A New Approach to Measure Italian Regional Trade Flows with Administrative Micro Firm-Level Data

Abstract

In this paper, we present a novel approach to measure domestic bilateral trade flows in intermediate and final consumption in the Italian regions. We reconstruct regional trade flows in final consumption by using administrative data from tax returns that limit the issues of survey-based measures. We also investigate the main determinants of domestic regional trade flows in Italy by adopting a spatial gravity modeling framework that allows for the inclusion of multiple spatial dependence effects. Moreover, we provide evidence of the presence of geographical heterogeneity and a specific focus on trade flows in the manufacturing sector. Our findings, which are robust to alternative specifications, suggest that the introduction of different sources of spatial dependence is relevant to understand the occurrence of multiple types of network effects when considering origin, destination, and origin–destination linkages in trade flows. We also document regional differences in the spatial concentration of trade flows in intermediate and final consumption. Finally, we discuss the main implications and future avenues of inquiry of our research.

Appendix

1.1 A.1 Dataset description

See Tables 9, 10, 11

Table 9 Italian regions

Table 10 NACE rev.2 Level 1

Table 11 CPA 20 labels

1.2 A.2 Reconstruction of B2C Flows: Additional Information

We report below the different sub-steps used to reconstruct bilateral trade flows in final consumption from administrative VAT data.

Step 1: Transformation of VT data from fiscal domicile to plant location of seller

In the first step, a bridge matrix derived from IRAP (Regional tax on productive activities) database²⁰ is applied to VT data $vt_{od,s}^{domicile}$ for the transformation from location by tax domicile to the location by the production plant. In detail, we use the information about the (net) value of production provided by fiscal domicile and plant location. This operation is conducted by calculating a re-proportioning coefficient $\xi _{o,s}$ for each origin region o and sector s, which is the ratio between the value of net production per plant $VNP_{o,s}^{plant}$ and the value of net production per domicile $VNP_{o,s}^{domicile}$.

$$\begin{aligned} \xi _{o,s} = \frac{VNP_{o,s}^{plant}}{VNP_{o,s}^{domicile}} \end{aligned}$$

(11)

Then, this coefficient is multiplied by the taxable transactions $vt_{od,s}^{domicile}$ obtaining a re-proportioned value of the total amount of taxable transactions $\sum _d vt_{od,s}^{plant}$ in the region o and each sector s.

$$\begin{aligned} \sum _d vt_{od,s}^{plant} = \sum _d vt_{od,s}^{domicile} \xi _{o,s} \end{aligned}$$

(12)

Finally, these values are distributed for each pair of regions od according to the weight of the taxable operations in the initial data $vt_{od,s}^{domicile}$.

$$\begin{aligned} VT_{od,s}^{plant} = \frac{vt_{od,s}^{domicile}}{\sum _d vt_{od,s}^{domicile}}\sum _d vt_{od,s}^{plant} \end{aligned}$$

(13)

Step 2: Converting VT Data From Activity Sector to Commodity

In a second step, an additional bridge matrix $\Phi$, derived from national accounting data,²¹ is applied to VT data by plant and sector $VT_{od,s}^{plant}$ to distribute them by commodities i. $\Phi$ is a bridge matrix with sectors s and commodities i as rows and columns, respectively. Each value $\Phi _{is}$ of this matrix represents the national output of commodity i produced by sector s, i.e., $Output_{nat,is}$ with respect to the total output of commodity i:

$$\begin{aligned} \Phi _{is} = \frac{Output_{nat,is}}{\sum _i Output_{nat,is}} \end{aligned}$$

(14)

Hence, we obtain the estimated output of commodities for each region.

$$\begin{aligned} VT_{od,i}^{plant} = \sum _{s} VT_{od,s}^{plant} \Phi _{is} \end{aligned}$$

(15)

Step 3: Cleaning VT Data From Margin Bias

In a third step, once the VT data framework by plant and commodity for each region are obtained, we need to correct for margins’ bias on electricity, trade, and transport. These three sectors represent a subset $g \in I$, where I is the commodities’ set. The VT data, by construction, attribute to margins’ commodities g some taxable transactions which actually belong to other commodities $i \ne g$, e.g., the agriculture commodity produced by region o and transported and sold to region d is attributed to the transport commodity instead of agriculture commodity.

To avoid this bias, we have to identify for each region o and commodity g, how much of the value of the taxable transactions (linked to g) should be attributed to margin and how much to the amount of produced goods and services. This operation is conducted by using data on national trade, transport margins²² and national output.²³ Then, we re-distribute the identified margin values to the commodities object of the effective trade flow.

The margins (labeled as Mar in the formulas) identification is obtained by applying shares $\sigma _g$ to the VT data on plant and commodity. $\sigma _g$ defines for each commodity g the amount of output to be considered as margin (e.g., 14%, 84%, and 15% of the electricity, trade, and transport commodities are margins and have to be allocated among all the other commodities as flows). These shares are calculated as follows:

$$\begin{aligned} \sigma _g = \frac{Mar_{nat,ii}}{\sum _s Output_{nat,is}}, \forall i = g \end{aligned}$$

(16)

This step leads to the estimation of the margins $M_ {od,g}$ for the three commodities g, which indicate the amount of sales of electricity, trade, and transport that has to be reallocated between all the other commodities.

$$\begin{aligned} M_{od,g} = VT_{od,i}^{plant} \sigma _g \end{aligned}$$

(17)

At this stage, we need to reallocate these quantities $M_{od,g}$ across all the other commodities. Therefore, we define $\eta _{ig}$ which represents the distribution of margin g among all the other commodities i. These values are calculated as follows:

$$\begin{aligned} \eta _{ig} = - \frac{Mar_{nat,ig}}{\sum _i Mar_{nat,ig}} \end{aligned}$$

(18)

Now we apply these percentages to the margins $M_{od,g}$, in order to reallocate the margins of commodity g to all the others as follows:

$$\begin{aligned} \Delta M_{od, ig} = \eta _{ig} M_{od,g} \end{aligned}$$

(19)

Hence, we correct the VT data by plant and commodities by reallocating the margins:

$$\begin{aligned} {\hat{VT}}_{od,i}^{plant} = VT_{od,i}^{plant} + \sum _g \Delta M_{od,ig}, \end{aligned}$$

(20)

where ${\hat{VT}}_{od,i}^{plant}$ represents the value of the trade flow of commodity i from region o (by plant) to region d (i.e., source-side).

Step 4: Reshaping the point of view from seller side to buyer side

At this stage, we need to change the point of view by switching from the source side to the destination side (i.e., ${\hat{VT}}_{od,i}^{plant} \rightarrow {\hat{VT}}_{do,i}^{plant}$).

Finally, to make the estimation of administrative data coherent with the national accounting data, we transform ${\hat{VT}}_{do,i}^{plant}$ in shares $\tau _{d,i}$. These shares are then applied to aggregated regionalized household expenditure by CPA²⁴ net to imports linked to households.²⁵

$$\begin{aligned} \tau _{d,i} = \frac{{\hat{VT}}_{do,i}^{plant}}{\sum _o {\hat{VT}}_{do,i}^{plant}} \end{aligned}$$

(21)

$$\begin{aligned} {\hat{VT}}_{do,i}^{plant} = (HH_{d,i} - Import_{d,i}^{HH}) \tau _{d,i}, \end{aligned}$$

(22)

where $HH_{d,i}$ is the households’ consumption in the destination region d.²⁶ We obtain the flows of the household’s final consumption by destination region and commodity ${\hat{VT}}_{do,i}$ which represent the response variable for our analysis, namely, $Y_{hh}$.

1.3 A.3 Estimation of B2B Flows: Additional Results

We report below additional information regarding the variables used to correct B2C trade flows in order to obtain B2B trade flows, as explained in the main text (Tables 12, 13).

Table 12 Origin–destination pairs inter-regions interaction

Table 13 Intra-commodities trade condition

1.4 A.4 Sensitivity Analysis

See Table 14.

Table 14 Lagrange multiplier test