Estimating Sourcing Premia Using Italian Regional Data

2.1 Global Sourcing under Contractual Incompleteness

For the sake of simplicity, consider a very stylized framework in which a firm is willing to produce a final good; the final-good production requires two tasks – input manufacturing and final-good processing – and the enterprise itself is responsible for processing. Under these assumptions, the firm can either manufacture components within its boundaries or purchase them from an independent supplier; moreover, inputs can be made or bought in either the home or the host country. Depending on whether the input supplier is a domestic or a foreign enterprise and whether it belongs to the firm or it does not, four sourcing strategies may emerge, namely, domestic integration (DI), domestic outsourcing (DO), foreign integration (FI), and foreign outsourcing (FO) ^[4] (Antras and Helpman 2004).

From a theoretical point of view, one candidate explanation of global sourcing stems from the wish to mitigate hold-up concerns in a context of contractual incompleteness.

The original intuitions of Grout (1984), Grossman and Hart (1986), and Hart and Moore (1990) are extended to the international context in McLaren (2000), Grossman and Helpman (2002, 2003, 2005), Antras (2003), Ottaviano and Turrini (2007), and Antras and Helpman (2004, 2008).

The framework maintained across these models is that a firm located in the north is the final-good producer and that final-good production requires relationship-specific investments to manufacture components. The difference is, instead, in the number of sourcing strategies that are analyzed. In particular, McLaren (2000), and Grossman and Helpman (2002) focus on DI as opposed to DO; Grossman and Helpman (2005) on DO rather than FO; and Grossman and Helpman (2003), Antras (2003), and Ottaviano and Turrini (2007) on FI rather than FO. Antras and Helpman (2004, 2008) ^[5] are the only contributions that offer the full characterization of all sourcing modes in a unitary framework.

Despite the subset of strategies that are considered in the different papers, a common message emerges from this theoretical literature. In choosing between in- and outsourcing, firms trade off the benefits of ownership under the former with the benefits of better incentives for the manufacturer under the latter. In choosing between domestic and foreign sourcing, final-good producers trade off the benefits of lower variable costs in the south with the benefits of lower fixed organizational costs in the north. By combining these effects, one obtains a clear understanding of the main factors that affect firm sourcing behavior.

An additional interesting question regards which firms are more likely to choose a specific sourcing strategy rather than the others. Clearly, this issue can be addressed only in a heterogeneous firm framework, as in Antras and Helpman (2004, 2008). Building on Melitz (2003), those authors assume that upon paying fixed entry costs, final-good producers draw a productivity level from a known distribution; after observing that level, each of them decides whether to exit the market or to start producing. In the end, the two papers demonstrate that in low-tech sectors, integration never occurs: Lower-productivity players choose DO, and higher-productivity players select FO. On the contrary, in high-tech sectors, all sourcing strategies may be undertaken: Lower-productivity firms buy inputs in the domestic market, and higher productivity firms buy inputs in the foreign market; among firms that source in the same country, the more productive ones vertically integrate, and the less productive outsource. These results provide unambiguous predictions about firm sorting into different sourcing modes. Put another way, firms are expected to self-select into one of the four mutually exclusive alternatives – DO, DI, FO, and FI – depending on their productivity levels.

In the last few years, a burgeoning empirical literature has grown rapidly that brings the intuition of Antras and Helpman (2004, 2008) to the data. The availability of large longitudinal datasets at the micro-level has recently helped obtain interesting findings about productivity and sourcing behavior and have offered a preliminary test of certain theoretical results and adding to previous industry-level evidence (Antras 2003; Nunn and Trefler 2008; Yeaple 2006; Federico 2012; Bernard et al. 2010).

Empirical papers of this sort differ in terms of their sourcing modes and geographical coverage. For the former issue, the majority of studies consider only a subset of strategies depending on the data availability. For instance, Tomiura (2005, 2009) and Ito, Tomiura, and Wakasugi (2011) focus on FO as opposed to DO, and Tomiura (2007a), Defever and Toubal (2013), and Corcos et al. (2013) study FO instead of FI, whereas Tomiura (2007b), Federico (2010), and Kohler and Smolka (2011, 2012) are the only papers to consider all strategies in the same econometric model, capturing the original spirit of Antras and Helpman (2004, 2008). For the geographical coverage, the existing studies provide evidence only for Japan (Tomiura 2005, 2007a, 2007b, 2009; Ito, Tomiura and Wakasugi 2011), France (Defever and Toubal 2013; Corcos et al. 2013), Italy (Federico 2010), and Spain (Kohler and Smolka 2011, 2012).

Despite these differences, important regularities also emerge throughout the empirical literature. First, heterogeneity is measured quite consistently in terms of labor and total factor productivity; second, the results are quite similar: Firms that source abroad are, on average, better than firms that source domestically; moreover, vertically integrated enterprises are, on average, better than outsourcers. By and large, these findings are consistent with the theoretical framework of Antras and Helpman (2004, 2008) in that the most productive firms choose FI and the least productive opt for DO. However, no definite ranking emerges that encompasses all sourcing strategies because firms with intermediate levels of productivity self-select into either DI or FO (see, for instance, Federico 2010; Kohler and Smolka 2011). Hence, the theoretical sorting of firms into sourcing modes is preserved only at the extremes.

How does our contribution fit into the literature on global sourcing under contractual incompleteness? Estimating sourcing premia, in the global sourcing jargon, means analyzing the matching between heterogeneous enterprises and organizational modes. Our contribution then lays in the measure of heterogeneity, which is not restricted to productivity, and the characterization of organizational modes, which includes all sourcing strategies dissected by production tasks. Addressing the whole set of organizational modes, heterogeneity dimensions and definitions of sourcing strategies, we develop a wide research framework that encompasses previous results as a special case.

2.2 Internationalization and Performance

Starting from the seminal contribution of Bernard and Jensen (1995), a large number of scholars have investigated the relationship between internationalization and performance at the micro-level.

They document that globally engaged enterprises tend to be a minority compared with purely domestic players, but they perform better on a number of economic variables, including sales, wage, value added, and productivity.

Those authors also suggest two alternative but not mutually exclusive explanations for why firms that are involved in international business are better than domestic enterprises. Under the so-called self-selection argument, there are ex-ante performance differences between firms that will become international and firms that will keep serving the domestic market. The intuition is that operating abroad involves additional costs related to transportation, marketing, human capital and production that provide a natural entry barrier to less successful firms.

Under the so-called learning-by-internationalization argument, ex post performance differences emerge, instead, as a result of firm exposure to international markets. This is because in interacting with foreign competitors and customers, firms derive information about reducing costs and quality rising processes; moreover, to keep pace with their rivals, they are strongly encouraged to innovate, increase their scales, and become more efficient.

Since the publication of the research of Bernard and Jensen (1995), a rapidly growing body of empirical literature has comprehensively investigated these issues, identifying typical internationalization premia, defined as the average percentage differences in performance between international and domestic players.

In the following, we focus more specifically on the papers that address the internationalization and performance of Italian enterprises to allow for comparing the results in Sections 4 and 5, in which we present original evidence from the same country. ^[6] For expositional convenience, we review the related literature according to the following issues. First, we portray the global involvement of Italian firms considering the relative prevalence of different internationalization strategies; second, we discuss the main internationalization premia as highlighted in the existing studies.

With regard to the first issue, evidence provides unambiguous results. Italian firms are active in a number of internationalization strategies, which range from importing to exporting, from FDI to international outsourcing, and from international partnerships to subcontracting. If one considers these strategies together, global firms prove to be the majority compared with purely domestic players (Basile, Giunta, and Nugent 2003; Benfratello and Razzolini 2009; Castellani and Zanfei 2007). If one instead considers one strategy at a time, different patterns emerge. In particular, exporters and importers tend to be quite numerous, representing more than 50% of the entire population (see, among others, Crinò and Epifani 2012; ISGEP 2008; Razzolini and Vannoni 2011; Serti and Tomasi 2008b, 2012; Serti, Tomasi, and Zanfei 2010). On the contrary, firms that engage in FDI represent a minority, accounting for less than 10% of the overall sample (Barba Navaretti et al. 2011; Casaburi, Gattai, and Minerva 2007; Castellani and Giovannetti 2010; Giovannetti, Ricciuti, and Vilucchi 2009), and the same is the case for international outsourcing (Barba Navaretti et al. 2011; Casaburi, Gattai, and Minerva 2007).

With regard to the second issue, the majority of the literature emphasizes the existence of some internationalization premia related to the foreign exposure of Italian enterprises. ^[7] Notably, each category of foreign involvement entails certain performance advantages compared with the purely domestic operations (see, among others, Basile 2001; Bratti and Felice 2012; Castellani 2007; Mayer and Ottaviano 2007). Furthermore, a systematic performance ranking emerges for players characterized by different degrees of internationalization or that pursue more than a single strategy, with FDI makers at the top and exporters at the bottom (see, for instance, Benfratello and Razzolini 2009; Casaburi, Gattai, and Minerva 2007; Castellani and Zanfei 2007; Serti and Tomasi 2008b, 2012; Castellani, Serti, and Tomasi 2010; Mayer and Ottaviano 2007; Castellani and Giovannetti 2010). Finally, there is evidence in favor of both self-selection and learning-by-internationalization (see, for instance, Basile 2001; Bratti and Felice 2012; Castellani and Giovannetti 2010; Ferragina and Quintieri 2001; ISGEP 2008; Serti and Tomasi 2008a, 2012).

How does our contribution fit into the literature on internationalization and performance? Estimating sourcing premia in the internationalization and performance jargon can be rephrased as searching for systematic performance differences between firms that choose foreign sourcing, our “international” players, and firms that choose domestic sourcing, our “domestic” players. Because domestic and foreign sourcing can be further dissected by firms’ ownership decisions, we end up with a wider framework in which different types of domestic and international enterprises can be considered at the same time. Therefore, we expect to obtain more general results to be compared with the FDI and international outsourcing premia that were detected before.

4.1 Definition of Sourcing Strategies

Defining sourcing strategies correctly is crucial to the present study. The main challenge of our survey is to obtain all of the relevant information to translate the theoretical definition into a convenient empirical proxy.

According to the theory, firms are uniquely mapped into four mutually exclusive sourcing strategies depending on their ownership and location responses to input procurement concerns. As mentioned in Section 2, this is the result of two simplifying assumptions: First, the production process involves only two tasks; and second, ownership and location decisions apply only to one task.

In designing the questionnaire, we relaxed these assumptions in a way that provided a richer taxonomy of sourcing strategies while preserving the stylized theoretical framework as a special case. More precisely, we disaggregate the production process into four tasks, rather than two: input manufacturing, final-good processing, research & development (R&D), and other services (e.g., informative, logistics). Moreover, we allow firms to engage in the ownership and location decisions for every task performed in their production process, not solely for input manufacturing. In this way, we are able to collect very detailed information on which tasks are performed within the boundaries of the firm, a foreign affiliate, an independent Italian firm, and an independent foreign firm. Unlike the stylized theoretical framework delineated in Section 2, these are not intended as mutually exclusive categories, which means that the same task may be performed both intra- and extra-firm and both domestically and abroad.

Drawing on our survey data, we give four definitions of DO, DI, FO, and FI.

First, we distinguish between the so-called conservative and liberal definitions of sourcing strategies (Table 2). Consistent with the theory, the former focuses only on input manufacturing, whereas the latter considers all production tasks except for final-good processing. ^[8] The main advantage of the conservative definition is that the results are likely to be neat and not difficult to confront with the theoretical priors; however, the definition exploits only part of the information that was collected from the survey interviews, thus offering a somewhat standard treatment of sourcing strategies. The liberal definition creates, instead, a more general overview of firms’ organizational solutions. Taking advantage of the rich dataset might disclose new facts about global sourcing; however, evaluating its results according to the theory is not clear.

Second, we distinguish between the so-called mutually inclusive and mutually exclusive definitions of sourcing strategies (Table 3). In the former case, we assign to each firm any sourcing strategy that results from the interviews, whereas in the latter case, we assign each firm only one sourcing strategy. Following Kohler and Smolka (2011), when a firm is active in two or more sourcing modes simultaneously, it is assigned the least prevalent category in the overall sample, according to the ranking of the most frequent choice, that is, DI-DO-FO-FI. ^[9] The main advantage of the mutually inclusive definition is that it preserves all information from the data; however, the results might be difficult to interpret in light of the theoretical models. The mutually exclusive definition is, in contrast, more arbitrary in that we make a choice, out of a set, regarding the strategy that should be assigned to each firm; however, the results are likely to be clear and not difficult to confront with the theoretical predictions.

In light of the above discussion, because each definition entails certain pros compared with the others, we prefer retaining all of them for empirical purposes. This eventually enables us to make comparisons and reach a better understanding of global sourcing.

4.2 Descriptive Statistics

Table 4 disaggregates the sampled firms into all possible combinations of DO, DI, FO, and FI – in italics – to provide a first observation of the data.

According to the conservative (liberal) definition, no-sourcing firms (NS) ^[10] represent only 25% (11%) of the overall sample, whereas sourcing firms are the majority.

If we observe the pure strategies, ^[11]DI is the most frequent choice, followed by DO, FO, and FI. Notice also that FO and FI are quite marginal in our data, accounting for a negligible number of observations. This result is strongly consistent with the Japanese firms portrayed by Tomiura (2007b), for which the same ranking emerges; on the contrary, Kohler and Smolka (2011, 2012) report that Spanish enterprises are primarily active in DO, followed by FO, DI, and FI, and the same is true for the sample of Italian firms that were analyzed in Federico (2010). If we instead observe the mixed strategies, the most frequent combination is DIDO. Interestingly, DI occurs quite often both alone and together with DO, FO, and FI, whereas the other strategies tend to be chosen only as part of a mixed design.

For the ownership and location decisions that underlie firm sourcing behavior, Table 4 reveals that in choosing between home and foreign, Lombard respondents tend to prefer home, and in choosing between make and buy, they tend to select make, consistent with the theoretical case of Antras and Helpman (2004, 2008).

Interestingly, our sampled firms’ sourcing strategies appear to be complementary rather than substitutes for each other. This becomes evident after observing the average number of modes per firm, which is 1.62 below the conservative definition and 1.8 below the liberal definition. These results are aligned with the previous evidence reported by Kohler and Smolka (2011, 2012) and indicate the benefit of maintaining both mutually inclusive and mutually exclusive proxies for empirical purposes.

For the sake of completeness, Figures 1 and 2 extract mutually inclusive and mutually exclusive sourcing strategies from the data reported in Table 4; this may help the reader focus on a few facts.

Figure 1:

Sampled firms’ mutually inclusive sourcing strategies, conservative versus liberal definitions.

Figure 2:

Sampled firms’ mutually exclusive sourcing strategies, conservative versus liberal definitions.

First, mutually inclusive and mutually exclusive definitions produce only a partially overlapping ranking of the most preferred sourcing strategies, that is, DI-DO-FO-FI in the former case and DO-DI-FO-FI in the latter case. Second, these rankings are robust to the conservative versus liberal perspective.

Taking advantage of our data, this evidence can be further dissected by province, industry, and type of firm. ^[12] In particular, we ask the following: How do sourcing behaviors differ for firms that are headquartered in the regional capital? Are “true-headquarters firms” more or less likely to engage in a certain sourcing mode? Does any sectoral pattern reflect the relative prevalence of DI, DO, FO, and FI?

Figure 3 answers these questions and displays mutually inclusive sourcing strategies.

Figure 3:

Sampled firms’ mutually inclusive sourcing strategies, by province, type, and industry, conservative versus liberal definitions.

Concerning the province, we distinguish between Milan – the regional capital – and the rest of Lombardy. Our descriptive statistics reveal that firms headquartered in Milan are the most likely to retain strategic high-value functions, such as R&D and other services, within the boundaries of a foreign affiliate. On the contrary, there seems to be no “metropolitan hub-effect”, given that the percentages of DI are very similar throughout the sample and do not display any sizeable differences based on the conservative versus liberal definitions.

For type of firm, we introduce the notion of “true-headquarters” to denote those companies that have less than 50% external participation in their joint capital (Kohler and Smolka 2011, 2012). Identifying true-headquarters firms is important for controlling for firms’ discretion over their input procurement choice. ^[13] Interestingly, companies that have more than 50% of external participation rely more on internationalization than do true-headquarters firms. Moreover, the former tend to perform strategic high-value functions abroad more often than the latter.

Finally, Figure 3 dissects the Lombard firms’ sourcing behavior according to the Bell and Pavitt (1993) industries. All sectors produce a ranking of the most preferred sourcing strategies that is consistent with the sample average displayed in Figure 1. However, some specific patterns emerge as well. To mention just a few, specialized supplier-dominated firms are those that rely the most on domestic sourcing; scale-intensive firms have the smallest percentage of foreign integration; and, not surprisingly, science-based firms seem quite reluctant to engage in outsourcing, especially that of R&D and other services.

For the sake of completeness, Figure 4 displays mutually exclusive sourcing strategies by province, industry, and type of firms. The results are broadly consistent with those described above for the mutually inclusive case.

Figure 4:

Sampled firms’ mutually exclusive sourcing strategies, by province, type, and industry, conservative versus liberal definitions.

5.1 Variables and Specification

Following the methodology of Bernard and Jensen (1995, 1999), we define sourcing premia as the average percentage difference of performance between firms engaged in a particular strategy compared with the others. To compute these premia, we regress firm performance on sourcing dummies, controlling for a number of additional variables that may also affect performance.

Our econometric model is set accordingly. For every performance measure, three equations are estimated, designated [1], [2], and [3].

Equation [1] has a very parsimonious specification because performance (performance_i) is regressed only on sourcing strategies (sourcing strategy_i) for the unrestricted sample of respondents, called Ω_unrestr:

This basic equation is then completed by firm controls (firmctrl_i), together with industry (industryctrl_i) and province (provincectrl_i) controls, as in eq. [2].

Equation [3] is the same as eq. [2], with the only difference being that it is estimated in the restricted sample of true-headquarters firms, called Ω_restr, to account for the potential effect of foreign ownership on performance. ^[14]

A number of variables are considered in eqs [1], [2], and [3] as a proxy for firm performance_i. They range from sales to value added and from profit to productivity (TFP) and skill intensity to investigate many potential premia related to sourcing strategies. With regard to productivity, we follow the estimation method of Levinsohn and Petrin (2003) to address the simultaneity and selection bias. ^[15] In particular, we assume the production function of firm i, at a given point in time to be Cobb-Douglas:

where Y_i is the logarithm of firm output, measured by value added; L_i and M_i are the logarithms of the freely variable inputs labor and intermediate input, approximated by the number of employees and the cost of raw materials, respectively; and K_i denotes the logarithm of the state variable capital, proxied by total assets. All variables in eq. [4] are deflated by the Producer Price Index for the appropriate two-digit NACE industry.

The independent variables can be distinguished into two main groups: the “core” and “control” regressors. Core regressors denote the main variables of interest, namely, sourcing strategy_i. A number of dummies are available (DO_con_incl, DI_con_incl, FO_con_incl, FI_con_incl, DO_con_excl, DI_con_excl, FO_con_excl, FI_con_excl, DO_lib_incl, DI_lib_incl, FO_lib_incl, FI_lib_incl, DO_lib_excl, DI_lib_excl, FO_lib_excl, and FI_lib_excl), depending on the conservative versus liberal and mutually inclusive versus mutually exclusive definitions of sourcing strategies. In addition to the core regressors, eqs [2] and [3] consider a number of “controls” to verify the robustness of the sourcing dummies to the inclusion of firm, industry, and province variables. In particular, firm-level controls include firm size, age, group affiliation, and export experience; industry effects are measured through sector dummies following Bell and Pavitt (1993), and a dummy for Milan, the regional capital, captures the province effect.

OLS estimations are retained throughout the empirical analysis, with the only exception being the skill intensity equation, which follows a probit model. Under the OLS estimation, sourcing premia can be directly discerned from the α coefficient, which captures the percentage change in performance due to a given sourcing solution ^[16]; under the probit estimation, the same is true for marginal effects.

At this stage, it is worth mentioning that survey estimation methods are employed throughout the empirical analysis to control for the potential bias originating from the response rate. Each combination of a single geographical location (out of four) and a single manufacturing activity (out of four) denotes a stratum (16 in total). In the econometric analysis, we use sampling information to obtain consistent and efficient estimates and to draw conclusions about Lombardy as a whole. Specifically, we weight each observation by the inverse of the probability of being sampled using, for every stratum, location- and industry-specific information on the total number of firms in the population and in the sample.

To avoid simultaneity, all dependent variables refer to 2009, whereas the regressors date back to 2007. Unfortunately, the cross-sectional design of our data does not allow us to implement rigorous econometric methods, apart from lagged variables, to account for endogeneity. ^[17] Therefore, one should not interpret regressions as indicating the exact direction of causality but as a convenient way of summarizing the statistical regularities among variables.

A detailed variables description is available in Table 9, in the Appendix, while Tables 10 and 11 provide some summary statistics and correlations between firm-level regressors.

5.2 Results

In this section, we comment on the main estimation results.

Tables 5 and 6 report the empirical evidence about sourcing premia when the conservative definition is applied to mutually inclusive (Table 5) and mutually exclusive (Table 6) sourcing strategies. According to eqs [1], [2], and [3], for every dependent variable, three specifications are displayed.

A first look at Tables 5 and 6 suggests some interesting facts.

With regard to core variables, FI_con_incl is found to be statistically significant, with a positive sign, in every specification. This means that FDI makers enjoy superior performance in terms of sales, value added, profit, capital, TFP, and skill intensity. FO_con_incl is found to be positive and statistically significant as well, but its explanatory power is limited to the sales, value added, and profit equations. These results are robust to firm, industry, and province controls, and they hold for both the unrestricted and restricted samples of true-headquarters firms. Our evidence thus unveils that foreign sourcing is key to Lombardy’s performance, regardless of whether it is pursued intra- or extra-firm. On the contrary, we find no effect for domestic sourcing because DO_con_incl and DI_con_incl are not significant. Consistent with Kohler and Smolka (2011, 2012) and Federico (2010), the magnitude of the FI effect is the largest, which means not only that FI discloses precious premia for Lombard enterprises but also that these sourcing premia are the highest possible.

Concerning control variables, size, age, and group affiliation seem to play a role in shaping firm performance. Indeed, they are positive drivers of many dependent variables such as sales, value added, profit, capital, and skill intensity, thus confirming the previous evidence on Italian enterprises reported by Barba Navaretti et al. (2008), Bugamelli, Cipollone, and Infante (2000), Conti, Lo Turco, and Maggioni (2010), and Bugamelli and Infante (2003). On the contrary, we find no significant effect of international experience on economic performance, which is quite surprising in light of the empirical literature reviewed in Section 2. ^[18]

Interestingly, moving from the mutually inclusive to the mutually exclusive definition does not alter the abovementioned results that can be considered typical of the conservative case.

To complete the picture, Tables 7 and 8 display our estimation results when the liberal definition is applied to mutually inclusive (Table 7) and mutually exclusive (Table 8) sourcing strategies.

Compared with the conservative case, the liberal definition makes a more restrictive argument. In terms of core regressors, sourcing premia seem to accrue only to FDI makers because FI_lib_incl is positive and statistically significant in every column of Tables 7 and 8, whereas FO_lib_incl is not significant, regardless of the parsimonious versus rich specification and the unrestricted versus restricted sample. With regard to control regressors, we find significant effects only of size on sales, value added, and capital and a significant effect of age on capital, but these effects vanish if we restrict attention to true-headquarters firms. On the contrary, group affiliation is found to be positive and statistically significant in the sales, value added, and capital equations.

These results are robust to the mutually inclusive versus mutually exclusive definitions; therefore, we retain them as typical of the liberal – as opposed to conservative – case.

As a further robustness check, one might run the same estimation exercise as in Tables 5–8 on the restricted sample of sourcing enterprises. The underlying idea is to restrict attention to firms that are actually making sourcing decisions, according to both the conservative and liberal views, and dispensing with those that do not perform the related tasks. Although we believe this is a proper way of estimating sourcing premia, it comes at the expense of a lower number of observations. Because the (unreported) results are fully consistent with those discussed above, we stick to Tables 5–8. ^[19]