, GVA per-worker, EU-27 regions: 1995–2006

Depended variable: gi, n=268 NUTS-2 regions, estimation method: OLS
	Equation 1	Equation 5
a	0.5746**	0.6161**
b1	–0.0753**	–0.0812**
b2	–0.0187*
Implied β	0.0065**	0.0070**
AIC	–291.105	–291.389
LIK	147.552	148.695
F-statistic for the overall significance of the regression		20.138
[p-value]		[0.000]
Wald test for omitted variables, based on the covariance matrix
χ^2 (2) [p-value]		40.277 [0.000]
F (2, 265) [p-value]		20.138 [0.000]
Variance Inflation Factors (VIF)
yi,0		1.113
TGi,0		1.100

Notes: **indicates statistical significance at 95% level of confidence, *90% level. AIC and LIK denote the Akaike information criterion and Log-Likelihood, respectively.

The results of the empirical modelling process are clear and revealing but must be interpreted with some caution. Considering first the results of testing for absolute convergence, it might be argued that there is a slow tendency for absolute convergence across the regions of Europe. The rate of convergence of labour productivity is, on average, about 0.65% per annum. Of particular importance to this paper, however, are the results obtained for the conditional convergence model, which according to the econometric test appears to be preferred from the specification given by Equation 5. This view is further supported by the facts of the Akaike information criterion (AIC).17 This criterion signifies that the specification in Equation 5 dominates the absolute convergence model in Equation 1. Furthermore, the Wald test rejects the null hypothesis, that the estimated parameters in Equation 5 are zero (H₀: b₁=b₂=0 with the alternative ), while the F-test confirms the overall significance of the regression.18 These two tests provide further support to the argument that regional convergence in Europe is related to the size of the technological gap. A potential problem with this model is related with the presence of multicollinearity, which can be detected by calculating the variance inflation factors (VIF). As a rule of thumb, if , then multicollinearity is high. High multicollinearity is not detected for any of the explanatory variables. The average VIF value is about 1.1, which should be interpreted as an indicator of the model's rather negligible levels of multicollinearity.

Conditioning for the technological variable tends to increase the estimated rate of convergence (0.7%). The variable TG_i,0 is negative in sign. A high technological gap does not necessarily imply that technologically lagging regions will be able to adopt technology – a large gap may constitute an obstacle to convergence. This proposition is supported by the empirical analysis, which suggests that, on average, regions with high technological gaps at the start of the period grow slower than regions with low gaps, ceteris paribus. A high initial technological gap is a factor that helps to sustain initial differences across regions, constraining any possibilities for overall convergence and, in turn, suggesting the possibility of convergence towards different equilibria following the predictions of the model, examined previously in the second section. If technologically backward regions of the EU were successful in adopting technology, then the estimated coefficient b₂ would be positive. Since b₂<0, this indicates that the adoptive abilities in regions with high technological gaps are inhibiting technology adoption. Nevertheless, this is a process that might be difficult for lagging regions, especially during the early stages of development when overall conditions are least supportive. In the case of the EU regions, there seems to be a certain ‘threshold’ level of technology and regions below that level are not able to assimilate technological innovation in an efficient way. Convergence towards a steady-state equilibrium growth path is feasible only regions with low technology gaps, relative to leading regions as represented by the growth rate of the leading region. Regions with relatively large technology gaps may fall progressively behind.

Concluding remarks

It is beyond argument that, although an increasing number of empirical studies have paid attention to issues of economic convergence in the EU, the impact of technology adoption in regional convergence has so far received more limited attention. We have attempted in this paper to address this question, using data for the NUTS-2 regions of the EU-27 over the period 1995 to 2006. The results reported in this paper suggest that the NUTS-2 regions of EU-27 exhibit a slow rate of convergence in terms of labour productivity. An important conclusion to emerge from the empirical application is that the EU-27 regions exhibit some tendency to converge faster, although not substantially, after conditioning for technological differences across regions. While the ‘conventional’ approach predicts in principle that the higher the technological distance from the leader, the greater the incentive to adopt technology, the results in this paper imply that not all the lagging regions of Europe are able to reap the ‘benefits of backwardness’. This inability can be attributed, possibly, to inappropriate conditions prevailing in lagging regions, which prevent or constrain convergence with the more technologically advanced regions. Catch-up to the leading regions is feasible only amongst those regions whose technological conditions are similar or close to those of the technologically advanced regions. Therefore, a primary aim of regional economic policy in the context of an enlarged Europe should be the promotion of high-technology activities and Research and Development (R&D) including universities and scientific and research institutions. Moreover, in order to enhance regional growth and convergence, policy should seek to reorient these activities. High-technological and knowledge-creating activities should be directed, if possible, at regions with unfavourable conditions, the purpose being to stimulate the production structures of those regions to shift to activities that implement high technology. Such activities will provide an appropriate environment in the lagging regions of the EU for diffusion of information and knowledge; a sine qua non for entrepreneurial success, as Fischer and Nijkamp (2009) aptly call it. Through information and knowledge, such an environment will put the lagging regions of the EU in a path of convergence with the leading regions. Regional growth is shaped to a great extent by innovative actions of risk-seeking firms. From this perspective, investment in knowledge creation will create dynamic spillovers that enhance growth in a region. Nevertheless, the relation between entrepreneurship and regional growth/convergence is a complex one and clearly more research is required. Quantitative models including a series of microeconomic factors such as networks, industrial structure/organisation, culture, and so on as explanatory variables might provide some interesting aspects of the relation between regional growth and entrepreneurship.

Although this paper has been concerned primarily with regional convergence focusing on the role of labour employed in advanced technological sectors, this is by no means to imply that this approach is the only route to understanding regional growth and convergence. While the empirical results are significant for the case of the EU-27 regions in their own right, they should nevertheless be placed in perspective. Indeed, it is not claimed that the foregoing analysis has provided an exhaustive account of all the factors that affect the process of regional convergence. Hence, improving the model developed in this paper by adding more explanatory variables would open up an interesting avenue for future research. However, the model developed in this paper is sufficiently flexible for it to be applied to other regional contexts such as the United States. Empirical studies in those contexts using alternative variables might reveal different and more interesting features in regard to regional growth and convergence. Nevertheless, the present work suggests possible avenues for future research in different contexts and that examines different factors shaping the pattern of regional convergence.

Conflict of interest and funding

The author has not received any funding or benefits from industry or elsewhere to conduct this study.

Appendix A: Appendix

Consider a Cobb-Douglas production function: A1 where Y stands for output while K, L, and A denote capital, labour, and technological inputs, respectively.

Equation A1 can be expressed in terms of effective units of labour by dividing Equation A1 with A_i L_i: A2 Defining Q_i=Y_i/A_iL_i and k_i=k_i/A_iL_i, then Equation A2 can be written as follows: A3 Labour force grows in accordance to the following relation: A4 Total investment less depreciation approximates the growth rate of capital stock, : A4 where s and δ denote the rate of saving and depreciation, respectively.

Dividing Equation A5 with K_i and using Equation A1 yields: A5 Given that and defining , and , where , , and capital stock per effective worker grows as follows: A6 Using Equation A6 it is possible to transform Equation A7 as follows: A7 Multiplying both sides of Equation A8 by k_i yields: A8 Taking a first-order Taylor approximation of Equation A9 around the steady-state value of k_i yields: A9 In the steady-state . Setting Equation A8 equal to zero and solving for s yields: A10 Substituting Equation A11 into Equation A10 and after some manipulations yields: A11 Assuming that , with , and , then and , then . Given that the gap between actual and steady-state level of k_i can be expressed as logk_i–logk*=log(k_i/k*) and , then Equation A12 can be written as follows: A12 Output per-worker, in effective terms (Q_i), grows as follows: A13 Given that logQ_i=αlogk_i and in the steady-state , then: A14 Equation A15 implies that A15 Using Equations A14 and A15, Equation A13 can be written as follows: A17 Equation A17 can be written as follows: A18 Equation A18 is a differential equation in log Q_i,t with the general solution: A19

Subtracting log Q_i,0 from both sides yields: A20 where g_i,T=log Q_i,t−Q_i,0 is the growth rate of Q_i over a given time period T=t−0, c=(1−e ^−z)Q*, and b=−(1−e ^z).

Notes

¹The issue of technology diffusion is examined in Fagerberg and Godinho (2005).

²A similar view is proposed by Cohen and Levinthal (1990) who state that “[T]he ability to evaluate and utilize outside knowledge is largely a function of the level of prior related knowledge” (p. 129). It should be noted, however, that the authors examine technology adoption in a microeconomic framework. Contrary, the concern in this paper is to examine the process of technology adoption from a macroeconomic perspective.

³This argument has been dealt with at length in Gerschenkron (1962), which is acknowledged as the initiator of this view. Nevertheless, as Fagerberg (1994) claims, the central conceptual apparatus stems from the earlier contribution of Veblen (1915).

⁴For a more detailed discussion see Alexiadis (2010).

⁵Castellacci and Archibugi (2008) put forward the idea of leaders and followers across national economies.

⁶Some technical details are relegated to the Appendix.

⁷Despite an extensive empirical literature on technology and innovation across the EU regions (e.g. Basile, 2009; Rodríguez-Pose, 1999b; Crescenzi, Rodriguez-Pose, & Storper, 2007, etc.), nevertheless the implications of technology adoption are not examined in an explicit way.

⁸Romer (1996) describes perfect convergence as occurring when b=−1 while at the other extreme, a value of zero indicates that the economies included in the data set may even exhibit divergence. Alternatively, b=0 implies g_i=a, which can be considered as an indication of an autonomous growth rate that maintains income differences across economies.

⁹The time at which output per worker (y_i,t) is halfway between the value during the initial year and the ‘steady-state’ (y*) satisfies the condition .

¹⁰The convergence coefficient is bounded to the sign of β, implied by log(y_i,t)=(1−e ^−βt )log(y*)+e ^−βt log(y_i,0).

¹¹For a more detailed analysis see Barro and Sala-i-Martin (1992, 1995).

¹²This proxy is in accordance with several models of endogenous growth (e.g. Romer, 1986, 1990) that emphasise the existence of a sector that deliberately produces technological innovations.

¹³The European Statistical Office (EUROSTAT) is the main source for data used in this paper.

¹⁴This is the region of ‘Berkshire, Bucks, and Oxfordshire’ in the UK.

¹⁵The plot density estimate is a standard fixed bandwidth, following Silverman's (1986) rule.

¹⁶The use of OLS estimation allows us to employ a battery of powerful tests to choose between different specifications of the convergence model.

¹⁷As a rule of thumb, the best fitting model is the one that yields the minimum value for the AIC.

¹⁸This is computed as , where ESS is the explained sum of squares, RSS is the residuals sums of squares (the total sum of squares is TSS=ESS+RSS), k is the number of parameters including the constant and n is the number of observations. The null hypothesis associated with this test is that all the regression coefficients are equal to zero. A low probability value implies that at least some of the regression parameters are non-zero and that the regression equation does have some validity in fitting the data (i.e. the independent variables are not purely random with respect to the dependent variable).