The dataset covers 21 Italian NUTS-2 regions over the period 1995–2022. Data on GDP, manufacturing value added, total value added, and population are taken from the Italian National Institute of Statistics (ISTAT). Data on the EAFRD and ERDF are obtained from the European Commission's Rural Development database. Table 1 provides descriptive statistics for the variables employed in the analysis.

To estimate the dynamic effects of EAFRD and ERDF expenditures on regional GDP per capita, we combine a two-step instrumental variable (IV) strategy with Jordà's (2005) local projection (LP) method.

This framework allows us to trace both short- and medium-term effects of the EU funds on regional GDP per capita while addressing the potential endogeneity of fund allocation. In particular, since funding decisions may depend on regional economic performance, business cycle conditions, or unobserved institutional characteristics, relying on ordinary regression could lead to biased estimates. By using an IV strategy, we isolate the exogenous component of EU funds and get consistent estimates of their dynamic response.

We begin by employing a standard empirical approach to estimate the heterogeneous effects of structural funds across regional groups—in our case, between Southern and Central-Northern Italy. The second-stage estimation is specified as follows:

For k = 0, 1,2,3,4,5

where is the log of GDP per capita in region i at time t + k. and are region and time fixed effects, controlling for time-invariant regional characteristics and for the shocks common to all regions in a given year, respectively. South is a dummy variable equal to 1 for southern regions and zero otherwise, which allow us to estimate the heterogeneous effects between South and Centre-North. denotes the exogenous component of log EAFRD (or log of ERDF) expenditure per capita, obtained from the first-stage regression. is the control variable, which includes 2 lags of the dependent variable and our shock and its interaction with the south dummy to control for regional economic dynamics.

The identification of exogenous changes in the structural funds expenditure closely follows Brueckner et al. (2023), who deals with local government expenditure. We apply this approach to the structural fund expenditure and proceed as follows. We regress the SF spending on regional GDP per capita for the south and north and extract the residual (shock), which isolates the exogenous component by purging the contemporaneous regional economic influences:

where is the log of GDP per capita, instrumented using the interaction between the log of oil price and the share of regional manufacturing value added, following Brueckner et al. (2023):

= log ) (3)

where log ) is the log of oil Price, is 1995 regional manufacturing value added, and is 1995 total value added. The rationale behind our instrument is straightforward. Oil prices are key drivers of business cycle fluctuations and have a significant impact on GDP, while remaining exogenous to structural funds policy. Regions with a larger manufacturing base tend to be more sensitive to energy price movements, thereby introducing exogenous variation in exposure to oil shocks. By interacting oil prices with the share of manufacturing value added in a pre-sample year—thus avoiding potential endogeneity arising from the influence of structural funds policy on the manufacturing share—we construct a source of exogenous variation in regional GDP per capita. This measure is employed in Eq. (2) to address the endogeneity of regional output. The residual from Eq. (2), , provides a series of changes in structural funds expenditures that are exogenous to regional business cycle fluctuations.

Figure 1 (Panels A and B) and Table 3 (in the Appendix) represent the Impulse Response Functions (IRFs) of the regional GDP per capita to EAFRD and ERDF spending, for the Centre–North and the South regions.

Starting with EAFRD (Fig. 1, Panel A, and Table 3 in the Appendix), the response of the regional GDP per capita for both northern and southern regions is positive and statistically significant over the time horizon. In Central-Northern, GDP per capita increases by approximately 4% after the shock, with the effect remaining positive and around 3% after five years. This indicates a strong and persistent impact of EAFRD expenditure on regional economic performance. For the southern regions, in the short term, the effect is similar to the northern regions (4% after the shock) but somewhat smaller over time. In particular, five years after the shock, the estimated gain in GDP per capita is approximately 2%. Our findings are in line with Fratesi and Perucca (2019) and Crescenzi and Giua (2020), who highlight that rural development funds tend to have slower and more context-dependent impacts, particularly in areas with weaker institutional capacity or lower economic density.

For the ERDF expenditure (Fig. 1, Panel B and Table 3 in the Appendix), our results show a short-term gain on regional economy of about 1%. However, the graphs demonstrate that their persistence effects differ; in the Central-Northern, the effect remains statistically significant even after five years (about 1.1%), whereas in the South it becomes insignificant in the medium term. These findings confirm and extend the existing literature on the effectiveness of EU SFs in promoting regional growth, but ERDF produces more durable effects, particularly in the North. In comparison with Coppola et al. (2023), who find that ERDF is particularly effective in more industrialized regions through its impact on industry, our findings extend their work by demonstrating that these gains are not only sectoral but also persist over time in overall GDP per capita. Compared to earlier static panel analyses (i.e., Aiello and Pupo 2012; Pellegrini et al. 2013), which report weaker or short-lived effects, our dynamic approach reveals more persistent impacts, especially in the North.

Taken together, our results highlight two key patterns. First, both funds have a positive effect on the regional GDP per capita, especially in the Centre-North, reflecting its persistent alignment with industrial structures. Second, Southern regions experience short-term gains from the ERDF funds, and their effects fade more quickly, suggesting that regions with weaker institutional setting limit the long-term impact of cohesion policy. However, applying this standard approach to disentangle the effects between the North and the South, we find that the heterogeneous impacts of structural funds are limited. Table 2 reports tests for differences in the estimated effects across the two areas, indicating that these differences are not statistically significant. Consequently, we cannot conclude that structural funds have heterogeneous effects between Southern and Central-Northern Italy. This finding motivates us to move beyond the conventional framework and adopt a novel approach that highlights the impact of structural funds on the distribution of regional per capita income.

Insert Table 3 here

We proceed by investigating the effects of SF on income convergence, following De Groot et al. (2023), applying a location-scale model. As already anticipated, the F-tests in Table 2 shows that regional differences are not statistically significant, motivating us to go beyond mean effects and examine distributional dynamics. Previous studies (i.e., Dall'Erba and Le Gallo 2008; Aiello and Pupo 2012) found that while SFs contributed to growth, they did not necessarily reduce regional disparities. Our findings complement this literature that EU funds, ERDF in particular, seems more effective than nationally funded programs (Del Bo and Sirtori 2016), possibly due to better targeting or stronger objectives.

Insert Table 2 here

To address this issue, we follow Frangiamore et al. (2025) and extend our baseline model by incorporating the location-scale framework of Machado and Silva (2019) within an IV–local projection setting. This approach enables us to estimate the effects of structural funds not only on the location (i.e., the average) but also on the scale (that is, the standard deviation) of the regional per capita income distribution. Similar methodologies have been employed in other contexts—for instance, De Groot et al. (2023) apply the location-scale model to examine the distributional effects of monetary policy shocks across European regions. Building on this methodological precedent, our study applies the framework to EU Cohesion Policy, an area where distribution-sensitive analyses remain relatively scarce (Fratesi and Perucca, 2019). The model is as follows:

The first part of the model estimates the effect of SFs on the expected level of regional GDP per capita (location), which is captured by the coefficient . By contrast, the second part captures the impact of structural funds on the scale of the GDP per capita distribution, estimated by the coefficient . To clarify the implications of the signs of the location and scale parameters for the distribution of regional GDP per capita, Fig. 3 presents an illustrative example that depicts how the distribution of the outcome variable changes following an increase in SFs expenditure. The example assumes a positive location effect, consistent with the results presented in the previous section. Panel (a) demonstrates that, in the absence of scale effects, an increase in SF expenditure shifts the entire distribution uniformly to the right, implying equal movement across the distribution, including the tails. In contrast, panels (b) and (c) illustrate how a non-zero scale parameter alters the shape of the distribution. A negative scale effect (panel b) reduces dispersion, indicating convergence: the left tail (representing regions with very low GDP per capita) shifts more markedly to the right than the right tail. Conversely, a positive scale effect (panel c) widens the distribution, suggesting divergence. This example highlights the capacity of the empirical strategy to uncover the distributional consequences of SF expenditure across regions. The estimation procedure follows the methodology proposed by Machado and Silva (2019), with the location and scale equations estimated via instrumental variables, as in Frangiamore et al. (2025).

Figure 2 (Panels A and B) show the results of the location-scale model. The left panels present the location effects, while the right panels display the scale effects.

For EAFRD (Fig. 2 Panel A, left chart), the result shows that the location effect is positive and statistically significant during the time horizon. We find that, in the short term, there is a jump increase in regional GDP per capita of about 2.7 percent, and after 5 years, reaching about 1.8 percent, while the scale reduces significantly only two years after the shock. This indicates that EAFRD expenditures contribute to increasing the average regional GDP per capita over time while mildly reducing disparities in GDP per capita across regions (right chart). These findings are consistent with Loddo (2010), who also reports that agricultural funds have short-term positive impacts that tend to diminish over time, suggesting that while rural funds stimulate growth, their convergence potential is more limited compared to other findings.

For the ERDF, shown in Fig. 2, Panel B, the results reveal a similar pattern to that observed for EAFRD, but with notably stronger distributional effects. The location effect (left chart) is positive and statistically significant, indicating that ERDF spending increases regional GDP per capita by approximately 0.8 percent after five years. More importantly, the scale effect is both more substantial and more persistent than that of EAFRD, implying a pronounced reduction in regional income disparities. Strikingly, the ERDF leads to a measurable decline in inequality—about 0.4 percent—just one year after the policy shock. This evidence aligns with the example depicted in Fig. 3, Panel (b), where negative scale effects compress the distribution by disproportionately boosting the lower tail. In other words, ERDF expenditure has a particularly strong effect on the poorest regions, accelerating their convergence toward the income levels of more developed regions. These results underscore the powerful role of ERDF not only in promoting overall economic growth but also in fostering regional equity through targeted structural investments.

Insert Table 4 here

A novel contribution of this paper lies in its application of the location-scale model, which captures not only mean effects but also changes in the distribution of regional GDP. While much of the prior literature has focused on average impacts, our results demonstrate that cohesion policy—especially ERDF—also plays a role in reducing territorial inequalities, thus supporting the EU's broader cohesion objectives. This methodological innovation offers a more nuanced picture of policy impact and addresses calls in the literature for evaluation tools capable of detecting distributional changes (Machado and Silva 2019; Rodríguez-Pose and Garcilazo 2018).

We did several robustness checks for our estimation. First, we check the sensitivity of our results by changing the number of lags. Fig. A1, in Appendix B, shows that our findings remain stable across different lag specifications. Second, we estimate the influence of potential outliers by removing observations in the top and bottom 1 percent of the distribution for the regional GDP per capita.

Fig. A2, in Appendix B, confirms that the results are consistent with the baseline. Third, we add additional controls motivated by growth theory. In particular, we add the lagged log investment share of GDP which is as a scale-free proxy for capital accumulation and investment-cycle dynamics and lag of the regional population which capture evolving market size and agglomeration forces documented in urban and regional growth research (Mankiw et al. 1992; Crescenzi and Giua 2020; and Kremer et al. 2021 among others). Fig. A3-A4, in Appendix B, confirm that our results are robust to the baseline.

Overall, our findings indicate that while both EAFRD and ERDF promote economic growth, ERDF is more effective in fostering convergence by reducing regional disparities, particularly in the short term. These results highlight the importance of tailoring cohesion policy to regional economic structures and strengthening institutional capacity to maximize long-term impacts, especially in the Mezzogiorno.