Total Factor Productivity, Intangible Assets and Spatial Dependence in the European Regions

Abstract

Dettori B., Marrocu E. and Paci R. Total factor productivity, intangible assets and spatial dependence in the European regions, Regional Studies. The aim of this paper is to analyse the determinants of the efficiency levels across the European regions. Firstly, a regression-based measure of regional total factor productivity is derived by estimating a spatial Cobb–Douglas production function. Secondly, the role played by intangible factors (human capital, social capital and technological capital) on total factor productivity levels is investigated by applying the spatial two-stage least-squares (2SLS) method and the spatial heteroskedasticy and correlation consistent (SHAC) estimator to account for both heteroskedasticity and spatial autocorrelation. It turns out that a large part of total factor productivity differences across the European regions is explained by disparities in the endowments of these intangible assets.

Dettori B., Marrocu E. and Paci R. 欧洲地区总要素生产力、无形资产以及空间依赖，区域研究。本文旨在分析欧洲区域决定其效率等级的不同要素。首先，通过估测空间生产职能本研究对区域总要素生产力进行了基于回归的测量。其次，通过应用空间两阶段最小二乘算法 （2SLS）以及考虑了异方差以及空间自相关的SHAC估算，研究考察了无形要素（人力资本、社会资本以及技术资本）对于总要素生产力水平的影响。

总要素生产力 人力资本 社会资本 技术 欧洲

Dettori B., Marrocu E. et Paci R. La productivité des facteurs de production globaux, les immobilisations incorporelles et la dépendance géographique dans les régions européennes, Regional Studies. Cet article cherche à analyser les déterminants des niveaux d'efficacité à travers les régions européennes. Primo, on établit une mesure de la productivité des facteurs de production globaux basée sur la régression en estimant une fonction de production géographique du type Cobb–Douglas. Secundo, on examine le rôle joué par les immobilisations incorporelles (à savoir, le capital humain, le capital social et le capital technologique) sur les niveaux de la productivité des facteurs de production globaux en appliquant la méthode géographique des moindres carrés à deux degrés (appelée 2SLS) et l'estimation SHAC pour tenir compte à la fois de l'hétéroscédasticité et de l'autocorrélation géographique. Il s'avère qu'une part non-négligeable des écarts de la productivité des facteurs de production globaux à travers les régions européennes s'explique par des disparités de la dotation de ces immobilisations incorporelles.

Productivité des facteurs de production globaux Capital humain Capital social Technologie Europe

Dettori B., Marrocu E. und Paci R. Gesamtfaktorproduktivität, immaterielle Vermögenswerte und räumliche Abhängigkeit in den europäischen Regionen, Regional Studies. In diesem Beitrag werden die Determinanten des Effizienzniveaus in verschiedenen europäischen Regionen analysiert. Zu Beginn wird ein regressionsbasierter Maßstab der regionalen Gesamtfaktorproduktivität durch Schätzung einer räumlichen Cobb–Douglas-Produktionsfunktion abgeleitet. Anschließend wird die Rolle von immateriellen Faktoren (Humankapital, Sozialkapital und technologisches Kapital) für das Niveau der Gesamtfaktorproduktivität untersucht, indem die Methoden der räumlichen zweistufigen Kleinstquadrat-Schätzung (2SLS) und des SHAC-Schätzers angewandt werden, um sowohl die Heteroskedastizität als auch die räumliche Autokorrelation zu berücksichtigen. Es stellt sich heraus, dass sich ein Großteil der Unterschiede bei der Gesamtfaktorproduktivität in den verschiedenen europäischen Regionen auf Disparitäten hinsichtlich der Ausstattung mit diesen immateriellen Vermögenswerten zurückführen lässt.

Gesamtfaktorproduktivität Humankapital Sozialkapital Technologie Europa

Dettori B., Marrocu E. y Paci R. Productividad total de factores, activos intangibles y dependencia espacial en las regiones europeas, Regional Studies. El objetivo de este artículo es analizar los determinantes de los niveles de eficacia en las regiones europeas. En primer lugar, derivamos una medición basada en la regresión de la productividad total de los factores regionales al calcular una función espacial de producción Cobb–Douglas. En segundo lugar, analizamos el papel desempeñado por los factores intangibles (capital humano, capital social y capital tecnológico) sobre los niveles de productividad total de los factores aplicando el método espacial de mínimos cuadrados en dos etapas (2SLS) y el estimador SHAC para tener en cuenta la heteroscedasticidad y la autocorrelación espacial. Parece ser que una gran parte de las diferencias en la productividad total de los factores en la regiones europeas se explica por las desigualdades en las dotaciones de estos activos intangibles.

Productividad total de los factores Capital humano Capital social Tecnología Europa

Keywords:

• Total factor productivity
• Human capital
• Social capital
• Technology
• Europe

JEL classifications:

• C31
• C33
• O47
• O52
• R11

Acknowledgments

The research leading to these results received funding from the European Community's Seventh Framework Programme (Grant Number FP7/2007-2013) under Grant Agreement Number 216813. The authors would also like to thank for their useful comments the participants at the 2008 ERSA Conference, 2009 SEA Conference and DECA-CRENoS seminar. The authors also benefited from fruitful discussions with Paola Zuddas. They thank Francesca Alberti, Giuliana Caruso and Marta Foddi for valuable assistance in preparing the database. They also thank J. Paul Elhorst for kindly making publicly available the Matlab routines for estimating spatial models.

Notes

Results are not reported for reasons of space, but they are available from the authors upon request and reported in Dettori et al. (2008) where an extended version of this paper can be found.

To check the robustness of the results, Moran's I test is calculated allowing for different specifications of the spatial weight matrix. A more detailed discussion on the economic aspects of matrix normalization is postponed to the next section.

When the panel data are stacked as T-subsequent cross-sections, one has:

where I_T is a (T × T) identity matrix; and _N is the normalized spatial weight matrix.

Anselin (1988) emphasizes that the aim of spatial econometrics should be on measuring spatial spillovers. The alternative spatial error model specification is just a particular case of non-spherical errors which eliminates spillovers by construction.

For a thorough discussion on normalization issues, see also Elhorst (2010).

Note that, as emphasized by Anselin et al. (2008), row standardization also has the side-effect that the sum of all the elements in W equals N, the number of cross-sectional observations, and that the induced asymmetry in the weights ‘is an unusual complication with significant computational consequences’ (p. 628).

For a discussion on the relevance of absolute distance versus relative distance in economic phenomena, see Baltagi et al. (2008).

For cross-section analyses, exceptions are represented by Kelejian and Prucha (2004, 2007), Anselin and Lozano-Gracia (2008), Fingleton and Le Gallo (2008), and Dall'Erba and Le Gallo (2008); for a panel application, see Elhorst et al. (2007).

TFP is estimated using measured inputs – a possible cause of the disparities among regions relies on measurement errors; moreover, there may be problems of misspecification of the production function (Caselli, 2005).

Similar results are found when testing for weak exogeneity of capital and labour within an error-correction model framework; only labour can be considered weakly exogenous (the p-value for the null hypothesis that the adjustment term is zero in the labour Error Correction Model (ECM) is equal to 0.293).

The panel version of the test is reported in Anselin et al. (2008).

Given that the sample refers to a period of twenty-two years, a panel co-integration analysis was also carried out in order to guard against the spurious regression problem. Evidence of a non-stationary kind of behaviour was detected for the individual variables by means of the cross-sectionally augmented Im Pesaran Shin (CIPS) unit root tests (Pesaran, 2007). The existence of a spurious relation between the variables of interest was ruled out by the results of the well-known co-integration tests developed by Pedroni (1999, 2004), which allowed the null hypothesis of no co-integration to be rejected. All detailed results are reported by Dettori et al. (2008).

Models that account for the two sources of endogeneity separately were also estimated. The model estimated by an instrumental variable without the spatial term, as expected, yields spatially autocorrelated residuals; this results points to the importance of modelling explicitly the spatial pattern. On the other hand, the spatial lag model for which the productive inputs are not instrumented (estimated by the maximum likelihood (ML) method) ensures that the residuals do not exhibit spatial autocorrelation, but the coefficients for the capital and labour regressors are quite similar to the ordinary least-squares (OLS) ones, signalling that the endogeneity bias is still present.

For 2SLS estimated models, Moran's I test is calculated as suggested by Anselin and Kelejian (1997) for the case of instrumental variable residuals.

For the same reason, the measures of dispersion for the impact estimates are not reported. Note also that for the normalized matrix adopted for each entry has a very small value.

All the other results are available from the authors upon request.

Note that possible changes in relationship 1.2 are accounted for by the temporal dummies. A subsample analysis was not carried out in order to check the robustness of the result, as splitting the sample would result in a loss of valuable information needed to estimate the fixed-effects accurately (it is well known that the fixed-effect estimator makes use of the within dimension of the sample information, which in the case is constituted by twenty-two time observations).

For some regions in France, Germany and the UK, data are available at the NUTS-1 level so that that value is assumed for the included NUTS-2 regions.

In the special case in which the omitted relevant variables follow a spatial autoregressive process and are also correlated with the included spatially correlated ones, LeSage and Pace (2009) show that this leads to a spatial Durbin model specification. The 2SLS method is more general as it allows one to tackle different sources of endogeneity bias. Note also that in applying the latter method, instruments are represented by the explanatory variables spatial lags, so that it is not possible to include them as regressors, as would be required by the spatial Durbin model estimation. Moreover, as reported in the discussion of the empirical results, evidence of omitted relevant spatially correlated variables was not detected by the spatial diagnostics carried out on the residuals.

In the base specification, research and development expenditure was also used (as an alternative to patent counts); however, research and development at the regional level is only available for different years in each country. Note that the correlation coefficient between patents and research and development is equal to 0.82.

An indicator was also used for a ‘low’ level of human capital (that is, the share of the population that has attained at most a primary level, ISCED 0–2). As expected, it shows a negative and significant influence on TFP. Moreover, its inclusion reduces the significance of the social capital variable.

Starting from the seminal contribution by Aschauer (1989), the literature has investigated the role of infrastructure, and more generally of public capital, on regional performances. See, among others, Eberts (1990) for the Unites States and Marrocu and Paci (2010) for the Italian regions; for a useful survey, see Gramlich (1994).

Detailed results are not reported here to save space, but are available in Dettori et al. (2008).

The same kind of results are obtained for all the specifications reported in Table 3. All detailed results are available from the authors upon request.