![Sample our Economics, Finance,Business & Industry journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5931/TOC-Subject-Sample-2021-Economics-Finance-Business-Industry.jpg"})
Abstract
We highlight the difference between the service sector and the manufacturing sector in regard to the determinants for a firm to start foreign direct investment (FDI) and the resulting productivity growth. This paper analyzes two questions: (1) whether a certain level of initial productivity explains a firm's choice and (2) how the productivity of such a multinational firm changes over time after FDI. Using the longitudinal panel data on Japanese firms from 1980 to 2005, we trace firm-level decisions over decades. We find that the total factor productivity (TFP) does not explain a firm's choice for starting FDI in manufacturing, but it does in service sector. In contrast, the size and the profitability of firms motivate their FDI in the manufacturing sector, but these do not matter in the service sector. In addition, after FDI, we observe 1.3% annual growth rates of TFP in service, whereas firms in manufacturing show only 0.9% growth rates. To eliminate the selection bias, each multinational firm is paired with domestic firm(s) with similar ex-ante characteristics in the same industry.
Acknowledgements
The author is very grateful to the referees for their helpful comments.
Notes
1. The world outward FDI stock has increased from 2.1 trillion dollars in 1990 to 20.9 trillion dollars in 2010. FDI volume in service comprises 14.2 trillion dollars in 2010.
2. Kimura and Kiyota Citation(2006) show the association between FDI, export, and the TFP growth using 1994–2000 observations of Japanese firms. The data source is The Basic Survey of Business Structure and Activity accessible by confidential agreements. Using the initial TFP levels as controls, they state that firms with foreign presence become more productive than others.
3. Head and Ries Citation(2003) use the data of listed firms by Toyo Keizai Inc., and Tomiura Citation(2007) uses The Basic Survey of Commercial and Manufacturing Structure and Activity (Sho-Kogyo Jittai Kihon Chosa, in Japanese) by Ministry of Economy, Trade, and Industry (METI).
4. Our research is based on Hijzen, Inui, and Todo Citation(2010) and Navaretti, Castellani, and Disdier Citation(2010). While their measures of TFP are based on a standard Cobb–Douglas function, our measure is called multilateral TFP as explained in the appendix. The Cobb–Douglas assumption is commonly used for manufacturing industries, but not for service industries. The TFP index in our paper therefore does not require any specific functional forms of production. Thus, we avoid using this strong assumption for service production.
5. Fukao, Kim, and Kwon Citation(2007) investigate the sources of productivity both in service and manufacturing in Japan, but they mainly consider the influence of domestic competition. As another evidence, Amiti and Wei Citation(2009) show that foreign service activities (service outsourcing) have enhanced the value added of manufacturing industries, using US data of 1992–2000. But they have not discussed the parallel analysis for US service industries.
6. We follow Fujita, Krugman, and Venables Citation(1999) for this setup.
7. In 1980, local sales to consumers (horizontal FDI) comprised 46.2% of the total manufacturing sales according to the information of Basic Survey of Business Activity Abroad. In the early 1980s the motive for manufacturing FDI was tariff-jumping, to escape restrictions on cross-border sales in developed economies. As the appreciation of Yen fuels export and FDI, accessing low costs of production abroad has gradually become the primary motive in manufacturing.
8. The price index in country j is .
9. The fixed cost here is proportional to the marginal production cost. This assumption reduces computations, but does not affect the theoretical indications.
10. If Z ∼ Pareto(f), . (zo is the cutoff productivity level for active operation.)
11. The price is a constant markup over cost, .
12. Initially, v = i, i.e. a firm exports from home.
13. The firm also needs to be more competitive than the producers in other countries (indexed as v ∈ Ω).
14. Export destinations are indexed as j. The competitors’ locations are indexed as v.
15. The time dimension of a firm's decision is in reality sequential and continuous. However, we assume that every entry decision is based on the myopic expected payoff.
16. One unit change in Z in service industry is 3.4–4.5 times as influential as that in manufacturing. We choose 24 major economies where Japanese foreign affiliates are located. We use the wage data from United Nations Industrial Development Organization (UNIDO) for ck and cv. We use CIF/FOB data of IMF (the ratio of import costs inclusive of insurance and freight, to compare costs exclusive of insurance and freight) as trade costs for country pairs: tvk, tkj, tvj. We use the real GDP data as the market sizes: Mj and Mk. The elasticity of substitution σ is assigned as 4 based on the empirical results of Lai and Chun Zhu Citation(2004). Other estimated values of σ in their paper are also applied.
17. Firms in finance or in insurance service are omitted from the database. We further omit firms in agriculture, mining, and construction, to make our definition of the service industry comparable to the Japan Standard Industrial Classification (JSIC) as of March 2003.
18. The combination excludes the listed firms in some new security exchanges (Jasdaq, Hercules, and Mothers). Thus, 73 firms in Development Bank of Japan (DBJ) Data Bank are left unmatched with the Toyo Keizai database.
19. Before 1980, we unfortunately did not have sufficient firm-level statistics for measuring TFP and other related industry-level deflators.
20. We alternatively apply probit estimation. The results about significance and relative impacts of coefficients are unchanged. Upon contact with the author, results using probit with parallel methods are available.
21. In this index, the hypothetical representative firm has TFP = 1, i.e. TFP = 0. The hypothetical firm's input shares are equal to the arithmetic means of input shares, and its output and input quantities are the geometric means of output and input quantities. The cumulative changes in
TFP (from 1980) is negative, showing the relative decline in the TFP growth rates during the sample period. The macro data show that Japanese economy has been stagnant since 1991.
22. We observe the negative effect of the past capital intensity on the entry in the service sector. This comes from the industrial specificity, and the lack of information on the industry-level deflators. In the service sector, real estate companies are ranked high in capital intensity, but they have low FDI frequencies. In addition, we deflate capital and labor values by industry-level price indices. However, the precision of deflators differ by industry.
23. The reason for low pseudo R2s in the results is the omission of industry dummies. However, this omission (and low R2) will not be a problem since propensity scores of firms are matched for each industrial category. The industry-level difference is taken into account at the matching stage.
24. See Cameron and Trivedi Citation(2005) for detailed descriptions. To implement the matching based on p(Xi), three issues are relevant: (1) whether to match with or without replacement, (2) the number of units to use in the comparison set, and (3) the choice of matching method. We treat these issues as follows. First, we apply matching with replacement. This means that not all the non-treated observations are matched with the treated. Second, we use a single closest match to a treated case. In this way, we are left with a relatively large variance while reducing the bias. Third, through propensity score matching, the matching is simply based on a scalar-valued metric. This method can avoid the unsuccessful match which will arise if we set some high dimensional factors to compare.
25. When there are more than one firms with the nearest propensity score (scores are calculated up to three decimal points), we include all of these firms as the reference for one FDI entrant. We also impose the common support, by dropping treatment observations whose propensity score is higher than the maximum or less than the minimum propensity score of the controls.
26. There are a couple of reasons for this. First, we do one-to-many matching (e.g. 813 versus 285 in service) so the distributions are different. Second, we match only in the same industrial category, so the sample sizes of the matched pairs are smaller than that of the unmatched and the distribution of the matched spreads out more. These cause the variances to be large, and the t-values remain high even for the matched pairs. This rejects the hypothesis of ‘the mean equivalence’ that distributions (population mean and population variance) are equal.
27. We choose this estimation after implementing the F-test and Lagrange multiplier test over pooled-OLS (Ordinary Least Squares), and the Hausman test over fixed-effects GLS.
28. Here we omit the firms which started FDI before 1980. Since some firms started their FDI as early as 1933, we consider these as not relevant for estimating the post-FDI effects on the TFP. This cutoff eliminates a certain number of manufacturing giants, which set up foreign affiliates in earlier years as pioneers. Therefore, the proportion of entrants in the selection is 49.2% in manufacturing and 26.1% in service. In manufacturing, there is a decline in the proportion.
29. The correlation coefficients between profitability and multilateral TFP are 0.24 in manufacturing, 0.23 in service. The correlation coefficients between capital intensity and multilateral TFP are 0.36 in manufacturing, 0.27 in service. Other independent variables have correlation coefficients between −0.06 and 0.20.
30. The productive effects of the FDI on non-FDI firms are reported for manufacturing industry in the literature, and we infer that the same competitive pressures applies to service as well. Bernard, Jensen, and Schott Citation(2006) show, with the US manufacturing data of 1977–1997, that domestic firms adjust their product mix in response to import pressures, especially those from low-wage countries such as China and India. For example, they switch to less competitive industries, or to products with greater skill intensity. These movements toward comparative advantage enhance the overall productivity level of the country. Keller and Yeaple Citation(2009) consider positive externalities through trade and (incoming) FDI. With the data of manufacturing firms operating in the US in 1987–1996, they show that the FDI leads to significant domestic productivity gain, explaining 14% of productivity growth in the US.
31. We extract data of listed Japanese firms from the JECR (http://www.jcer.or.jp/eng/index.html). I would like to thank Young Gak Kim for his instruction on data and provision of related deflators.