Productivity and spatial proximity: evidence from the Italian food industry

ABSTRACT

The aim of this paper is to examine the existence of productivity spillovers across firms in the Italian manufacturing food industry. To this end, using a sample of Italian food firms over the period 2008–2015, first the total factor productivity (TFP) is computed by considering the Levinson-Petrin approach and then a spatial econometric model is employed. Results show evidence of productivity spillovers across firms due to spatial proximity.

KEYWORDS:

• Italian food industry
• productivity spillovers
• spatial econometrics

JEL CLASSIFICATION:

• O30
• C21
• D22

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. The data sample was restricted to active firms with a known number of employees over the period 2011–2015 and a known value of value added over 2008–2015. Observations with value added and employees below and upper 1 and 99 percentiles, respectively, were eliminated. Also, observations with the growth rate of value added and employees below and upper 1 and 99 percentiles, respectively, as well as those with missing values, were dropped. A balanced panel of Italian 1,746 firms in the food processing industry over the period 2008–2015 was, then, obtained.

2. A random-effects model is also considered, with ${\mathrm{v}}_{\mathrm{i}\mathrm{t}}={\alpha }_{\mathrm{i}}+{\mathrm{u}}_{\mathrm{i}\mathrm{t}}$ denoting the composite error term.

3. Cost of labor is deflated by considering consumer price index for families of workers and office workers from ISTAT.

4. The stock of patents is 1-year lagged since there is likely to be a time lag between the innovation and its effect on productivity.

5. Information source on patents provided by the Orbis dataset by Bureau van Dijk is the European Patent Office’s (EPO) PATSTAT database.

6. Due to availability of the IQI variable, computed by Nifo and Vecchione (2014), Nifo and Vecchione (2015), in these estimations, the 2008–2012 period is considered.

7. The stock of patents is computed for each firm in the sample according to Equation (3). In the computation of the average patent stock in each province, the period t-2 is considered; moreover, firm i is excluded.

8. http://clisun.casaccia.enea.it/Pagine/Comuni.htm (last accessed: April 2018). For the towns not included in the previous dataset, information was taken from http://www.tuttitalia.it/. Also, firm coordinates are replaced by municipal ones when there is a difference between the two values higher than 1 or lower than −1. In case of duplicated coordinates, in order to compute the inverse distance matrix, we add a small value (0.01 to 0.04) at longitude or latitude.

9. The STATA command spmat (Drukker et al, 2013) is used to compute the haversine distance matrix. With row-standardization, the rows sum to one. This transformation is used in most related firm-level literature (e.g. Baltagi, Egger, and Kesina 2016; Lamieri and Sangalli 2013). Spatial lag model with fixed effects is estimated by quasi-maximum likelihood estimators employing the STATA command xsmle developed by Belotti, Hughes, and Mortari (2017).

10. Estimates are obtained through the levpet command in Stata (Petrin, Poi, and Levinsohn 2004). Material costs and value added are deflated by using food sector production price indices from Eurostat. As regards the tangible fixed assets, data have been deflated by using the average production price indices from Eurostat of the following sectors: machinery and general equipment, electrical machines and electrical equipment, electronics and optics and motor vehicles.