Supporting agro-food enterprises in Piedmont: a counterfactual analysis

This paper aims at analysing the role played by the European Rural Development Programme (ERDP) in supporting the agro-food industry in the Piedmont region (North-West Italy), which represents that part of the agricultural production chain characterised by the highest added value. This is a first attempt to extend the in itinere evaluation (Milanetto et al., 2011) to an ex post quasi-experimental counterfactual evaluation of the net impact of the subsidy in the medium term. This is particularly fundamental, since the effect of this kind of subsidy becomes reasonably apparent only after a few years from its supply. Since the Piedmont agri-food industry is characterised by a wide variety of firms, and the in itinere evaluation established that the treated firms do not share the average characteristics of the agri-food population, the proper counterfactual group has to be carefully selected by matching methodologies. Balance sheet data are extracted from AIDA database and carefully pre-processed in order to avoid distortions. This paper compares the performance of different matching techniques, investigating the effect of their adoption on the final estimate of the subsidy net effect. In particular, pros and cons of the Coarsened Exact Matching algorithm are considered, which is a recent monotonic imbalance-reducing matching method (Iacus, King, and Porro, 2012). It is computationally fast, robust to measurement errors, and it allows an internal identification of the common support. The performance of CEM is compared with other more conventional algorithms (nearest neighbour, caliper, radius, kernel). Since our current interest is most of all methodological, data are limited to the longest available time series: the outcomes of the beneficiaries in 2006 are compared to the matched controls for the period 2006-2012. The final results do suggest a stabilizing effect of the subsidy in a period characterized by a sever worldwide economic crisis. However, since results are quite uncertain, we expect that on-going further research (on the data-base, the model, and balance sheet indicators) will lead to stronger conclusion on the effectiveness of the policy. Nonetheless, this exercise already shows that the selected matching set and methodology, the chosen timing, and the quality of the available data do strongly influence the impact analysis.