Skip to main content

Internationalization choices: an ordered probit analysis at industry level


This paper extends a highly influential contribution by Helpman et al. (Am Econ Rev 94: 300–316, 2004), on the link between within-industry heterogeneity, exports and foreign direct investment (FDI), by examining the robustness of their results using a more comprehensive dataset and relaxing the assumptions regarding the distribution of firm sales. The results of an ordered probit model—estimated on exports and mergers and acquisitions data for a sample of 25 origin countries, 91 destination countries and 57 manufacturing industries, over the period 1994–2004—confirm that sectors characterized by larger firms and more dispersed firm sales show higher probability of internationalization, through both exports and FDI.

This is a preview of subscription content, access via your institution.


  1. 1.

    For a recent survey of this literature, see Greenaway and Kneller (2007).

  2. 2.

    Another strand of the literature focuses on the distinction between horizontal and vertical FDI (see, e.g. Carr et al. 2001; Conconi et al. 2013). However, this issue is out of the scope of our paper.

  3. 3.

    The Melitz (2003) and HMY theoretical framework have also been adopted to analyse the internationalization choices of tradable services industries in India (Bhattacharya et al. 2012) and in Germany (Vogel and Wagner 2009, 2011), as well as to compare self-selection in the export markets in France, Germany and the UK (Temouri et al. 2013).

  4. 4.

    Data available do not allow us to replicate HMYs’ methodology, since while we have data on exports, we can only proxy for FDI using data on the value of M&A. These are not directly comparable to foreign sales.

  5. 5.

    There is a large literature comparing the performances of domestic, export-oriented or multinational firms. However, since it is not explicitly aimed at explaining internationalization choices, we will not mention it here.

  6. 6.

    Additional empirical evidence, surveyed by Bernard et al. (2007) and Greenaway and Kneller (2007), confirms the theoretical hypothesis that firms self-select into internationalization strategies depending on their productivity level.

  7. 7.

    A partly contrasting result is that of Todo (2011) who, allowing firm heterogeneity in unobserved characteristics by estimating a multinomial logit model with random intercepts and random coefficients (a mixed logit model), finds a small economic impact of productivity on the probability that a firm exports or invests abroad for Japanese firms.

  8. 8.

    The ordered probit model is described in the “Appendix”.

  9. 9.

    The (very) few cases of industries that have M&A, but no trade are dropped from the sample.

  10. 10.

    Variables used in the analysis and their sources are listed in Online Resource 1.

  11. 11.

    In the original UN Comtrade database, countries do not necessarily report their trade statistics for each year. This means that aggregations of data into groups of countries may involve countries with no reported data for a specific year. In addition, UN Comtrade does not contain estimates for missing data. In order to rule out a possible sample selection bias, we construct a matrix of all possible pairs of countries/industries, for each year between 1994 and 2004, and we replace not available data with zero flows.

  12. 12.

    The main sources of information of data on M&A are financial newspapers and specialized agencies like Bloomberg and Reuters. It should be kept in mind that until the mid-1980s, Thomson focused very much on M&A for the USA only, and it is only for about the last 20 years that (systematic) M&A data gathering took place for other countries (Brakman et al. 2005).

  13. 13.

    Similar to trade data, missing M&A relationships for a country–industry pair are replaced by zero values.

  14. 14.

    Domestic M&A, namely acquisitions with acquirer and target located in the same country, could still provide access to foreign markets if the target firm is active abroad or if the acquirer is controlled by a foreign firm. However, in the former case, we do not know what are the foreign markets (possibly) involved, while in the latter case, we have no information about foreign controls: as a consequence, we exclude domestic M&A from our sample.

  15. 15.

    To allow the comparison of our distributions with those provided in HMY, we provide the plots and regression coefficients for Manufacture of rubber products and Manufacture of motor vehicles industries in the USA (Online Resource 2).

  16. 16.

    While countries might use different capital–labour ratios, in absence of detailed information, we follow Bernanke and Gurkaynak (2002) in calculating TFP growth rates under the assumption that labour share is the same and fixed for both developed and developing countries. Lack of data prevents us from using more refined approaches for the estimation of TFP (e.g. Levinsohn and Petrin 2003). Nevertheless, since our analysis covers several industries and capital and labour share might differ among them, in the robustness checks, we use different values of \(\alpha \) to calculate TFP.

  17. 17.

    Since the original data on patents is classified according to the US Patent Classification, we combined it with other information adopting the correspondence scheme between the US Patent Classification and the International Patent Classification and between the latter and the ISIC3 provided by Johnson (2002).

  18. 18.

    The CEPII follows the great circle formula and uses latitudes and longitudes of the most important cities (in terms of population) to calculate the average of distances between pairs of cities. Data on distances are available at: We also adopt distances between capitals as an alternative measure and the results remain unchanged.

  19. 19.

    Concordance tables are available at: Online Resource 3 lists concordances between SIC and ISIC classifications at the 3-digits.

  20. 20.

    Descriptive statistics are computed on the largest sample, the one including Germany in the group of domestic countries.

    Table 1 Summary statistics (whole sample)

    Among the bilateral characteristics, tariffs show a high variability, with values ranging between 0 and 58 % and an average level of 12 %. The average number of common partners in trade is 58, with values ranging between 0 and 117, while the average number of common partners in M&A is much lower and the range narrower (between 0 and 30). This difference highlights that the two “networks” are quite different, and the former is much larger than the latter.

    Online Resource 6 reports the summary statistics (means and standard deviations) for all variables in our data set, distinguishing among “domestic”, “exports” and “exports and M&A” cases. The first category, grouping 5890 observations, includes cases not involved in an international relationship at all; the second, by far the most numerous (62,706 observations), includes cases involved in exports only; the third category, featuring 4229 observations, includes country pairs-industries involved in both exports and M&A. The distribution of the key explanatory variables in the three samples implies that the higher the internationalization involvement of industries, the higher the level of within-industry dispersion, in size and also in productivity, independently on the measure adopted. This suggests, as expected, that industries involved both in trade and in foreign investment show the highest presence of large and more productive firms. Industries which are only active in exports represent 86 % of our sample. Domestic industries represent 8 % of the total and exporter and foreign investor industries represent 6 %.

    Table 2 Correlation matrix
  21. 21.

    All estimations reported include three sets of dummies, controlling for: the domestic country, the foreign country and the industry-specific fixed effects. This is described in Sect. 3.

  22. 22.

    Qualitatively similar results, available on request, are obtained using different variations, for example from \(10\mathrm{th}\) to \(90\mathrm{th}\) percentile, as in Benfratello and Razzolini (2009), from \(20\mathrm{th}\) to \(80\mathrm{th}\) or from \(30\mathrm{th}\) to \(70\mathrm{th}\) percentile.

  23. 23.

    In Online Resource 7, we have also verified our main hypothesis by using the number of large firms in an industry, by adopting the world distribution of firm sales. The main results remain unchanged if we consider the number of firms in the \(9\mathrm{th}\) and \(10\mathrm{th}\) decile, or in the \(4\mathrm{th}\) and \(5\mathrm{th}\) quintile, instead of the dispersion.

  24. 24.

    These results are consistent with those of HMY as far as capital intensity is concerned, but not in the case of innovation activity. It should be noted, though, that we differ from them in terms of the variable used to proxy for innovation: the number of patents rather than R&D expenses.

  25. 25.

    Lists of advanced economies and developing and emerging countries are available from the IMF website, at:

    Table 4 Internationalization from advanced countries and developing countries to all other countries
    Table 5 Internationalization from advanced countries and developing countries to developed countries
    Table 6 Internationalization from advanced countries and developing countries to developing countries
  26. 26.

    In Online Resource 8, we also re-estimate the model by using different threshold of export and/or M&A to distinguish between domestic and internationalized industries. In particular, we take the following thresholds based on the world distribution of sales: (i) exports higher than the \(50\mathrm{th}\) percentile, (ii) M&A higher than the \(90\mathrm{th}\) percentile and (iii) exports higher than the \(50\mathrm{th}\) percentile and M&A higher than the \(90\mathrm{th}\) percentile. We find that our main results remain unchanged.


  1. Barba Navaretti G, Venables AJ (2004) Multinational firms in the world economy. Princeton University Press, Princeton

    Google Scholar 

  2. Basile R, Giunta A, Nugent J (2003) Foreign expansion by Italian manufacturing firms in the nineties: an ordered probit analysis. Rev Ind Organ 23:1–24

    Article  Google Scholar 

  3. Bhattacharya R, Patnaik I, Shah A (2012) Export versus FDI in services. World Econ 35:61–78

    Article  Google Scholar 

  4. Benfratello L, Razzolini T (2009) Firms’ productivity and internationalisation choices: evidence for a large sample of manufacturing firms. In: Piscitello L, Santangelo G (eds) Multinationals and local competitiveness. Franco Angeli, Milano

    Google Scholar 

  5. Bernanke BS, Gurkaynak RS (2002) Is growth exogenous? Taking Mankiw, Romer and Weil seriously. NBER Macroecon Annual 16:11–72

    Article  Google Scholar 

  6. Bernard AB, Jensen JB, Redding S, Schott PK (2007) Firms in international trade. J Econ Perspect 21:105–130

    Article  Google Scholar 

  7. Blonigen B (2002) Tariff-jumping antidumping duties. J Int Econ 57:31–49

    Article  Google Scholar 

  8. Blonigen B (2005) A review of the empirical literature on FDI determinants. Atl Econ J 33:383–403

    Article  Google Scholar 

  9. Brakman S, Garretsen H, Van Marrewjk C (2005) Cross-border mergers and acquisitions: on revealed comparative advantage and merger waves. CESifo working paper no. 1602. Accessed 3 Mar 2010

  10. Brainard SL (1993) A simple theory of multinational corporations and trade with a trade-off between proximity and concentration. NBER working paper no. 4269. Accessed 3 Mar 2010

  11. Brainard SL (1997) An empirical assessment of the proximity-concentration trade-off between multinational sales and trade. Am Econ Rev 87:520–544

    Google Scholar 

  12. Bougheas S, Görg H (2008) Organizational forms for global engagement of firms. University of Nottingham, GEP research paper no. 2008/33. Accessed 27 Feb 2012

  13. Calia P, Ferrante MR (2010) How do firms combine different internationalization modes? A multivariate probit approach. Paper presented at the ETSG 2010 annual conference, 9–11 September 2010, Lausanne, Switzerland. Accessed 14 Feb 2012

  14. Carr DL, Markusen JR, Maskus KE (2001) Estimating the knowledge capital model of the multinational enterprises. Am Econ Rev 91:693–708

    Article  Google Scholar 

  15. Chaney T (2011) The network structure of international trade. NBER working paper no. 16753. Accessed 14 Feb 2012

  16. Conconi P, Sapir A, Zanardi M (2013) The internationalization process of firms: from exports to FDI. CEPR discussion paper no. 9332. Accessed 27 June 2012

  17. Demirbas D, Patnaik I, Shah A (2013) Graduating to globalisation: a study of Southern multinationals. Indian Growth Dev Rev 6:242–259

    Article  Google Scholar 

  18. Disdier AC, Head K (2008) The puzzling persistence of the distance effect on bilateral trade. Rev Econ Stat 90:37–48

    Article  Google Scholar 

  19. Eeckhout J (2004) Gibrat’s law for (all) cities. Am Econ Rev 94:1429–1451

    Article  Google Scholar 

  20. Engel D, Procher V, Schmidt CM (2009) Foreign market dynamics and the symmetric role of firm-specific characteristics—evidence for French firms. Accessed 30 Mar 2013

  21. Francois J (2010) Who trades with whom. Paper presented at the 12th European Trade Study Group, Lausanne 9–11 September, 2010

  22. Greene WH (2008) Econometric analysis. Prentice-Hall, Upper Saddle River

    Google Scholar 

  23. Greenaway J, Kneller J (2007) Firm heterogeneity, exporting and foreign direct investment. Econ J 117:F134–F161

    Article  Google Scholar 

  24. Head K, Ries J (2003) Heterogeneity and the FDI versus export decisions of Japanese manufacturers. J Jpn Int Econ 17:448–467

    Article  Google Scholar 

  25. Head K, Mayer T, Thoenig M (2014) Welfare and trade without pareto. Am Econ Rev 104:310–316

    Article  Google Scholar 

  26. Helpman E, Melitz MJ, Yeaple SR (2004) Export versus FDI with heterogeneous firms. Am Econ Rev 94:300–316

    Article  Google Scholar 

  27. Helpman E, Melitz MJ, Rubistein Y (2008) Estimating trade flows: trading partners and trading volumes. Q J Econ 73:441–487

    Article  Google Scholar 

  28. Herger N, Kostoggiannis C, McCorriston S (2008) Cross-border acquisitions in the global food industry. Eur Rev Agric Econ 35:563–587

    Article  Google Scholar 

  29. Isaksson A (2009) The UNIDO world productivity database: an overview. Inter Prod Monit 18:38–50

    Google Scholar 

  30. Johnson DKN (2002) The OECD technology concordance (OTC): patents by industry of manufacture and industry of use OECD science, technology and industry working papers no. 2002/5. http://www.oecd-ilibraryorg/science-and-technology/the-oecd-technology-concordance-otc_521138670407. Accessed 14 Feb 2012

  31. Kimura F, Kiyota K (2006) Exports, FDI, and productivity: dynamic evidence from Japanese firms. Rev World Econ/Weltwirtschaftliches Archiv 142:695–719

    Article  Google Scholar 

  32. Levinsohn J, Petrin A (2003) Estimating production functions using inputs to control for unobservables. Rev Econ Stud 70:317–41

    Article  Google Scholar 

  33. Levy M (2009) Gibrat’s law for (all) cities: comment. Am Econ Rev 99:1672–1675

    Article  Google Scholar 

  34. Markusen JR, Maskus K (2002) Discriminate among alternative theories of the multinational enterprise. Rev Intl Econ 10:694–707

    Article  Google Scholar 

  35. Melitz MJ (2003) The impact of trade on intra-industry reallocation and aggregate industry productivity. Econometrica 71:1695–1725

    Article  Google Scholar 

  36. Oberhofer H, Pfaffermayr M (2012) FDI versus exports: multiple host countries and empirical evidence. World Econ 35:316–330

    Article  Google Scholar 

  37. Oldenski L (2010) Export versus FDI: a task based approach. Working paper, Georgetown University, 2010. http://www9.georgetownedu/faculty/lo36/Oldenski_XvsFDI_Nov2010pdf. Accessed 14 Feb 2010

  38. Rajan RG, Zingales L (1998) Financial dependence and growth. Am Econ Rev 88:559–586

    Google Scholar 

  39. Slangen AHL, Beugelsdijk S (2010) The impact of institutional hazards on foreign multinational activity: a contingency perspective. J Int Bus Stud 41:980–995

    Article  Google Scholar 

  40. Slangen AHL, Beugelsdijk S, Hennart JMA (2011) The impact of cultural distance on bilateral arm’s length exports: an international business perspective. Manag Int Rev 51:875–896

    Article  Google Scholar 

  41. Temouri Y, Vogel A, Wagner J (2013) Self-selection into export markets by business services firms—evidence from France, Germany and the United Kingdom. Struct Change Econ Dyn 25:146–158

    Article  Google Scholar 

  42. Todo Y (2011) Quantitative evaluation of the determinants of exports and FDI: firm-level evidence from Japan. World Econ 34:355–381

    Article  Google Scholar 

  43. Vogel A, Wagner J (2009) Exports and profitability: first evidence for German business services enterprises. University of Lüneburg working paper series in economics no. 129. Accessed 08 April 2014

  44. Vogel A, Wagner J (2011) Robust estimates of exporter productivity premia in German business services enterprises. Econ Bus Rev 13:7–26

    Google Scholar 

  45. Wang C, Wei Y, Liu X (2010) Determinants of bilateral trade flows in OECD countries: evidence from gravity panel data models. World Econ 33:894–915

    Article  Google Scholar 

  46. Wooldridge JM (2010) Econometric analysis of cross section and panel data, 2nd edn. MIT Press, Cambridge

    Google Scholar 

  47. Yeaple S (2003) The role of skill endowments in the structure of US outward foreign investments. Rev Econ Stat 85:726–734

    Article  Google Scholar 

Download references


We acknowledge financial support from the “New Issues in Agricultural, Food and Bio-energy Trade (AGFOODTRADE)” (Small and Medium-scale Focused Research Project, Grant Agreement no. 212036) research project funded by the European Commission. We would also like to acknowledge the comments of participants at the Italian Trade Study Group conference (Catania, 15–16 June 2012), at the European Trade Study Group conference (Leuven, 13–15 September 2012), at the Riunione della Società Italiana degli Economisti (Bologna, 24–26 October 2013) and at the Research Meeting of NIPFP-DEA Research Program (New Delhi, 13–14 March 2014). In particular, we wish to thank Luigi Benfratello, Roberto Patuelli, Filippo Vergara Caffarelli and Maurizio Zanardi, for helpful comments on a previous version of this paper. A special thanks to the Electronic Resources Area of Bocconi University’s library for providing the access to the Worldscope Database.

Author information



Corresponding author

Correspondence to Filomena Pietrovito.

Electronic supplementary material



The ordered probit model for \(y\) can be derived from a latent or unobserved continuous variable, \(y^{*}\), related to a set of explanatory variables according to a standard linear model:

$$\begin{aligned} y^{*}=\beta _0 +\beta _1 x_1 +\beta _2 x_2 +....+\beta _K x_K +\varepsilon \end{aligned}$$

where \(x_{1,\ldots , K}\) are the explanatory variables, which may include industry and country characteristics influencing the probability of different internationalization involvements, \(\beta _{1\ldots k}\) are the associated parameters, and \(\varepsilon \) is a random error term drawn from a standardized normal distribution. Although \(y^{*}\) is unobserved, \(y\) is observed and related to \(y^{*}\) by the following relationship:

$$\begin{aligned} {\begin{array}{lll} {y=0}&{} \quad {\hbox {if}}&{} {y^{*}\le \alpha _1 } \\ {y=1}&{} \quad {\hbox {if}}&{} {\alpha _1 <y^{*}\alpha _2 } \\ {y=2}&{} \quad {\hbox {if}}&{} {y^{*}>\alpha _2 } \\ \end{array} } \end{aligned}$$

where \(\alpha _{1}<\alpha _{2}\) are the unobserved cut points identifying the boundaries between the different levels of international involvement. Therefore, given the standard normal assumption for the error term, we can derive each response probability of observing an industry as being “domestic” (i.e. the dependent variable \(y\) taking the value of 0) as:

$$\begin{aligned} \Pr [y=0]= & {} \Pr [y^{*}\le \alpha _1 ] \nonumber \\= & {} \Pr [\beta _0 +\beta _1 x_1 +\beta _2 x_2 +....+\beta _K x_K +\varepsilon \le \alpha _1 ] \nonumber \\= & {} \Pr [\varepsilon \le \alpha _1 -(\beta _0 +\beta _1 x_1 +\beta _2 x_2 +....+\beta _K x_K )] \\= & {} \Phi (\alpha _1 -(\beta _0 +\beta _1 x_1 +\beta _2 x_2 +....+\beta _K x_K )) \nonumber \\= & {} \Phi (\alpha _1 -\mathbf{x \upbeta } )\nonumber \end{aligned}$$

where \(\Phi (.)\) is the standard normal distribution function. Similarly, we can obtain the probability of \(y\) = 1 and \(y\) = 2 in the following way:

$$\begin{aligned} \begin{aligned} \Pr [y=1]&=\Pr [\alpha _1 <y^{*}\le \alpha _2 ]=\Phi (\alpha _2 -\mathbf{x\upbeta })-\Phi (\alpha _1 -\mathbf{x\upbeta })\\ \Pr [y=2]&=\Pr [y^{*}>\alpha _2 ]=1-\Phi (\alpha _2 -\mathbf{x\upbeta }) \end{aligned} \end{aligned}$$

The \(\beta \) parameters and the threshold levels on the latent variable which characterize the transition from one observed categorical response to the next (cut points \(\alpha \) ) can be obtained through the maximum likelihood estimation.

In general, in a \(J\)-choice ordered probit model, \(y\) is an ordered response where the values we assign to each outcome represent a specific order along a continuum, but not the magnitude of difference between the options. In our specification (1), \(y\) is an indicator of international involvement at industry level ranging between zero and 2, with: \(y\) = 0 for industries that are not internationalized at all (“domestic”), \(y\) = 1 for industries that internationalize only through trade (“exports”) and \(y\) = 2 for industries that have both trade and M&A (“exports and FDI”). That the coefficient 2 indicates a higher international involvement than 1 (and 0) even though the index itself has only an ordinal meaning, suggesting useful information. Whereas in a linear regression, an industry with an index equal to 2 would be twice as internationalized as one with an index equal to 1, in the ordered probit model, no such presumption of cardinality is made: a value of 2 simply indicates more internationalization than a value of 1.

For such an ordinal dependent variable, using multinomial probit or logit would not be efficient. In fact, these models would mis-specify the data-generating process, assuming that there is no order in the different categories that the dependent variable can take. An OLS regression estimation would also be inappropriate, since it would consider the difference in the dependent variable between a 0 and a 1 as equivalent to the difference between a 1 and a 2. Greene (2008) summarizes the previous remarks pointing out that when “the outcome is discrete, the multinomial logit or probit model would fail to account for the ordinal nature of the dependent variable. Ordinary regression analysis would err in the opposite direction, however” (Greene 2008, 831).

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Pietrovito, F., Franco Pozzolo, A. & Salvatici, L. Internationalization choices: an ordered probit analysis at industry level. Empir Econ 50, 561–594 (2016).

Download citation


  • Internationalization choices
  • Exports
  • Mergers and acquisitions
  • Ordered probit

JEL Classification:

  • C25
  • D21
  • F10
  • F14
  • F20
  • F23