1. Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana Firm Heterogeneity and FDI Productivity Spillovers: The Case of Central and Eastern European Countries Jože Damijan, Matija Rojec, Boris Majcen, Mark Knell

2. Motivation  Despite theoretical justification of potential spillovers, the evidence of technology spillovers from foreign affiliates to host country firms is weak.  Potential reasons for failing to find significant spillovers: effectiveness of MNCs in protecting their technology advantages, low absorption capacity of host country firms, a lack of differentiation between vertical and horizontal spillovers, factors related to firm heterogeneity. Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana 2

3. Objectives  The first objective is to provide a comparative study on the importance of direct technology transfer and spillovers through FDI using an exhaustive firm-level dataset on a group of comparable countries by using a common methodology and appropriate methods to account for selection and simultaneity problems.  The second objective is to account for the inherent heterogeneity of firms:  Most of existing empirical work neglects the fact that local firms are not homogeneous.  Some recent studies demonstrate that firm heterogeneity might explain a significant portion of the differential impact of FDI on firm performance.  We explicitly accounts for different aspects of firm heterogeneity, including size, absorptive capacity, productivity and the technology gap relative to foreign affiliates. Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana 3

4. Methodological approach & data  We differentiate between direct effects of FDI and horizontal and (backward) vertical spillovers from foreign affiliates to local firms.  To calculate horizontal and vertical – backward and forward – spillovers we use the methodology developed by Blalock (2001) and Damijan et al (2003a, 2003b).  The importance of different channels of technology transfer is estimated in the framework of the growth-accounting approach using a unique firm-level database that consists of a panel of some 91,500 firms in 10 transition countries.  We use several correction methods to account for possible biases in the data:  The simultaneity problem that typically arises in the growth-accounting approach in a panel data framework is dealt with by using the Olley – Pakes method.  We correct for potential selection bias that arises due to possibly endogenous foreign investment decisions using a generalized Heckman two-step procedure. Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana 4

5. Channels of technology transfer through FDI (1): Direct technology transfer from parent to affiliate  Foreign investors can transfer technology in two ways: directly to the affiliates under their ownership and control, and indirectly to other firms in the host economy through spillovers.  Ample empirical evidence on positive direct technology transfers from MNCs to their foreign affiliates in terms of higher productivity levels and growth (for example, Haddad and Harrison, 1993; Blomström and Wolff, 1994; Blomström and Sjöholm, 1999; Aitken and Harrison, 1999; Girma et al, 2001; Barry, Görg and Strobl, 2002; Alverez et al, 2002; Blalock, 2001; Damijan et al, 2003b; Arnold and Smarzynska-Javorcik, 2005; Girma and Görg, 2006).  The extent and scope of technology transfers from MNCs to their foreign affiliates heavily depends on the position of foreign affiliates in the MNCs’ international production network (see, for instance, White and Poynter 1984, Bartlet and Ghoshal 1989, Young, Birkinshaw and Hood 1998). This points to the importance of including parameters of foreign affiliates' heterogeneity in the analysis. Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana 5

6. Channels of technology transfer through FDI (2): Spillovers  FDI spillovers take place when the entry/presence of foreign affiliates, which typically have better technologies and organizational skills than domestic firms, increases the knowledge of domestic firms and foreign investors do not fully internalize the value of these benefits (Griliches 1979, 1992).  Vertical (backward, forward), inter-industry spillovers versus horizontal, intra-industry spillovers.  Kokko (1992) identifies at least four ways of FDI spillovers: the demonstration-imitation effect, the competition effect, the foreign linkage effect and the training effect.  Not all spillovers are positive, foreign firms with superior technology may force domestic firms to exit. Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana 6

7. Channels of technology transfer through FDI (3): Spillovers  Empirical literature on FDI spillovers has produced mixed results; it shows positive, neutral, and negative spillovers from foreign subsidiaries to domestic firms.  Three types of empirical FDI spillover analysis: (i) case studies, (ii) sectoral studies, (iii) lately primarily micro, firm-level data based studies.  Case studies (Rhee and Belot 1989, Larrain et al 2000, Hanson 2001, Moran 2001) : mixed results, their conclusions are not easily generalized  Sectoral studies (Caves 1974, Globerman 1979, Blomstrom 1986, Blomstrőm and Persson 1983, Blomstrőm and Wolff 1994, Kokko 1994, 1996, Blomstrom, Kokko and Zejan 1994, Sjoholm 1999, Xu 2000, Blomstroem 1996, Cantwell 1989) : mostly show positive correlation between foreign presence and sectoral productivity. Two problems:  Difficulty in establishing the direction of causality (Smarzynska 2003) => FDI may tend to locate in high productivity industries.  Do not take account of firm heterogeneity (Keller, 2004) Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana 7

8. Channels of technology transfer through FDI (4): Spillovers  Firm level panel data studies:  The main reason to move from sectoral to firm level data was a heterogeneity problem (Keller 2004)  Most firm level studies cast doubt on the existence of FDI spillovers in developing countries, the picture more optimistic for industrialized countries  For transition countries, the firm-level panel data suggest that there are few intra-industry spillovers from FDI or that they are restrictited to certain countries, categories of firms (Konings 2001, Djankov and Hoekman 2000, Kinoshita 2000, Damijan et al 2003, Nicolini and Resmini 2006, Tytell and Yudaeva 2005). Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana 8

9. Channels of technology transfer through FDI (5): Spillovers  Reasons for the lack of evidence on FDI spillovers in firm level analysis  There may really be no or even negative spillovers (Görg and Greenaway 2001, Gőrg and Strobl 2001, Sgard 2001, Aitken and Harrison 1999, Caves 1996).  Lack of absorption capacity in host countries (Kokko, 1994; Borensztein et al, 1998; and Kinoshita, 2000).  Spillovers may not occur horizontally (intra-industry) but through vertical (inter-industry) relationships (Blalock, 2001; Schoors and van der Tool, 2001; Kugler, 2006; Smarzynska-Javorcik, 2004; Damijan et al, 2003a, 2003b; Halpern and Murakozy, 2007, etc.).  Data and methodology related reasons: data quality, limited samples, linear relationship hypothesis, inappropriate econometric techniques (Gőrg and Greenaway, 2001, 2004; Keller and Yeaple, 2003; Gőrg and Strobl, 2001; Castellani and Zanfei, 2007; Knell and Rojec, 2007).  Firm heterogeneity => Positive spillovers may only affect a sub-set of firms. Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana 9

10. Channels of technology transfer through FDI (6): Spillovers  Studies that further disaggregate data into more homogenous groups of firms find more encouraging FDI spillovers results (Gőrg and Greenaway, 2004).  Firm heterogeneity includes many aspects, which act in different directions:  Geographical distance between foreign affiliates and domestic firms,  Time/dynamic dimension of FDI spillovers,  Heterogeneity of foreign affiliates,  Heterogeneity of foreign investors,  Heterogeneity the absorption capacity of domestic firms.  Introduction of firm heterogeneity in the analysis proves to be a very important development in empirical studies of FDI spillovers.  We specifically tackles the (i) distinction between vertical and horizontal spillovers, and (ii) heterogeneity of domestic firms as far as technological capacities, productivity and human capital are concerned. Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana 10

11. Channels of technology transfer through FDI (7): Spillovers  Vertical and horizontal spillovers:  Explicit differentiation between vertical and horizontal spillovers in empirical studies is relatively recent.  Horizontal intra-industry spillovers are less likely to take place than vertical spillovers (Blalock, 2001; Schoors and van der Tool, 2001; Kugler, 2006; Smarzynska- Javorcik, 2004; Damijan et al, 2003b; Gorodnichenko et al, 2007; Halpern and Murakozy, 2007).  Message 1: empirical studies on technology spillovers should differentiate between horizontal and vertical spillovers,  Message 2: analysis of vertical spillovers should differentiate between backward and forward linkages. Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana 11

12. Channels of technology transfer through FDI (8): Spillovers  Firm’s absorption capacity = productivity, technological capacity, human capital.  Level of technological capacity: Knowledge spillovers occur more frequently if the technology gap between domestic and foreign firms is not too large (Perez, 1998; Halpern and Murakozy, 2007; Ben Hamida and Gugler, 2007; Abraham et al, 2006; Girma et al, 2006).  Productivity level as a proxy for technological capacity: empirical findings linking productivity with spillovers go in both directions (Keller and Yeaple, 2003; Nicolini and Resmini, 2006; Haskel et al, 2001; Castellani and Zanfei, 2003).  Human capital capacity increases the ability of domestic firms to benefit from positive spillovers (Borensztein et al, 1998; Meyer and Sinani, 2001; Ben Hamida and Gugler, 2007; Girma et al, 2006).  Company size also seems to have a positive influence on the absorption capacity (Knell and Rojec, 2007; Veugelers and Cassiman, 1999, Ornaghi, 2004). Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana 12

13. Modeling direct and spillover effects of FDI (1) 13  We include technology variables (T = technology stock) directly in the production function on the firm level.  In this way we study various factors that affect productivity growth, including technological accumulation => by using the growth-accounting approach and decomposing TFP into factors external and external to the firm, such as R&D activity, human capital and channels of technology transfer.  (1)  (2)  (3)  G: Internal technology variables  Z: Spillover effects

14. Modeling direct and spillover effects of FDI (2) 14  Internal technology variables:  F: Foreign ownership  H: Human capital  Spillover effects:  ES: potential home market spillovers (external economies of scale at the industry level)  HS: Horizontal spillovers at the industry level  VS: Vertical backward spillovers at the industry level

15. Modeling direct and spillover effects of FDI (3) 15  Scope for horizontal spillovers: mainly reflect competitive pressures:  The higher the value added of foreign affiliates the higher the competitive pressure  The higher the share of exports in total foreign affiliates value added the lower the competitive pressure  FA: Value added of foreign affiliate  DF: Value added of domestic firms  EX: Exports of foreign affiliates

16. Modeling direct and spillover effects of FDI (4) 16  Scope for vertical spillovers - backward linkages: the impact of foreign affiliates on their upstream suppliers:  VSb kt : the sum of the output of upstream industries r purchased by firms in industry k weighted by the share of total foreign output HS kt :  Α krt : the proportion of industry r’s output consumed by industry k  IM ikt and MC ikt : imports and material costs of foreign affiliate i; the higher the share of imports in material costs the lower the backward linkages

17. Modeling direct and spillover effects of FDI (5) 17 i  External knowledge spillovers generated at the industry level:  The larger the industry, the larger the scope either for inter-firm exchange of components or for competition among differentiated firms.  Castellani and Zanfei (2007): a more accurate specification of externalities yields larger (positive and significant) spillover effects  ES: aggregate value added of the industry

18. Data  Amadeus database (Bureau van Dijk) for 1995-2005 for Bulgaria, the Czech Republic, Croatia, Estonia, Latvia, Lithuania, Poland, Romania and Ukraine. For Slovenia, local statistical office.  Manufacturing firms only, with no limitations on the size threshold.  Dataset consists of more than 90,000 firms with up to 11 annual observations. We only include firms with 5 or more annual observations (required by Olley-Pakes corrections) => the size of our data is some 315,000 annual firm observations.  Share of foreign affiliates range from 4% in Bulgaria to 10% in Poland; share in value added from 7% to 29%.  Data on input-output accounts are conducted at the NACE 2-digit level. We also apply NACE 2-digit input-output coefficients to the NACE 3-digit sectors when calculating the vertical spillovers. Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana 18

19. Econometric approach (1)  The empirical model, with both Heckman and Olley-Pakes corrections:  tfp is the logarithmic growth rate of  F ikt is a dummy for foreign ownership  w ikt denotes the stock of human capital in the firm (proxied by the average wage bill)  hs ikt and vs ikt stand for horizontal and vertical spillovers from FDI at the sectoral level  es ikt denotes the impact of sector economies of scale (proxied by the sector size)  w ikt. is interaction term of the spillover variables with the human capital variable, to control for the impact of firm absorption capacity on firm ability to reap the benefits of spillover effects  λ ikt is the inverse Mill’s ratio from the Heckman correction for sample selection.  T and R denote the year and regional dummies, and ε ikt is the remaining error term. 19 Institute of Economic Research and University of Ljubljana Milan, 27-28 February 2014

20. Econometric approach (2)  The model is estimated in OLS.  Spillovers are measured both at the NACE 2-digit (21 sectors) and NACE 3-digit (129 sectors) levels.  Firm specific effects are wiped out as we estimate the model with the dependent variable defined in first differences.  Year dummies are included to control for common external policy shocks and regional dummies for region specific shocks.  Regions are defined at the NUTS 3-digit level.  The estimations are performed and reported for each country separately. 20 Milan, 27-28 February 2014 Institute of Economic Research and University of Ljubljana

21. Econometric approach (3)  We first estimate the model on the whole sample of firms, and then proceed with separate estimations (firm heterogeneity):  for each size class (micro, small, medium and large) of firms (Bekes et al 2007 for Hungary)  for each quintile of productivity across sectors (Q1 through Q5) (Bekes et al 2007 for Hungary)  for each class of technology gap between domestic and foreign owned firms (Girma et al, 2006)  Technology gap is defined as the ratio of average productivity of domestic firms to the average productivity of foreign owned firms within each sector. Gap1, Gap2 and Gap3 refer to domestic firms with a productivity level below 80%, between 80 and 120% and more than 120% of the average productivity of foreign owned firms within each sector, respectively. 21 Institute of Economic Research and University of Ljubljana Milan, 27-28 February 2014

22. Results (1): Summary of results (Number of countries with significant direct effects/spillovers) 22 AllMicroSmallMediumLargeQ1Q2Q3Q4Q5G1G2G3 Direct effects Positive 3213120222220 Negative 0000000000000 Horizontal spillovers Positive spill. Nace-2D0001001000200 Dw6573544475562 Nace-3D1000012010001 Dw7652534446641 Negative spill. Nace-2D4441344351231 Dw1002001001101 Nace-3D5352104341521 Dw0001001000001 Vertical spillovers Positive spill. Nace-2D2120101211030 Dw3023041112303 Nace-3D2020111101421 Dw2111011201401 Negative spill. Nace-2D4023022010401 Dw3022211113241 Nace-3D2001012210102 Dw1020311101422 :Each cell gives a number of countries with a significant coefficient (at 10 per cent at the least). * Foreign affiliates are excluded from the summary, except for direct effects.

23. Results (2): Summary of results  Direct effects of foreign ownership on firm performance are only present in 3 out of 10 countries => when present, strictly positive.  Horizontal spillovers mostly negative when not controlling for the absorptive capacity of firms. When accounting for firms’ absorptive capacity, in 6 out of 10 countries, they are positive.  Positive horizontal spillovers are equally distributed across size classes of firms, negative horizontal spillovers seem to be more likely for smaller firms.  Positive horizontal spillovers seem more likely to be present in medium or high productivity firms with higher absorptive capacities, negative horizontal spillovers are more likely to affect low to medium productivity firms.  Vertical spillovers are less frequent than horizontal spillovers from FDI. If present, smaller and more productive firms are more likely to benefit from positive vertical spillovers, while larger and less productive firms are more likely to suffer from negative vertical spillovers. 23 Institute of Economic Research and University of Ljubljana Milan, 27-28 February 2014

24. Results (3): Direct effects - Matching  Measuring direct effects by the matching approach:  Apart from measuring direct effects by ‘OLS on first differenced log TFP’ we also measure them by using the matching and the average treatment effect techniques. This allows to controll for more heterogeneity.  We use firm propensity to become foreign owned to match foreign owned firms with otherwise similar non-foreign owned firms in order to evaluate the effect of foreign ownership on productivity growth  We compare cohorts of fairly similar foreign and domestic firms over time. 24 Institute of Economic Research and University of Ljubljana Milan, 27-28 February 2014

25. Results (4): Direct effects from FDI – Impact of foreign ownership on firm TFP growth [ATT effects with nearest neighbor matching] 25

26. Results (5): Direct effects -Matching  Foreign owned firms persistently outperform domestic firms in terms of TFP growth only in the Czech Republic and Slovenia.  In Estonia, Latvia and Poland, productivity gains are only observed in the first year after the change in ownership, and seem to dissipate afterwards.  In Croatia, Romania, and Bulgaria benefits of foreign ownership become significant in the second, third and fourth year after the switch in ownership and do not persist (except in Bulgaria).  Direct productivity improvements from foreign ownership are far from being general, but are subject to foreign affiliate heterogeneity => productivity gains widely differ across size and productivity classes, and with regard to the time period after the ownership change. 26 Institute of Economic Research and University of Ljubljana Milan, 27-28 February 2014

27. Conclusions  Heterogeneity of firms is important  Horizontal spillovers have become increasingly important over the last decade  even more important than vertical spillovers  Positive horizontal spillovers:  equally distributed across size classes  in medium or high productivity firms with higher absorptive capacity  Negative horizontal spillovers:  in smaller firms  in low to medium productivity firms Budapest, 20-21 June 2013 Institute of Economic Research 27