Explaining Total Factor Productivity at Firm Level in Italy: Does Location Matter?

Abstract

This study analyses how firms' internal variables and regional factors affect total factor productivity (TFP) of Italian manufacturing firms. Due to the hierarchical structure of our data, we employ a multilevel model that allows for a clear distinction between firm and region-specific effects. Results refer to 2004–2006 and show, as expected, the importance of firm-specific determinants of TFP. At the same time, they indicate that location matters in the sense that the context where firms operate plays a crucial role in determining the level of TFP. In more detail, we find that the regional endowment of infrastructure, the efficiency of local administration and the investments in R&D exert a positive effect on firms' performance.

Abstract

Cette étude analyse la façon dont les variables internes des entreprises et des facteurs régionaux affectent la productivité totale des facteurs des firmes manudacturières italiennes. En raison de la structure hiérarchique de nos données, nous employons un modèle à multi-niveaux qui distingue clairement les effets spécifiques à l'entreprise des facteurs propres à la région. Les résultats se rapportent à 2004–2006 et montrent, comme prévu, l'importance de facteurs déterminants spécifiques à l'entreprise de la productivité totale des facteurs. En outre, ils indiquent que le lieu est important, car le contexte dans lequel les entreprises exercent leurs activités joue un rôle essentiel sur la détermination du niveau de productivité totale des facteurs. De façon plus détaillée, nous estimons que la dotation régionale de l'infrastructure, l'efficacité de l'administration locale et les investissements dans la recherche et le développement (R&D), exercent un rôle positif sur le rendement des entreprises

Extracto

Este estudio analiza cómo los factores regionales y las variables internas de las empresas afectan a la Productividad Total de Factores (PTF) de empresas italianas de fabricación. Debido a la estructura jerárquica de nuestros datos, empleamos un modelo de múltiples niveles que facilita una distinción clara entre la empresa y los efectos específicos de la región. Los resultados hacen referencia a 2004–2006 y muestran, como se preveía, la importancia de los determinantes de PTF específicos de las empresas. Al mismo tiempo, indican que la ubicación importa, en el sentido de que el contexto de operación de las empresas desempeña una función crucial en determinar el nivel de PTF. Con más detalle, descubrimos que el legado regional de infraestructura, la eficiencia de la administración local y las inversiones en I + D ejercen un efecto positivo en el rendimiento de las empresas.

摘要

本研究分析了企业的内部变量和区域要素如何影响意大利制造企业的全要素生产率 : (TFP)。由于数据的层级结构 , 我们采用能够明确区分企业和区域特定影响之间的差别的多层模型。结果涉及 2004 年到 2006 年, 并且正如所料, 证实了企业特定的 TFP 决定因素的重要 性。同时 , 它们还表明了地理位置具有重要意义 , 在某种程度上来说 , 企业运营的地点在决定 TFP 水平方面起着关键作用。更为详细地 , 我们发现当地的基础设施建设力度、地方行政效率和对研发的投资都会对企业的绩效产生积极影响。

Keywords:

• Manufacturing firms
• total factor productivity
• Italian regional divide
• multilevel models

JEL classification:

• L60
• R11
• C31

Notes

1. The survey covers a sample of firms with 11–500 employees and all firms with more than 500 employees. The Xth Capitalia-UniCredit survey questionnaire refers to 2004–2006 and contains information on firm structure, ownership, workforce and investments in physical and technological capital, as well as the degree of internationalization. Data from balance sheets refer, instead, to 1998–2006. It is worth noting that the unit of analysis in Capitalia-Unicredit survey is the firm and no information is reported on the number of each firm's establishments. While this limits us to having to make the most of the heterogeneity within the micro-data, there is no possibility of using plant-data level when the source of data is the Capitalia-Unicredit survey. Thus, results have to be interpreted cautiously, although it is also important to bear in mind that more than 50% of our data-set is formed by small-sized firms which are probably single-plant firms (see Table 1).

2. Although the original Capitalia-Unicredit data refer to 5,100 firms, we use a sample of 3,000 firms which is obtained after carrying out a data cleaning procedure. The firms with negative values of value added have been eliminated from the original archive. Moreover, in order to eliminate outliers, firms with a growth rate of value added and of employees below the first or above the 99th percentile of the distribution have also been eliminated. Finally, when building the sample used in estimating TFP, we excluded firms for which at least 7 years data regarding the number of employees was not available. It is important to point out that the distribution of our sample overlaps with the distribution of the original data-set when aggregating firms by macro-area, region and sector. On the other hand, in the sample the proportion of medium-large firms is higher than that observed in the original data-set, although small firms still make up more than 50% of the sample. Finally, we have also checked the representativeness of our sample by comparing the sample and data regarding the national manufacturing industry which are provided by ISTAT, the National Institute of Statistics. From this comparison, it emerges that northern firms are slightly over-represented in our sample, especially firms located in Lombardy, while the opposite holds for firms located in southern regions. With regards firm size, the sample is somewhat unbalanced in favour of medium and large firms.

3. Labour productivity is calculated as the weighted average of firms' productivity, using as weights the firm's value added with respect to the group of reference (the whole sample or the value added of the area in the case of averages relative to the territory).

4. The results for science-based firms operating in the South are not really interpretable because there are only seven such firms.

5. The generalized improvements in TFP observed for 2002–2006 can mainly be explained by the performance of firms operating in specialized suppliers (3% in Italy, 4% in the North-West and 3.7% in the North-East), while the influence of science-based sector was minimal. In the South, the positive impact in specialized suppliers is neutralized by the dynamics of the traditional sector, whose TFP declines. When it is combined with the sectorial composition of macro-region economies (where, in the South, traditional sectors weight more than in the other areas of the country), this evidence helps to explain the effect on TFP (data are available upon request). Another potential reason beyond the recovery of 2001–2004 TFP in the South might be the boom in exports from Abruzzo, Molise, Campania and Basilicata which, at the beginning of 2000s (ICE, 2002), performed even better than the national average, but returned to their normal trend after a few years. More in general, the increase in TFP depends on the restructuring process in the Italian productive system which occurred after the introduction of the euro and the increasing entry of new competitors into the international market (Bugamelli et al., 2009). In the South, this process was less intense (Cannari et al., 2009) than elsewhere and this might explain the territorial technological differences found in this study.

6. In the random effect specification, smaller groups have a smaller impact on the estimation results than larger groups (Snijders & Berkhof, 2008). This approach recognizes that there is little information for small groups by ‘shrinking’ their residual estimates towards zero and, therefore, pulling their mean towards the overall mean (Bickel, 2007).

7. A breakdown of firms' distribution by region is provided in the appendix (see Table A1).

8. In multilevel models, when level-one coefficients are permitted to vary across groups, the number of groups, not the number of level-one observations, is used to test the significance of level-one slopes. Consequently, cross-level interaction terms, which emerge in the full specification where slope coefficients are also allowed to vary across regions, are likely to have unstable coefficients and uncertain inferential properties unless there is a comparatively large number of cases at both levels one and two (Bickel, 2007).

9. The possibility to employ contextual factors (Z_j) to explain variability in random components is the main difference between the multilevel model and random coefficient regression.

10. When replicating the estimation of equation (7) using the TFP obtained from the application of Levinsohn and Petrin for the years 2004–2006, the results do not change significantly from those displayed in Table 4. This table uses as dependent variable the TFP estimated over the period 1998–2006 and expressed as average of the last three years as a dependent variable. More importantly, grounds of the reasons for this choice refer to the commonly-shared result according to which all proxy estimators, including Levinsohn-Petrin, require a reasonable panel as they are based on the use of lagged values which, typically, going back more than one year. This is an important reason to retrieve TFP after estimating the production function by using a long span period.

11. Equation (7) may suffer from omitted variable problems since unit heterogeneity is not considered. One way to allow for unobserved heterogeneity is the fixed effects model. However, panel data analysis cannot be performed due to the lack of time series in variables such as white collar share and exports. With regards endogeneity the specification of the multilevel model used in this analysis includes three TFP determinants defined at regional level which, acting as exogenous factors, limit the endogeneity issue, because it is unlikely that individual firms' decision may alter them, while vice versa holds. However, in order to increase the reliability of estimates, we perform the test proposed by Mundlack (1978) and applied to the case of Italian provinces by Fazio & Piacentino (2010). This test consists of augmenting the model with the group means of the level 1 explanatory variables. We find that the parameters associated with these group means are not significant and, as pointed out by Grilli & Rampichini (2006), this outcome suggests that there is indication of no endogeneity bias (results are available upon request).

12. One of the basic empirical facts relating to productivity is a strong positive association between productivity and exporting activity and, therefore, we include the share of exports in total sales among firms' characteristics (Melitz, 2003; ISGEP, 2008). Similarly, it is widely argued that a firm's performance improves as a result of its innovative behaviour and in the presence of skilled workers (see, i.e., Krueger & Lindahl, 2001; Sveikauskas, 2007).

13. The Cnelstats database, built in cooperation with the Guglielmo Tagliacarne Institute, provides both information and statistical indicators on economic trends, the productive network and social situation for Italy and the EU countries (http://www.cnel.it/cnelstats/index.asp).

14. See note [1] above.

15. Looking at the index proposed by Golden & Picci (2005), we note that northern regions exhibit less corruption than southern ones and, thus, their index is in many cases above unity (Table 2). There are two exceptions, Liguria and, to a lesser extent, Valle d'Aosta. The Liguria discrepancy is due to unmeasured aspects of its orography: Liguria houses a large population on a very narrow strip of land between the mountains and the sea. In such a setting, public works construction often requires daunting and expensive techniques. Valle d'Aosta, on the other hand, is remote and small (Golden & Picci, 2005).

16. We augment Model 2 by using the dummy South. Thus, Model 3 includes the variables at firm level, the varying regional intercepts and the variable South, and is meant to provide some evidence on the following issue: does an unobservable South effect remain even after controlling for regional non-observed heterogeneity? In other words, the idea is to understand the relevance of non-observable heterogeneity at individual and at macro region levels

17. Clustering data at regional level relaxes the assumption of independence and, therefore, increases the error term to accommodate the lack of independence of firms within regions. However, while clustered OLS leaves both the noise associated with difference between firms and noise associated with differences between regions in the error term, the multilevel model goes further by allowing these two error components (see equation 4) to be separated.

18. The multilevel analysis was implemented using the ‘xtmixed’ routine of STATA. All models were estimated using restricted maximum likelihood (REML) over maximum likelihood (ML) since the latter is more sensitive to loss of degrees of freedom when dealing with a small number of groups (Bickel, 2007).

19. See note [16] above.

20. The null hypothesis is that or that there is no random intercept in the model. If the null hypothesis is true, an ordinary regression can be used instead of a variance-components model.

21. For Italy, our findings are in line with Fazio & Piacentino's (2010) results at the provincial level. For the Netherlands, Raspe & van Oort (2011) find that 2.3% of firm productivity can be related to location and more than 97% to internal characteristics.

22. The coefficient of determination for the two-level model is given by:

where N stands for the null model and M for the model of interest.The proportional reduction in each of the variance components can be calculated separately. The proportion of the level-2 variance explained by the covariates is:

and the proportion of the level-1 variance explained is:

23. In the literature, two hypotheses about the positive correlation between export activity and productivity are investigated. The first hypothesis is that the most productive firms self-select into foreign markets because they can overcome sunk costs associated with foreign sales (Melitz, 2003). The second hypothesis raises the possibility of ‘learning by exporting’. Firms participating in international markets acquire knowledge and technology with positive feedback effects on firms' knowledge and technology accumulation. Furthermore, firms which are active in world markets are exposed to more intensive competition than firms which only sell their products domestically. In summarizing the results achieved in this field of research, it can be said that the more productive firms self-select into export markets (Melitz, 2003; ISGEP, 2008).