Introduction

How much domestic and how much foreign value added do Italian households consume? And where does the foreign component come from?

Answering these questions is relevant for at least three reasons. The first one is the extent to which household consumption feeds into GDP. Indeed, the foreign content embedded in the goods and services bought by consumers rewards by definition foreign factors of production and therefore does not directly contribute to gross domestic product.

The second reason relates to the exposure of households, through the foreign content of consumption, to a variety of shocks originating overseas. In general, consumption of domestic productions that use local inputs intensively is more likely to be insulated from these external pressures than consumption of manufactures that rely heavily on imported intermediates (or of final imported goods). Depending on the specific shock of interest, however, the geographical origin of foreign value added (FVA) is crucial. For instance, investigating the exposure of Italian households to exchange-rate movements requires that FVA be broken down between the component originating in the euro-area, which is insulated by definition form these shocks, and FVA originating elsewhere.Footnote 1 Similar considerations hold, for euro-area monetary-policy analysis, for imported inflation shocks. In the case of trade-policy shocks, instead, the main concern—again by definition—is with FVA originating outside the European Union. Within this class of shocks, however, the researcher may need to single-out specific economies: for instance, FVA originating in the US is of the uttermost relevance in analysing the effects of EU-US trade attritions, like those emerged in 2019 over the Airbus-Boeing dispute; FVA originating in the UK is instead centre stage when addressing the potential effects of Brexit. In perspective, should the European Union introduce carbon tariffs, other specific countries may be prominent. Finally, thinking of current post-pandemic supply bottlenecks and of the inflationary pressures they are igniting, yet other specific origins of FVA may be of interest.

As for the third reason, the amount of foreign value added consumed by individual households is likely to vary with their income and total expenditure because of the heterogeneity in consumption bundles, both in terms of expenditure categories (non-tradable services vs manufactures, for instance) and in terms of “varieties” within such categories (high vs low-quality wearing apparel, for instance, or Italian car brands vs German ones). This implies, for example, that income- or demand-support programmes will have different aggregate effects depending on the households (and expenditures) targeted; it also implies that, all else equal, exchange-rate shocks or foreign-price movements or after-tariff domestic-price changes will have distributive effects through their differential impact on households’ purchasing power.

In this paper, we address these issues and offer evidence on the geographical origin of the value added embedded in Italian household expenditure, both at the aggregate level and along the distribution, focusing on exchange-rate shocks and the benefits of belonging to a common currency area, so that the pertinent distinction is that between foreign value added originating within the euro area and foreign value added accrued in the rest of the world.

In a globalized world, the foreign value added of domestic consumption is only loosely related to imports of final consumption goods. On the one hand, imports do not entirely consist of foreign value added, as they incorporate themselves a domestic value-added component, to the extent that their production requires (also) domestic intermediates. On the other hand, also domestic final consumption goods incorporate foreign value added, through the underlying foreign intermediate inputs.

We measure the foreign value added content of consumption using the World Input–Output database (WIOD). That is, we start from final consumption expenditure in Italy and trace back, following the production linkages through global value chains (GVCs), how this expenditure is distributed among the various countries’ contributions in terms of their value added. Seen in the opposite direction, we record how value added stacks up in subsequent stages of production and distribution around the globe, starting from the most upstream ones to the most downstream ones before final use by households. Our approach thus also accounts for the fact that what households actually pay for their consumption includes an array of mostly domestic services, in the form of transportation, logistics and distribution, wholesale and retail activities.

Importantly, WIOD allows splitting the foreign value added content of consumption among each of the 43 foreign economies included in the database. This is extremely relevant in light of all the above-mentioned policy questions, among which we have chosen to focus on those for which the relevant distinction is between euro-area and non euro-area origins. On the one hand, the expenditure of the Italian household sector directly propagates to income of other euro-area members through their share of value added embedded in the goods and services bought by Italian consumers; assessing such spillovers sheds light on the interconnections across member countries and on the potential effects of policies that support household income and demand at the euro-area level. On the other hand, disentangling the component of domestic consumption that is potentially affected by price developments in other euro-area countries allows, in a euro-area-wide policy perspective, to assess endemic inflationary pressures in the Eurozone. Moreover, belonging to the euro area and sharing a common currency limits the exposure to exchange-rate fluctuations exclusively to the foreign value added generated in countries outside the euro area (and, it could be argued, implies the use of a much stronger and more stable currency than a hypothetical national one had Italy not joined the monetary union).

We explore the distributive dimension of these aspects with two complementary approaches. The first is a data-driven approach, which focuses on how different households allocate their expenditure across consumption categories (e.g. food vs travel), based on the individual data from the Household Budget Survey (HBS) conducted by the Italian National Statistical Institute (Istat). The second is a model-driven approach, accounting for the fact that, within each consumption category (e.g. cars), different households consume varieties of different quality, which we proxy with the product’s geographic origin (e.g. German brands vs Chinese brands). This is a dimension that the first approach cannot capture, despite its merit of being of a quasi-accounting nature: HBS expenditure shares across COICOP categories are combined with an estimate of the FVA content within each COICOP category, in turn derived from the FVA content broken down by the CPA categories implemented in WIOD. HBS however provides no indication on geographical origin within consumption categories: in the example above, it provides the household’s expenditure share on cars, but no distinction between Italian, German and Chinese brands. WIOD tables instead tell us that the FVA (i.e. non-Italian) content is quite different among cars produced in these three economies. In order to leverage on this type of information, we need to know how households allocate their expenditure over product-origin pairs, which we attain by updating the estimates of Fajgelbaum and Khandelwal (2016).

We find that between one and two fifths (depending on the data and method used) of the expenditure of Italy’s household sector “buys” foreign value added. In other words, of each euro spent for consumption, about 60 to 80 cents accrue to domestic income, whereas about 20 to 40 cents remunerate foreign production factors.Footnote 2 Slightly less than half (around 40%) of this foreign value added originates in other euro-area countries: Italy’s participation in the monetary union overall almost halves the share of “consumption at risk”, i.e. exposed to foreign pressures trough exchange-rate shocks.

According to our results, the breakdown between the domestic and the foreign content of expenditure is broadly constant along the distribution of households, irrespective of their affluence. However, the geographical origin of the foreign component is heterogeneous: the expenditure of more affluent households flows to a larger extent to euro-area production factors.

We also find significant heterogeneity across the distribution in terms of consumption bundles: our results show that the FVA in the expenditure of less affluent households is mainly embedded in price-inelastic varieties and in necessities; the opposite happens for more affluent households. Hence, consumers at the lower end of the distribution have less margins of adjustment in reaction to, for example, external shocks that raise the price of the foreign component of their expenditure. Moreover, compared to households at the higher end of the distribution, they buy relatively more extra-euro area value added for each euro spent on consumption, so that their share of “consumption at risk” is larger: this means that they are relatively less insulated from exchange-rate movements while, as regards the level of their purchasing power, they benefit relatively more from euro-area membership thanks to the global status of the euro and to the strength and stability that go with it.

Our work speaks to both macro and micro strands of the literature, which we discuss in Sect. 2. Relative to the existing macro literature, our analysis extends the concept of import content of consumption to a more encompassing measure of foreign (value added) content of consumption. Relative to the few micro-based studies that delve with distributional issues, our paper is, to the best of our knowledge, the first to focus on Italian households. Indeed, while the Covid-19 pandemic, the recent resurgence of protectionist policies and the escalation of trade disputes have revived the academic debate on countries’ exposure to trade-induced shocks—from ‘supply chain contagion’Footnote 3 to new tariffs, to possible exchange-rate shocks due for example to Brexit, to the geographical allocation or reallocation of production stages—few quantitative studies have so far dealt with the quantification of household exposure to such risks, which we summarize by the FVA content of consumption, and with their distributional implications.

The paper is organised as follows. Section 2 exposes the major measurement challenges our analysis needs to face, and casts it in the context of the existing literature. Section 3 presents our quantitative assessment of the foreign content of consumption at the aggregate level. Section 4 first spells the details of how we bridge the aggregate picture into the micro dimension and then analyses how the content of foreign value added varies across households, depending on the composition of their consumption bundles, (i) in terms of expenditure categories, on the basis of HBS data, and (ii) in terms of the geographical origin of “varieties” within expenditure categories, on the basis of a model-based exercise. Section 5 wraps-up our main conclusions.

Measurement Issues and Related Literature

This paper broadly falls in the vast field of studies on GVCs (see Antràs and Chor 2021 for an updated and exhaustive literature review), although its focus on distributional issues is closely shared by a relatively small number of contributions.

The assessment of countries’ exposure to trade-driven shocks and, more in general, of their dependence on foreign productions, is subject to strong measurement challenges. Indeed, due to the international fragmentation of production, standard trade statistics are incapable of fully capturing cross-country supply and demand relationships. The quantitative literature has dealt with these issues through the use of inter-country Input–Output (ICIO) tables such as the WIOD or the OECD-WTO TiVA database, that allow tracking value added along the supply chain from the country of origin to the country of final use. While standard trade statistics only deal with the import content of, say, consumption, ICIO tables allow for a more encompassing measure of its foreign (value added) content.

Within the existing macro literature most closely related to our paper, a gross line can be drawn. On the one side, many authors have focused on exports and have successively refined the methodology in order to properly measure their FVA content by using ICIO tables,Footnote 4 with contributions by, among others, Johnson and Noguera (2012), Koopman et al. (2010, 2014), Nagengast and Stehrer (2016). A fully coherent framework was later put forward by Borin and Mancini (2015, 2019)Footnote 5 and popularised by World Bank (2020). Studies based on ICIO tables have mostly focused on the measurement of the domestic vs foreign value-added content of exports or on identifying the countries where exports are ultimately absorbed, or on measuring the domestic value added “activated” by another country’ final demand through its imports.Footnote 6

On the other side, a number of studies stressed the importance of taking into account also the import content of internal demand components, but mostly remained in the realm of standard trade statistics: examples include Bussière et al. (2013) and, specifically for consumption, Burstein et al. (2005), Gopinath (2015) and Hale and Hobijn (2016).

Hale and Hobijn (2016) find, on the basis of gross trade flows, that about 14% of the expenditure by US consumers goes to imported goods and services. As an example of how biased the picture depicted by standard trade statistics can be, due to not fully capturing cross-country supply and demand relationships, a later study (Hale et al. 2019) shows that, once the analysis is performed in value-added terms and the origin of value added is accounted for, the foreign content of US consumption results as being lower than previously estimated, just over 10%. Evidence for European countries is rather scant: Bourgeois and Briand (2019) quantify the foreign content of French household consumption at just below 20%; Chandler et al. (2018) focus on the UK consumer-price basket, computing an import intensity that ranges between 16 and 24% depending on price index considered (and thus on the underlying consumption basket).

Even less emphasis has been put in the literature on the distributional aspects related to the heterogeneous impact of global value chains on individual household consumption bundles, through their different foreign value-added content. Recent studies on the heterogeneous exposure of households to trade or trade-induced shocks focus either on specific events or on model-based estimations, and only consider the import rather than the foreign component of consumption. Cravino and Levchenko (2017), for example, study the impact of (large) exchange-rate devaluations on the cost of living at different points on the income distribution, whereas Fajgelbaum and Khandelwal (2016) and Atkin et al. (2018) examine the distributional impact of economic integration across consumers by modelling income elasticities and demand non-homotheticities. A few empirical papers more closely related to ours have recently been triggered by the Brexit referendum. Breinlich et al. (2017) show that the pound depreciation following the referendum led to higher inflation for products with a larger import share in consumer expenditure; they find that the inflation burden was shared evenly throughout the income distribution but unevenly across regions. In a related study, Breinlich et al. (2016) focus on how prices would react after Brexit due to changes in trade barriers, concluding that the negative consequences of higher import prices would be slightly harder for middle-income households than for the richest and for the poorest.

The Aggregate Picture

How much does the Italian economy pay out to other countries for the final consumption of its households? For correctly answering the question it is not sufficient to compute the fraction of consumption expenditure that is absorbed by final products made abroad, i.e. imported, as opposed to products that are “made in Italy”. Stated differently, it is not sufficient to consider only the direct import content of consumption. Conceptually, three more ingredients have to be included in order to quantify the foreign (value-added) content of consumption.

First, as already mentioned in the introduction, one also needs to take into account the indirect import content, namely the imported intermediate inputs used in domestically produced final goods.Footnote 7 Second, in connection with the ramification of GVCs, both the direct and the indirect import contents need to be properly measured, by acknowledging their content of Italian value added in the form of exported domestic intermediates that were used abroad to produce the goods and services that were then imported into Italy. Third, the value of imported final goods is measured at the border, while consumption expenditure is measured “at the store shelf”, hence including an array of mostly domestic services, in the form of transportation, logistics and distribution, wholesale and retail activities.

Our approach takes simultaneously into account all these three factors, since it allows disentangling the domestic and the foreign value added embedded in each consumption good and service. We use the November 2016 release of WIOD data at current prices and US dollars for the period 2000–2014; the database covers 43 countries (including the 28 European Union member states) and 59 product groups at the 2-digit CPA (version 2) classification detail. WIOD data are crucial for tracing back, by following the production linkages through GVCs, all countries that contributed (in value-added terms) to the foreign component. We compute the latter by implementing the Hypothetical Extraction approach of Los et al. (2016).Footnote 8

The distinction between the direct import component and the foreign component of consumption is quantitatively very relevant. Indeed, according to gross trade data, the fraction of imported goods and services in Italy’s consumer spending was just above 11% in 2014 (Table 1 left-hand-side panel). This figure doubles when the geographical origin of the value-added content of both domestically produced and imported final goods is taken into account: the share of foreign value added consumed by Italian households rises by around 10% points, reaching 21.1% (Table 1 right hand side panel). In other words, out of the 878.2 billion euros spent by Italian households in 2014 for final consumption in WIOD data, 186 billion were ultimately paid out to foreign productive inputs.

Table 1 Composition of Italy’s household consumption expenditure (1) (percentages)

In the 2000–2014 period, the incidence of foreign goods and services in consumption expenditure recorded a gradual increase until the outburst of the global trade collapse, diminished strongly in 2009 and recovered thereafter. The share in 2014 is close to the peak it had reached in 2008.

Around two fifths of the foreign content in 2014, namely 8.2% of consumption expenditure, can be traced back to value added that originated in euro-area countries (other than Italy). Measured in terms of gross imports, the euro-area weight is lower (6.0%), but it accounts for more than half of the import content of consumption. However, final consumption products imported from euro-area partners actually incorporate a large part of extra-euro area countries’ value added.Footnote 9

Table 2 Composition of household consumption expenditure in 2014 in the four main euro-area economies and in the US, by country of origin of value added (percentages)

Table 2 zooms into the breakdown of Italy’s consumption expenditure in 2014 by country of origin of value added, isolating the most important contributors. It also provides a comparison with the corresponding figures computed for the other three major euro-area countries. Domestic value added accounts for a slightly lower share of household expenditure in Germany and France than in Italy and Spain; among the foreign contributors, the weight of the euro area ranges between roughly 8 and 10% across the four countries, followed by China and the US. The fraction of consumption that can be traced back, in value-added terms, to the UK is rather low (between 0.8 and 1.5%), still higher than the weight of value added originating in Russia and Switzerland.

The last row of Table 2 presents results for the US based on the computations by Hale et al (2019) we have mentioned in Sect. 2. The comparison between these figures suggests that euro-area households consume a much larger share, at least twice as large in fact, of foreign value added than US households: in Italy 21.1% against only 10.7%. Unsurprisingly, the weight of value added originating from the euro area is very modest in the case of US consumption, at 1.2%, around one tenth of the overall foreign content (in our computations, the share is only marginally higher, at 1.6%, as shown in the second-to-last row in the table).

As expected, the average share of household expenditure that can be traced back to non-domestic origins is very heterogeneous across types of goods and services. For Italy the foreign content is generally higher in manufactured goods, around 55% in 2014 (about 20% points more than in 2000) and much lower in services (11%, almost unchanged since 2000; Table 3).

Table 3 Sectoral composition and foreign content of Italy’s household consumption expenditure (percentages)

Euro-area countries account for more than 40% of the foreign value added embedded in Italian consumers’ expenditure on manufactures, consistently with the evidence provided in studies of “factory Europe” (Amador et al. 2015; Baldwin and Lopez-Gonzalez 2015). The weight of euro-area value added is instead relatively low in the category of raw materials and energy products.

Table 5 in Appendix 1 shows that, among the different manufacturing categories, the foreign component is especially large (more than two-thirds) for the expenditure in chemical and pharmaceutical products, electronics and electrical equipment, gasoline and fuels, as well as motor vehicles. Notably, expenditures in traditional “made in Italy” productions, such as textile, wearing apparel and leather products are for almost three-quarters related to external value added, whereas those in manufactured food products, beverage and tobacco for almost half. A few categories of expenditure among services are also characterized by a sizeable intensity of foreign value added: air transport, architectural and engineering activities, R&D services for more than two-thirds, activities auxiliary to financial and insurance services for about 40%. Origin wise, only for textiles, wearing apparel and leather products, within manufactures, is the weight of China almost at par with that of the euro area, though it is substantial also for electronics and for electrical equipment. UK value added accounts for a significant fraction of the expenditure in activities auxiliary to financial and insurance services and in R&D services, whereas the US is a prominent source of the foreign value added embedded in architectural and engineering services.

Towards the Micro Picture: Data, Hurdles and Methods

Using WIOD data, we have traced the value added embedded in Italy’s aggregate household expenditure back to the countries of origin. We now consider explicitly the heterogeneity across households and try to quantify how the foreign value added consumed by the household sector is distributed across different consumers. To this end, the ideal dataset—that would enable replicating at the individual household level the quantitative strategy pursued at the aggregate level in the previous section—should contain sufficiently detailed descriptions of expenditure bundles for a representative sample of households, hence allowing tracing single items to their production inputs and supplier countries. However, available microdata are far from ideal. For one thing, even at its most detailed level, data collected by household budget surveys do not allow to distinguish expenditures by country of origin and quality, two dimensions known to be widely heterogeneous across households. For example, it is impossible to tell apart expenditures on a luxury German car from the purchase of an Italian clunker. The aggregate WIOD data only partly share this limitation, in that they do not allow distinguishing products and services based on their quality. To overcome these drawbacks we follow two complementary empirical strategies, each with its own merits and limitations.

First, we adopt a purely data-driven approach (Sect. 4.2). We complement data on individual expenditures collected in the Italian household budget survey (HBS) with the product-level measures of foreign value added content developed in the previous section, and we estimate household-specific shares of foreign value added in consumption. This approach has the merit of using observed data on household expenditure but it is limited by the lack of consistency between the classification of consumption expenditure items (COICOP) adopted in the HBS and that of (CPA) products used in WIOD tables. We overcome this hurdle by making assumptions on the linkages between the two classifications, as detailed in Sect. 4.1.

Second, we adopt a model-driven approach (Sect. 4.3). We exploit the aggregation properties of a standard theoretical representation of individual demand functions to econometrically estimate household-specific budget shares from the observation of (i) the composition of aggregate consumption expenditure and (ii) the household’s relative income. Applied to WIOD data, this approach has the merit of making full use of available information. In particular, by defining a consumption item as the combination of a product category and a supplier country (à la Armington 1969) we can partly address the fact that, within product categories, households consume varieties of different quality, which we proxy with the product’s geographical origin. The merits of this approach come at the cost of making strong assumptions on household behaviour and of using a coarse product classification for consumption items.

Since these approaches combine different data sources, it is useful to start by clarifying the main discrepancies between the HBS-based and the WIOD-based composition of the household sector expenditure.

In general, HBS and WIOD use different definitions, implement different classifications for the goods and services that households consume and look at a different population of agents. We review these three aspects in turn.

WIOD data are sourced from National Accounts (NAs) and hence are broadly aligned with them.Footnote 10 In turn, NAs do estimate household consumption expenditure based on HBS, but the latter is just one among a variety of sources, which are further complemented with imputation proceduresFootnote 11; these inputs then undergo an overall re-balancing in order to guarantee internal consistency. HBS data are instead disseminated as collected and are thus subject to the usual caveats of survey-based statistics, like measurement error and under-reporting issues.

Concerning the reference population, WIOD/NAs data refer to expenditure in Italy by resident and foreign consumers, thus including the expenses of inbound international tourists. HBS targets a different population, as it collects the expenditure of resident consumers only, be it incurred in Italy or abroad (thus including the expenses of outbound national tourists).Footnote 12 We are bound to accept the assumption that both sets of data are a good proxy for the variable we are interested in, namely the expenditure in Italy by Italian households, which unfortunately cannot be disentangled given the available information.

As for classifications, WIOD provides the breakdown of consumption expenditure by CPA products, while NAs are available with a breakdown both by CPA and by 3-digit COICOP (Classification of Individual Consumption by Purpose) and HBS uses the 4-digit COICOP.

With these caveats in mind, Fig. 1 compares the composition of household consumption expenditure for Italy in NAs and HBS in 2017 across ad-hoc macro categories that we have rearranged (mainly) from the standard two-digit COICOP classification, as detailed in Appendix 2. Panel A considers all categories and compares the percentage composition of consumption expenditure in NAs with that in HBS in 2017. The starker difference is the lower share of imputed rents in NAs; this is consistent with the fact that NAs, as mentioned above, include the expenditure in Italy of foreigners, which are unlikely to be imputed rental payments. Imputed rents is, however, a category we want to abstract from: since we are primarily interested in the foreign content of household expenditure, we see as appropriate the exclusion of items that do not correspond to actual payments. Panel B provides the comparison between the two sources after excluding imputed rents: as these account, due to the high homeownership rate, for a large share of resident household consumption (over 20% in HBS), their exclusion raises by about one fourth the expenditure shares on the other items.

Fig. 1
figure 1

Household consumption in National Accounts and in the Household Budget Survey (percentages; data for 2017)

Bridging Classifications

Given the different classifications adopted by WIOD and HBS, the joint use of the two sources as in our data-driven approach requires mapping CPA products into four-digit COICOP categories. Unfortunately, no public data for bridging the two classifications is available for Italy, whereas the UK Office for National Statistics (ONS) publishes a very detailed cross-over (or bridge) matrix that allocates CPA consumption (two-digit CPA 2008) across the various COICOP categories (four-digit COICOP 1999).Footnote 13 Even if the bridge matrix we use is based on UK rather than Italian data, insofar as it can be interpreted as a “technology” that maps CPA products into COICOP items we do not expect, at least to a first approximation, major differences with the confidential analogous matrix used by Istat to compile NAs.

Notice that in this mapping neither the CPA leg nor the COICOP one display a geographical breakdown. To exemplify, in this framework German and French cars are simply “Cars”, produced using the same combination of CPA products, again irrespective of their geographical origin. This implies, in particular, that using the mapping published by ONS implicitly assumes that imported COICOP goods and services are produced according to the same technology (CPA combinations) as domestic ones.

With this caveat in mind, we compute the share sc of foreign value added in each COICOP consumption category c in HBS as follows. For each c, we combine each of the shares mp of foreign value added in CPA product p (56 product categories, computed from WIOD) with the corresponding weight qpc that, according to the UK cross-over matrix, product p bears in category c, with the property Σp qpc = 1 holding:

$$ s_{c} = \Sigma_{p} \;m_{p} \;q_{pc} $$

As for the country of origin of value added, a similar algebra holds: it suffices to replace the foreign value-added share mp with the country-specific share mpi of foreign value added originating from country i embedded in product p. The weighted sum of these shares across expenditure categories—each sc being assigned the weight of category c in overall expenditure—gives the share of foreign value added in aggregate household consumption.

Panel A in Table 4 compares the results obtained with those based on WIOD/NAs (already discussed in Sect. 3). It shows that the differences between the two sources appearing in Fig. 1 carry over to the share of foreign value added in aggregate household consumption and to its decomposition by country of origin of value added. In particular, the foreign value-added content of total consumption computed from HBS is higher than that computed from WIOD/NAs (28.7 against 21.1%) and climbs up to nearly 37% if imputed rents are excluded.Footnote 14 Of this foreign value added, less than half originates from other euro-area countries (14.7%, as opposed to 22% from the rest of the world).

Table 4 Origin of value added in consumption expenditure: WIOD/NAs and HBS (percentages)

The foreign content varies quite significantly across expenditure categories (panel B): it is nearly 70% in clothing and footwear and in ICT; between 40 and 50% in energy, transports, food and health; around one fourth in recreation and leisure activities, housing related expenditures and remaining services; it is almost nil in rents and education and in (paid and imputed) rents. Except for food and beverages, whose foreign value-added content originates almost equally within and outside the euro area, all other consumption items contain value added originating to a relatively larger extent from extra-euro area countries.

A Data-Driven Measurement

We start out by documenting the heterogeneity of consumption bundles across the household distribution using the micro dimension of the HBS. Figure 2 reports, for each of the main categories of goods and services used in Sect. 4.1, the expenditure share of households ranked according to their total equivalent consumption, for ventiles of the distribution.Footnote 15 Imputed rents represent a large share of expenditure, which decreases only mildly along the distribution due to the widespread ownership of the main residence; consistently, less “affluent”Footnote 16 households (on the left-hand side of the distribution) display a higher share of paid rents. The share of food and beverages as well as that of fuel and energy decrease with affluence (consistently with Engle’s law), while the opposite holds for the incidence of recreational and transportation goods, services (other than ICT), clothing and footwear. Once again, these patterns are amplified when we exclude imputed rents (Fig. 2, panel B), which is the option we follow in the rest of the analysis as we focus on the composition of actual, “out-of-the-pocket” expenditures.

Fig. 2
figure 2

Consumption bundles by total equivalent expenditure (percentages; data for 2017)

In summary, less affluent households display, as expected, a higher share of expenditure in necessary items, which are typically characterised by a lower direct (compensated or uncompensated) elasticity of demand to prices. For example, Rondinelli (2015) estimates that in 2012 the compensated price elasticity of food consumption was −0.5 while that of more easily avertable restaurant expenditure was nearly triple (−1.5).

To assess the heterogeneity of the foreign content of consumption across ventiles, we exploit the full micro-level detail of HBS data and combine the composition of individual consumption bundles (panel B of Fig. 2) with the share of foreign value added in each consumption category (summarized in panel B of Table 4). In detail, we compute for each household h the fraction xh of foreign value added in his overall expenditure:

$$ x^{h} = ({\Sigma}_{c} s_{c} v_{c}^{h} )/V^{h} $$

where vhc is the level of expenditure of household h on category c (at the 4-digit COICOP level), sc is the share of foreign value added in category c (as defined in Sect. 4.1) and Vh is household h overall expenditure.

Results are shown in Fig. 3, which presents the share of foreign value added by ventiles of the distribution, along with the composition of foreign value added across consumption macro-categories.Footnote 17

Fig. 3
figure 3

Foreign value added in consumption expenditure (percentages; data for 2017)

Overall, the incidence of foreign value added is broadly stable (apart from the very first ventiles) along the distribution, around 37% (panel A). A closer look at its composition by expenditure category, however, reveals striking differences across households (panel B). Two thirds of the foreign value added consumed by households in the bottom fifth of the distribution are traceable to food and beverages (42%) and energy and fuels (21%); the share of these items in foreign value added consumed by households in the upper fifth is only one third (respectively, 21 and 13%). On the contrary, more than one fourth of the foreign value added consumed by affluent households is embedded in recreational, travel and transport goods and services, against only about 5% among poorer ones.

To gauge the heterogeneity, along the distribution, of the geographical allocation of foreign value added, Fig. 4 splits the latter into that originating from euro-area countries (except Italy; panel A) and that originating from the rest of the world (panel B).Footnote 18

Fig. 4
figure 4

Origins of foreign value added in consumption expenditure (percentage shares of foreign value added in consumption expenditure; data for 2017)

Although the two components are overall broadly constant along the distribution (excluding the initial ventiles), with about 15% of value added stemming from the euro area and 22% from elsewhere (consistently with Table 4, panel A, last row), the heterogeneity by origin across consumption categories has a significant distributional dimension. Indeed, the extra-euro area value added embedded in expenditures for energy and fuels as well as for food and beverage—items characterized by a low price-elasticity of demand—has a larger share in the consumption bundle of less affluent households.

In summary, according to our micro-based computations based on HBS data, slightly more than one third of the expenditure of Italian households—once imputed rents are excluded—“buys” foreign value added; this overall share does not change substantially with individual consumption levels, but its distribution across expenditure categories and foreign origin varies dramatically. Indeed, households in the lower part of the distribution spend proportionately more on energy and food products, which not only are characterized by a larger content of foreign value added (especially for the component originating from countries outside the euro area), but are also relatively inelastic to prices.

A Model-Driven Approach

The different shares of foreign value added across households detected by the data-driven approach stem only from differences in the composition of expenditure across COICOP categories. However, while providing a highly detailed description of consumption expenditures, HBS data is likely to miss other relevant dimensions of heterogeneity across the household distribution, connected to variability within consumption categories.Footnote 19 Possibly, the most relevant such dimension is quality; for example, luxury sport cars and cheap economy ones fall in the same four-digit COICOP category. This is clearly a relevant limitation for our analysis, as less affluent households are likely to select lower-quality varieties within a given consumption category and since the foreign content of low-quality varieties is likely to be very different from that of high-quality varieties. For instance, the incidence of foreign value added in cheap garments made in East Asia is presumably much higher than the incidence in luxury made-in-Italy garments; vice-versa, the incidence in an Italian economy car may be much lower than the incidence in luxury German cars. To complicate further the picture, we notice that the foreign content in expenditure is unlikely to be monotonically related to quality also across categories, given the consumption patterns of Italian households. For example, both low-quality textiles and high-quality cars are likely to have a higher foreign content than the opposite varieties within the same categories.

To address this limitation, we draw inspiration from the empirical trade literature that has established a strong association between quality and the products’ geographic origin (for example, Fajgelbaum et al. 2011, Feenstra and Romalis 2014). Building on this notion, in the following we expand the analysis of households’ expenditure patterns to consider also the geographic origin of each product they consume by adapting to our purposes the approach proposed by Fajgelbaum and Khandelwal (2016) to study the distributive effects of trade.

Fajgelbaum and Khandelwal (2016) embed an Almost Ideal Demand System (AIDS; Deaton and Muellbauer 1980) into a standard multisector model of international trade. Relevant to our purposes, this allows estimating how consumption expenditure, on a given category and at a given point of the household income distribution, is distributed across qualities/origins; the corresponding foreign value added contents by quality/origin can then be computed from WIOD data, a dimension that we were not able to exploit in the previous section.

More formally, consider a generic consumption item j (only later shall we associate it with a product-origin pair); AIDS implies that for household h, the expenditure share on j can be written as:

$${s}_{j}^{h}={\alpha }_{j}+\sum_{k=1}^{J}\left[{\gamma }_{jk}ln({p}_{k})\right]+{\beta }_{j}ln\left[{y}_{h}/\alpha (P)\right],$$
(1)

where yh is household h total expenditure, a(P) is a well-defined price aggregator of the vector price P of individual prices \({\left\{{p}_{k}\right\}}_{k=1,\dots ,J}\), γjk is the cross price (semi)elasticity between items j and k, and βj is the income (semi)elasticity of item j, assumed to be identical across households. Parameters satisfy Σj αj = 1, Σj βj = Σj γjk = 0 and γjk = γkj. This demand system can be aggregated at the market level so that Sj, the aggregate expenditure share on item j, can be interpreted as the behaviour of a representative agent:

$${S}_{j}={\alpha }_{j}+\sum_{k=1}^{J}\left[{\gamma }_{jk}ln({p}_{k})\right]+{\beta }_{j}ln\left[{\varvec{y}}\boldsymbol{ }exp(T)/\alpha (P)\right],$$
(2)

where y is the average household expenditure and T is the Theil index of the associated distribution.

Fajgelbaum and Khandelwal (2016) embed this demand system in a simple standard international trade model in which, to start with, consumption items are identified by a product-origin pair (g,i). The authors then assume that:

  1. (i)

    exporting item (g,i) (that is, product g from country i) to country n entails an iceberg trade cost tgin (tgin > 1, tgii = 1); perfect competition then implies that prices in country n are related to prices in country i by pgin = pgii tgin;

  2. (ii)

    labour is the only factor of production and its productivity Zgi is product-country specific;

  3. (iii)

    wage per efficiency unit in country i is equalized across products at wi = pgii Zgi;

  4. (iv)

    the distribution of units of labour efficiency across households is country-specific;

  5. (v)

    households spend all their labour income;

  6. (vi)

    cross-price elasticities γ are assumed to be zero across products and constant within products across origins.

These assumptions and the model equilibrium conditions yield an estimable gravity equation relating the aggregate household expenditure share for item (g,i)—i.e. for product g from country i—in country n to the geographic and sectoral distribution of world output, to the structure of bilateral trade costs and to features of the income distribution in country n: for the exact specification, please refer to Eq. 45 in Fajgelbaum and Khandelwal (2016) and to its lengthy derivation.Footnote 20

We update the estimates of Fajgelbaum and Khandelwal (2016): relative to their original dataset, the only major difference is that, while they used a “reference WIOD table” obtained as the average of the November-2013 vintage of WIOD tables for the years 2005 to 2007, we use the new (November 2016) vintage of WIOD table for year 2014, in continuity with the previous sections.Footnote 21 WIOD data allow exploiting the origin-product detail; with respect to the previous analysis based on HBS data, this additional flexibility comes at the cost of losing detail on consumption items, from four-digit COICOP to two-digit CPA.

All we need for our purposes are estimates for the income (semi)elasticities βj appearing in Eq. (2); since index j now should be thought of as spanning all product-origin pairs (g,i), we switch to notation βgi for clarity. These semi-elasticites are indirectly obtained as non-linear functions of parameters estimated from Fajgelbaum and Khandelwal (2016) gravity equation; these estimates are flexible enough to allow for, say, a negative income elasticity of the demand for textiles imported from China and a positive one for those imported from France.

They enable us to compute item-specific expenditure shares for households at any point of the income distribution. Indeed, Eqs. (1) and (2) can be rearranged, after substituting the item-j index with the product-origin pair index (g,i), as

$${s}_{gi}^{h}={S}_{gi}+{\beta }_{gi}ln\left[\frac{{y}_{h}}{{\varvec{y}}\boldsymbol{ }exp(T)}\right],$$

By plugging-in estimated values (denoted by a hat), the estimated expenditure share of household h on any product-origin pair (g,i) can be easily obtained as a simple function of the aggregate share Sgi observed from WIOD and of agent-h’s “relative income” (the term in square brackets):

$${\widehat{s}}_{gi}^{h}={S}_{gi}+{\widehat{\beta }}_{gi }ln\left[\frac{{y}_{h}}{{\varvec{y}}\boldsymbol{ }exp(T)}\right],$$
(3)

where \({y}_{h}\), the average household income y and the Theil index T of the associated distribution are all computed on data for 2016 from the Bank of Italy’s Survey of Household Income and Wealth (SHIW).

For each household h the fraction xh of foreign value added in his overall expenditure is then computed as:

$${x}^{h}=\sum_{g,i} {\widehat{s}}_{gi}^{h} {f}_{gi},$$
(4)

where \({f}_{gi}\) is the share of foreign value added in expenditure in product g sourced from country i, computed from the WIOD table for year 2014 (November 2016 vintage). Combining Eqs. (3) and (4) yields a relationship between the fraction of foreign value added in overall expenditure on (g,i) combinations by household h and its “relative income”.

In Fig. 5 we report the share of foreign value added—split between the component originating in the euro area (except Italy) and that originating outside—for two distinct expenditure categories, based on whether they display a negative (necessities) or positive (luxuries) income elasticity (βgi).Footnote 22 These shares are displayed for fictitious households h corresponding to ventiles of the distribution of equivalent consumption: for each ventile of the distribution of equivalent consumption in SHIW data for 2016, we compute the ratio between the equivalent income of households in the specific ventile and the overall mean equivalent income and use this ratio as the “relative income” of households in the hth ventile in Eq. (3).Footnote 23

Fig. 5
figure 5

Foreign value added by origin and income elasticity along the distribution of equivalent consumption (percentage share of foreign value added in consumption expenditure)

Figure 5 confirms that the overall share of foreign value added in consumption expenditure (the sum of the four bars) does not vary much across households once the extreme ventiles are disregarded; the distribution is slightly U-shaped, whereas it was slightly monotonic increasing in Fig. 3, left panel. There is, however, significant heterogeneity along the distribution both in terms of consumption bundles (necessities vs luxuries) and in terms of origin of foreign value added. As for the heterogeneity in consumption bundles, while expenditure by households in the last ventile of the distribution consists of luxuries for 98%, the corresponding share is only around 78% for households in the first ventile (the median is around 88%). Hence, less affluent households are relatively more exposed to global production chains through their expenditure on necessities while more affluent ones through their expenditure on luxuries. In the aggregate, necessities are more foreign-value-added-intensive than luxuries: the foreign-value-added content for necessities is around 55% (of which around one third from the euro area), just around 18% for luxuries (of which around two fifths from the euro area). The compounding of all these composition effects along the distribution implies that around 78% of the foreign value added consumed by less affluent households through their expenditure on necessities originates outside the euro area while that consumed by more affluent ones through luxuries is around 62%.

Conclusions

This paper moves from the consideration that items consumed by households are the result of global production chains. Because of the heterogeneity in consumption bundles across households and in supply chains across consumption items, different households are likely to consume different fractions of foreign value added, “originating” from different (foreign) countries. In other words, for every euro spent in consumption, both the share that flows abroad to remunerate foreign production factors and how such share is split among the various (foreign) economies varies from one household to another.

The implications are twofold. On the one hand, the extent to which individual expenditure contributes to domestic and to foreign aggregate income varies across households. On the other hand, households display a differential exposure to foreign shocks, depending on where they originate: in this paper we focus on Italian households and on the distinction between shocks endemic to the euro-area and those that hit the Eurozone from outside, like exchange-rate movements or foreign price developments that end up affecting the price of consumption goods and services (i.e. Italian households’ purchasing power).

We explore these issues by exploiting a broad range of sources and methods. We heavily rely on WIOD tables, which allow to compute the overall foreign value-added content of each euro spent on a given consumption item and to allocate it among the countries that generated it.

Our calculations show that, at the aggregate level, between one and two fifths of Italian household expenditure “buy” foreign value added. All our results suggest that this share is broadly constant across households with different levels of consumption.

These findings bear important implications. For instance, they quantify in 60 to 80 cents the direct impact on domestic GDP of each euro spent in Italy on private consumption-support programmes; the direct impact for income-support programmes would be even lower, once the aggregate propensity to consume out of income is factored in. Similarly, our findings unveil that only 20 to 40% of aggregate household consumption is exposed to the inflationary pressures that originate abroad.

Heterogeneity across consumers does instead matter in terms of the geographical composition of the foreign value added embedded in consumption: the share of expenditure flowing to euro-area production factors (i.e. “buying” foreign value added generated by euro-area countries other than Italy) is lower for less affluent households. We also unveil a significant heterogeneity across the distribution in terms of consumption bundles, namely in terms of which expenditure items account for the largest portion of foreign value added: the expenditure of less affluent households embeds a larger share of foreign value added through more price-inelastic varieties and through necessities.

The implications of this heterogeneity are especially relevant when assessing the potential effects of exchange-rate movements, which directly affect only the portion of foreign value added originating outside the euro area. Indeed our results imply that households at the lower end of the distribution have less margins of adjustment in reaction to, for example, an exchange-rate shock that raises the price of the foreign component of their consumption.

Our analysis shows that euro-area membership contributes both to shield Italian consumers from exchange-rate shocks and—we believe—to reduce their size. Indeed, we find that around 40% of the foreign-value-added content of consumption originates in euro-area countries (other than Italy) and therefore is not subject to exchange-rate fluctuations. As for the size of shocks, the euro is a global currency, which plausibly warrants that it is much more stable (and stronger) than a hypothetical national currency had Italy not joined the monetary union. Importantly, the benefits of these two mechanisms are different across the household distribution. We find that, compared to less affluent households, more affluent ones buy more euro-area value added for each euro they spend on consumption; this implies that they benefit relatively more from the direct effect of the common currency. Less affluent households, on the contrary, buy more extra-euro area value added for each euro spent on consumption; hence, the level of their purchasing power benefits relatively more from the global status of the euro and the strength and stability that go with it.

In measuring foreign value added embedded in household consumption, we have focused on the distinction between value added originating in the euro-area vs that accruing elsewhere. There are of course other ways to split up overall foreign value added, also depending on the research question of interest, and each of them is likely to highlight different heterogeneities among households. One example we have discussed, but not pursued, is the case of “tariff wars” or, more generally, trade-policy shocks: in this case the relevant distinction would be between the EU vs extra-EU value added.