DP
RIETI Discussion Paper Series 15-E-143
Input-Output Linkages of Japanese Affiliates in Mexico within NAFTA
KONDO Keisuke
RIETI
The Research Institute of Economy, Trade and Industry
http://www.rieti.go.jp/en/
RIETI Discussion Paper Series 15-E-143
December 2015
Input-Output Linkages of Japanese Affiliates in Mexico within NAFTA*
KONDO Keisuke †
RIETI
Abstract
This study empirically investigates input-output linkages of Japanese affiliates in Mexico
between the United States, Canada, Mexico, and Japan. Constructing a panel dataset at the
affiliate level between 1995 and 2012, this study identifies significant linkages for
international input-output flows between these countries. It finds that Japanese affiliates'
imports of inputs from the United States and Canada significantly increase their exports of
output to those countries. On the other hand, there is little evidence that their purchases of
inputs in Mexico increase their exports to the United States and Canada. In turn, local sales
of Japanese affiliates in Mexico are sustained by diversified intermediate inputs sourced
from the United States, Canada, Mexico, and Japan.
JEL classification: F14, F21
Keywords: Input-output linkages, Foreign direct investment, NAFTA
RIETI Discussion Papers Series aims at widely disseminating research results in the form of
professional papers, thereby stimulating lively discussion. The views expressed in the papers are
solely those of the author(s), and neither represent those of the organization to which the author(s)
belong(s) nor the Research Institute of Economy, Trade and Industry.
I am deeply grateful to Hisamitsu Saito for his helpful and insightful comments. I thank Sonoe Arai, Shota Araki,
Masahisa Fujita, Nobuaki Hamaguchi, Keiichiro Honda, Hiroshi Ikari, Satoshi Kawamura, Kozo Kiyota, Toshiyuki
Matsuura, Masayuki Morikawa, Atsushi Nakajima, Atsushi Ohyama, Yukiko Umeno Saito, Kensuke Teshima, Nobu
Yamashita, Hongyong Zhang, and participants of the RIETI-KEO Workshop “Productivity of Japanese Firms: Current
Status and Challenges,” RIETI Discussion Paper Seminar, and 29th annual meeting of the ARSC for their useful
comments and suggestions. Naturally, any remaining errors are my own. I am grateful to the Ministry of Economy,
Trade and Industry for providing the micro-data of the Basic Survey of Japanese Business Structure and Activities
(BSJBSA) and the Basic Survey on Overseas Business Activities (BSOBA). This research was conducted under the
project “Data Management” at the Research Institute of Economy, Trade and Industry (RIETI). This research uses the
RIETI converter table developed by the “Data Management” project in order to link two databases of the BSJBSA and
the BSOBA at the firm level.
†Research Institute of Economy, Trade and Industry (RIETI): 1-3-1 Kasumigaseki, Chiyoda-ku, Tokyo, 100-8901, Japan.
(E-mail: kondo-keisuke@rieti.go.jp)
*
2
1 Introduction
In recent decades, increasing numbers of firms have expanded their business worldwide alongside
decreasing trade costs, including communication costs as a result of the Internet. Some firms
import intermediate inputs from foreign counties and export products to the countries; other firms
undertake foreign direct investment (FDI). In particular, multinational firms do both. Among the
key features of the modern globalized economy is the global production network (Feenstra, 1998).
Firms construct international networks by dividing their production process into several stages
and locating them separately in optimal countries/regions, which makes the FDI of firms more
complex.1
Recent studies, such as Yeaple (2003), Ekholm et al. (2007), Neary (2009), Ito (2013), and Baldwin
and Okubo (2014), have paid much attention to export-platform FDI in addition to horizontal and
vertical FDI.2 The traditional concern of the international trade literature has been the relationship
between home and partner countries. Put another way, the standard two-country theoretical
model cannot capture a third country’s effect. Therefore, the importance of analyzing foreign
affiliates’ behavior, especially input–output structure with third countries, has grown increasingly
as the fragmentation of the production process is spread widely across different countries.
Since the free trade agreement (FTA) plays an important role in constructing global production
networks, this study aims to investigate connections between inputs and output for foreign
affiliates in terms of FTA networks. A new aspect of this study is to uncover how foreign affiliates
source inputs not only from the home country but also from third countries to export output
to third countries within the partner country’s FTA networks. For example, it is considered that
export-platform affiliates in the partner country have two choices for sourcing intermediate inputs:
from inside and outside the FTA network of the partner country. However, when these affiliates
export their output produced with non-FTA materials to the FTA member countries, the latter
channel could be relatively costly due to the rules of origins. Krishna and Krueger (1995) point
out that, although an FTA intends to promote trade liberalization, the rules of origin of FTA bring
about “hidden protection” to countries outside the FTA. As analyzed by Johnson and Noguera
(2012), Bombarda and Gamberoni (2013), and Conconi et al. (2015), FTA networks can have big
1
Fujita and Thisse (2006) emphasize that decreasing trade costs of goods as well as communication costs between
headquarter and production affiliates facilitate fragmentation of the production process.
2
See Greenaway and Kneller (2007) for a review of this literature.
3
impact on trade structure in intermediate goods.
This study focuses on Japanese affiliates in Mexico and their input–output linkages within the
North American Free Trade Agreement (NAFTA) by exploiting a unique panel dataset of Japanese
affiliates. The micro-dataset of Japanese foreign affiliates covers the post-NAFTA period between
1995 and 2012 and includes additional information on export and import values to/from the US
and Canada. A favorable feature of this dataset is that it can distinguish local sales and purchases,
exports and imports to/from Japan, and exports and imports to/from the US and Canada from
total sales and purchases of the affiliates, enabling us to identify significant linkages between
these international input–output flows. In other words, distinction can be made between inputs
sourced from the NAFTA area and non-NAFTA countries.
The launching of the NAFTA posed a twofold problem for the export-platform affiliates in
Mexico. First, foreign affiliates in Mexico that source inputs and materials from outside Mexico for
exporting output to the US and Canada faced additional payment of their import tariff. Before the
NAFTA was enacted in 1994, Maquiladora programs in Mexico, which permitted duty-free imports
of inputs and materials under condition that output therefrom must be exported outside Mexico,
attracted FDI from around the world.3 However, article 303 of the NAFTA, which started in
January 2001 after a transition period of seven years, stipulates the elimination of duty-drawback.
Second, the rules of origin of the NAFTA affect cost competitiveness of export-platform
affiliates in Mexico within the NAFTA in terms of whether the NAFTA preferential tariff is
applicable when they export their products from Mexico to the US. The use of high value-added
intermediate inputs imported from non-NAFTA countries might violate the local content rate
stipulated in the rules of origin. In other words, firms that have production networks within the
NAFTA became relatively competitive.
Although the first issue was solved by some policies, the second issue arising from the
rules of origin still has a big impact on production networks within the NAFTA.4 To increase
3
The maquiladora program originally started in 1965 to promote industrialization in the northern border area of
Mexico. It was reformed in 2006 and is currently known as the the Manufacturing, Maquila and Export Services
Industry (IMMEX) program. See Bergin et al. (2009) for additional information on the maquiladora industry.
4
Regarding the first issue, the Mexican government started the Sectoral Promotion Programs (PROSEC) in January
2001 to strengthen industrial competitiveness and source intermediate inputs from outside the NAFTA in response to
the elimination of duty-drawback. In addition, Japan and Mexico started negotiations of the Japan–Mexico Economic
Partnership Agreement (EPA), which was launched in April 2005. As mentioned by Reyes (2012), the purpose of the
4
competitiveness within the NAFTA, export-platform Japanese affiliates in Mexico have incentives
to source intermediate inputs within the NAFTA. Thus far, an increasing number of Japanese firms
have undertaken FDI into the US, and thus, there is a growing need for uncovering input–output
linkages between the US and Mexico.
Figures 1 and 2 show Japan’s exports to the US and Mexico, and Japan’s FDI to the same
countries, respectively. The figures capture an important change in that the launch of the NAFTA
might have affected the export and FDI behavior of Japanese firms. Panel (a) of Figure 1 shows
that the export values of intermediate goods from Japan to the US increased until 1994 but were
sluggish from 1994, whereas export values of intermediate goods from Japan to Mexico show an
increasing trend from 1994 in Panel (b) of Figure 1.5 This might reflect a structural change in
Japanese firms’ exports to NAFTA countries. On the other hand, Panels (a) and (b) of Figure 2
show that an increasing number of Japanese firms undertook FDI into the US and Mexico in late
2000s.6 Thus, it is considered that production networks between Japan and NAFTA countries
have changed since the NAFTA was launched.
[Figures 1–2]
This study contributes to the existing literature in two ways. First, by focusing on Japanese
affiliates in Mexico, it clarifies foreign affiliates’ export and import behavior toward the US and
Canada. The literature still lacks understanding of how export-platform affiliates in Mexico source
inputs from inside and outside the NAFTA. This study finds that imports from the US and Canada
significantly increase exports to the US and Canada, suggesting that Japanese affiliates in Mexico
export their output mainly by combining inputs imported from the US and Canada with Mexican
labor. Second, this study uncovers how Japanese affiliates in Mexico source their inputs for local
sales in Mexico. Unlike the Japanese affiliates’ exports from Mexico to the US and Canada, their
local sales in Mexico are sustained by diversified inputs imported from the US, Canada, Mexico,
and Japan.
The remainder of this paper is organized as follows.
Section 2 explains the empirical
Japan–Mexico EPA negotiation originally differed from the promotion of bilateral trade between Japan and Mexico:
the Japanese business community was worried about losing competitiveness within the NAFTA.
5
Panel (b) of Figure 1 shows further increase in export values of intermediate goods from Japan to Mexico around
2001. The PROSEC might have promoted exports to Mexico.
6
Data on the number of Japanese foreign affiliates are limited to after 2006.
5
framework. Section 3 describes the data. Section 4 discusses the estimation results. Finally,
Section 5 concludes the discussion.
2 Empirical Framework
2.1 FDI into NAFTA and Japanese Firms’ Characteristics
To begin the analysis of international input–output linkages within the NAFTA, this study
examines what type of firms make FDI into NAFTA and Mexico. As shown in Helpman et al.
(2004), more productive firm can make FDI. To clarify the relationship between FDI and total
factor productivity (TFP), two empirical approaches are adopted. First, this study compares TFP
distributions between domestic firms and FDI firms into NAFTA countries/Mexico. This analysis
illustrates the differences between the two distributional patterns. The TFP distribution for FDI
firms is expected to be located on the right-hand side of that for domestic firms.
Second, this study explores what characteristics the FDI firms show by estimating the probit
model:
Pr(FDIik = 1|Zi ) = Φ(Zi βk ),
(1)
where FDIik takes a value of 1 if firm i has any affiliates in area/country k (k = NAFTA, Mexico)
and 0 if firm i has neither affiliates nor export in foreign countries; Zi is a vector of explanatory
variables including the firm average TFP, employment size, ratio of foreign capital, and firm age;
and Φ denotes the cumulative distribution function of the standard normal distribution. Note
that FDIik is constructed if firm i undertook FDI into area/country k at least once during the study
period. Similarly, the explanatory variables are constructed as the firm average value during the
period.
2.2 International Input–Output Linkages within NAFTA
The main concern of this study is to uncover the international input–output linkages of Japanese
affiliates in Mexico within NAFTA. This study focuses on the international input–output linkages
at the aggregate industry level. An advantage of the dataset is that foreign affiliates’ export and
import values to the US and Canada are included.7
7
The regional division of third-country exports and imports in the BSOBA is North America (the US and Canada),
Asia, Europe, and other countries. In addition, the dataset includes exports and imports to/from Japan as well as local
6
The empirical framework of this study is essentially similar to that of Feng et al. (2012), who
investigate the relationship between imported intermediate inputs and export performance of
Chinese manufacturing firms. However, in the present study, the export and import values are
disaggregated by country. The regression models to be estimated are:
J
N
NN
MN
JN
ON
N
N
N
log PN
log PM
log PO
log SN
it = α
it + α
it + α log Pit + α
it + X it β + ψi + τ jt + uit ,
M
NM
MM
JM
M
M
M
log PN
log PM
log PJit + αOM log PO
log SM
it = α
it + α
it + α
it + X it β + ψi + τ jt + uit ,
(2)
where Ssit (s = N, M; N=the US and Canada, M=Mexico) represents sales of Japanese affiliate i to
country s at year t, Prit represents purchases from country r (r = N, M, J, O; N=the US and Canada,
M=Mexico, J=Japan, O=other countries), X it is a vector of control variables (log of employment
size and operating years), ψsi is a fixed effect of affiliate i, τsjt is a cross effect between industry j
and year t, and usit is an error term.8 The industry-year effect τsjt is intended to control for complex
institutional changes in tariff and trade policy.
The parameters of interest are a set of αrs , which denotes the input–output elasticity between
country r and s. For example, αMN indicates by what percentage a 1% increase in purchases
from Mexico increases exports to the US and Canada. It is expected that αrs is essentially nonnegative (αrs ≥ 0). This empirical analysis aims to find a significantly positive sign for αrs , which
would uncover significant linkages in the international input–output flows of Japanese affiliates
in Mexico.
As a benchmark estimation, this study separately estimates regression models (2) without
affiliate fixed effects ψsi . However, an estimation issue here is that parameters cannot be exactly
estimated when the level-variation is used.
This issue is highly related with that regional
breakdowns of the total sales and purchases are used in the regressions. For example, we cannot
identify whether inputs imported from the US and Canada increase exports to the US and Canada
by αNN even if there is a positive relationship between them. Consider that inputs sourced
from Mexico increase exports to the US and Canada and that inputs sourced from the US and
Canada increase sales in Mexico. In this case, the parameters αMN and αNM should be estimated
sales and purchases.
8
To take the logarithm of sales Ssit and purchases Prit , this study calculates log(Ssit + 0.1) and log(Prit + 0.1). The unit
of sales and purchases are in millions of JPY. In addition, the regression models include dummy variables that take a
value of 1 if sales or purchases are 0 (Ssit = 0; Prit = 0) and 0 otherwise (Ssit > 0; Prit > 0).
7
significantly positive. However, αNN and αMM might be also estimated significantly positive as
spurious correlations if growing affiliates show bigger sales and purchases overall. To strictly find
significant input–output linkages, within-variation needs to be used, which means that this study
controls for affiliate fixed effects in the estimation.9
Estimates of αrs obtained by the fixed-effect estimation are expected to capture linkages on
input–output flows. The comparison between the benchmark OLS and fixed-effect estimations
provides robust evidence on input–output linkages of Japanese affiliate in Mexico within NAFTA.
3 Data
This study uses two data sources. First, for the FDI and TFP analysis, this study uses a firm-level
confidential micro-dataset from the Basic Survey of Japanese Business Structure and Activities
(BSJBSA), which has been conducted annually by the Ministry of Economy, Trade and Industry
(METI) of Japan since 1994 after the first survey in 1991. The BSJBSA covers all firms with 50
employees or more and capital of 30 million JPY or more. Firm-level TFP is estimated by the data
of the BSJBSA. The details of the TFP estimation are discussed in Appendix A.
Second, to develop a list of headquarters with affiliates locating in the NAFTA and Mexico,
this study takes the confidential micro-dataset of Japanese foreign affiliates from the Basic Survey
on Overseas Business Activities (BSOBA), which is annually conducted by the METI. The BSOBA
covers firms that have foreign affiliates, except for the insurance/finance and real estate industries.
The definition of a foreign affiliate in this survey is a foreign company in which the Japanese
parent firm owns more than 10% share of investment or in which a foreign company with 50% or
more of investment from a Japanese firm has 50% of the investment.
The BSOBA has two types of questionnaires, one for headquarters and another for those
affiliates. To make FDI dummy variables at the firm level, this study develops list of headquarters
with affiliates in the NAFTA and Mexico using the BSOBA. Then, this information is linked to the
BSJBSA using the converter table developed by the RIETI. Variable FDIik takes a value of 1 if firm i
has at least one affiliate in region/country k during the study period, 1995–2011, and 0 otherwise.10
9
First difference estimation may be used instead of fixed-effect estimation. The preference in this study is for
fixed-effect estimation because some affiliates are dropped owing to the unbalanced panel dataset. Some Japanese
affiliates do not continually appear each year.
10
As for the FDI and TFP analysis, the 2012 BSJBSA and BSOBA are not used owing to the unavailability of deflators
8
The comparison group comprises domestic firms that do not export and have no affiliates in
foreign firms. The BSJBSA asks the number of affiliates that firms have in foreign countries and
the export values of firms.11 Therefore, this study focuses on the comparison between FDI firms
and domestic ones.
For the analysis of input–output linkages within NAFTA, I make a panel dataset at the affiliate
level between 1995 and 2012 by using the questionnaires for foreign affiliates. More importantly,
the BSOBA includes sales and purchases by aggregate region, such as Japan, North America
(the US and Canada), Asia, and Europe. Exploiting this data structure of the BSOBA, this study
distinguishes export and import values to/from the US and Canada from total sales and purchases.
Furthermore, the BSOBA contains local values of sales and purchases—that is, sales and purchases
in Mexico—by focusing on the Japanese affiliates in Mexico. This data structure allows us to
clarify international input–output linkages of the Japanese affiliates in Mexico within the NAFTA.
Nominal sales and purchases are deflated by price levels of gross output and intermediate inputs
in Mexico (1995=100). See Appendix B for more details of deflators.
Table 1 presents descriptive statistics of variables for FDI and TFP analysis. Each 0.25% upper
and lower observations for value-added, labor, capital, and purchases of intermediate inputs are
excluded as outliers. In addition, firms that appear only once in the BSJBSA are excluded. The
final sample size is 30,936. In our sample, the numbers of firms undertaking FDI into the NAFTA
and Mexico are 1,620 and 115, respectively, and the number of domestic firms is 29,316.
Table 2 presents descriptive statistics of variables for input–output linkage analysis. The
BSOBA has 2,968 observations of Japanese affiliates in Mexico between 1995 and 2012. This
study excludes observations with missing values for key variables and inconsistent breakdown
compared to total sales and purchases. In addition, the upper 1% of observations of distributions
in each variable is excluded from the sample except for operating years. Finally, the number of
observations is reduced to 1,013.
The ratios of sales in each country to total sales are shown at the bottom of Table 2. On
average, Japanese affiliates in Mexico have 74.3% share of local sales to total sales and 21.4% share
of exports to the US and Canada to total sales. Thus, the sum of local sales and export values to
the US and Canada reaches 95.7% of the total sales. On the other hand, intermediate inputs are
for gross output, intermediate inputs, and investment goods.
11
There are firms with missing values on affiliates in foreign countries and export values. For simplicity, it is
assumed that these firms are domestic.
9
sourced equally among Mexico, the US, Canada, and Japan.
Figure 4 shows the trend of intranational and international input–output linkages in terms
of shares of sales and purchases. Essentially, there is no big change in both sales and purchases
between 1995 and 2012. However, the share of exports to the US and Canada rises slightly between
2003 and 2008, while the share of sales in Mexico drops. On the other hand, the share of exports to
Japan is steadily low during the study period, while the share of imports from Japan is relatively
large.
Figure 5 presents sales-sourcing box plots, which are suggested by Baldwin and Okubo
(2014). Panels (a) and (b) show that the tertiary sector tends to have higher share of sales in
Mexico, whereas the machinery industry tends to have higher share of exports to the US and
Canada. In particular, the electric machinery industry exports output to the US and Canada.
The services sector, such as transportation and other services, mostly purchases inputs in Mexico.
The machinery industry sources relatively more intermediate inputs from the US and Canada,
compared to other industries.
[Tables 1–2 and Figures 4–5]
4 Estimation Results
4.1 More Productive Firms Undertake FDI into NAFTA
Figure 3 presents the TFP distributions for FDI firms into NAFTA/Mexico and domestic firms.
Panel (a) shows that the TFP distribution for FDI firms into NAFTA is right-shifted, and thus, the
FDI firms have higher TFP than the domestic firms. In Panel (b), the TFP distribution for FDI
firms into Mexico is slightly right-shifted, but the right-shift is not as big as the case of FDI into
NAFTA. Therefore, the right-shift in Panel (a) is because more productive firms undertake FDI
into the US and Canada within the NAFTA, rather than into Mexico. On the whole, these results
are consistent to Helpman et al. (2004).
Table 3 presents the results of probit estimations that investigate what type of firms undertake
FDI into NAFTA/Mexico. As discussed in Figure 3, Column (1) confirms that firms with higher
TFP undertake FDI into the NAFTA. Column (2) shows that this result holds even after controlling
for employment size, foreign capital ratio, and firm age. Column (3) shows firms with higher TFP
undertake FDI into Mexico. However, this is not true after controlling for other factors in Column
10
(4). Although it is true that productive firms undertake FDI into Mexico, higher TFP of these firms
is explained by larger employment size, higher ratio of foreign capital, and older firms.
[Figure 3 and Table 3]
4.2 Input–Output Linkages of Export-Platform Japanese Affiliates in Mexico
Table 4 presents the results of input–output analysis for the case of export to the US and Canada.
Columns (1) and (2) show the estimation results for all sectors; Columns (3) and (4) show the
estimation results for the manufacturing sector. Benchmark estimation results by OLS estimation
in Column (1) do not show that imports from the US and Canada have an impact on exports to the
US and Canada. However, fixed-effect estimation in Column (2) captures this significant linkage
between exports to the US and Canada and imports from the US and Canada. This difference
between OLS and fixed-effect estimations implies that the controlling for affiliate fixed effects is
important to estimate αrs .
An interesting finding in Columns (2) and (4) is that imports from Mexico and Japan do not have
significant impacts on exports to the US and Canada. In other words, export-platform Japanese
affiliates in Mexico do not necessarily source inputs from Japan. As pointed out by Krishna and
Krueger (1995), the estimation results might suggest that the rules of origin of NAFTA make
exports of Japanese affiliates in Mexico to the US and Canada difficult—that is, high value-added
inputs imported from Japan cannot satisfy the local content rate stipulated by the rules of origin of
the NAFTA and, as a result, the export-platform Japanese affiliates in Mexico need to source high
value-added inputs from the US and Canada to apply NAFTA preferential tariffs to exporting
output.
Furthermore, export-platform Japanese affiliates in Mexico do not significantly source inputs in
Mexico. Indeed, Hoshino (2014), who investigates the global supply chain of Japanese automobile
companies in Mexico by conducting detailed interviews, points out that Mexican suppliers in the
automobile sector have difficulty entering multinational firms’ supply chains. The estimation
results of this study might reflect this background.12
The estimation results for the manufacturing sector are similar for all sectors—that is, imports
from the US and Canada have significant impact on exports to the US and Canada in Column (4).
12
However, a recent rise of Japanese firms’ FDI into Mexico, especially Japanese suppliers’ entry into Mexico, may
increase local purchases in Mexico.
11
Another finding is that employment size significantly increases exports to the US and Canada. In
addition, the magnitude of employment size in the manufacturing sector is bigger than that in
all sectors, implying that export-platform Japanese affiliates in the manufacturing sector exploit
Mexican labor, compared to other sectors.
[Table 4]
4.3 Input–Output Linkages of Japanese Affiliates in Mexico for Local Sales
Table 5 presents results of input–output analysis for the case of sales in Mexico. Benchmark
estimation results in Column (1) show that inputs sourced within NAFTA and from non-NAFTA
countries have significant impacts on the sales in Mexico. However, the fixed-effect estimation
in Column (2) shows that the inputs sourced from the non-NAFTA countries except Japan have
no impact. In addition, there are big differences between OLS and fixed-effect estimates. For
example, the coefficient estimate of employment size is 0.133 in Column (1); however, it is 0.367 in
Column (2). As mentioned earlier, using within-variation apparently leads to the correct estimates
of parameters.
The manufacturing sector essentially shows similar estimation results to all sectors. Column
(4) of Table 5 shows that the sales in Mexico are sustained by diversified intermediate inputs
sourced from the US and Canada as well as Mexico and Japan. Comparing between Columns (2)
and (4), Japanese affiliates in the manufacturing sector tend to have higher labor share, compared
to those in other sectors.
To sum up, the export-platform Japanese affiliates in Mexico, which mainly export their
products to the US and Canada, source intermediates inputs from the US and Canada. The
Japanese affiliates in Mexico that mainly sell their products in Mexico source intermediates inputs
from the US and Canada as well as Mexico and Japan.
[Table 5]
4.4 Extension Using Shares of Sales and Purchases
Table 6 shows the estimation results obtained from the variations in export and import shares
for all sectors. To confirm the previous finding in Tables 4–5, this section focuses on the linkages
between export share to the US and Canada and import share from the US and Canada. Note
12
that changes in the import share from the US and Canada contains simultaneous changes in the
purchase share in Mexico or the import share from Japan and other countries.
In Columns (1) and (2) of Table 6, OLS and fixed-effect estimation results show that the increase
in the import share from the US and Canada has positive impact on the export share to the US
and Canada, whereas it has negative impact on the sales share in Mexico, as shown in Columns
(3) and (4) of Table 6. This implies that the import share from the US and Canada has a substitute
relationship between the export share to the US and Canada and sales share in Mexico. However,
note that, as shown in Tables 4 and 5, the increase in the import from the US and Canada has
positive impacts on both exports to the US and Canada and sales in Mexico.
Table 7 shows the estimation results obtained from the variations in export and import shares
for the manufacturing sector. Results in the manufacturing sector are essentially similar to those
in all sectors: the increase in the import share from the US and Canada has positive impact on the
export share to the US and Canada, whereas it has negative impact on the sales share in Mexico.
The magnitude of the import share from the US and Canada obtained by fixed-effect estimation
is smaller than that of the OLS estimation.
Another difference between the OLS and fixed-effect estimation appears in the impact of
employment in Tables 4 and 5. Although the employment has substitute effect between the export
share to the US and Canada and sales share in Mexico in OLS estimation, this is not found in the
fixed-effect estimation, suggesting that the increase in employment does not change the structure
of sales shares in each country.
As shown in Table 4, imports of inputs from the US and Canada increase exports of output to
the US and Canada relatively larger than sales in Mexico, which is also supported by variations
in the export and import shares. These results suggest that export-platform Japanese affiliates in
Mexico has stronger input–output linkages with the US and Canada than with Mexico and Japan.
[Tables 6–7]
5 Conclusion
The purpose of this study is to uncover significant linkages on input–output flows of Japanese
affiliates in Mexico. The launching of the NAFTA gives Japanese firms with export-platform
affiliates in Mexico the incentive of reconstructing production networks within the NAFTA to
13
strengthen their competitiveness. Against such a background, focusing on international input–
output linkages between the US, Canada, Mexico, Japan, and other third countries, this study has
investigated export and import behavior of Japanese affiliates in Mexico.
This study finds that more productive firms undertake FDI into the NAFTA countries and
Mexico, compared to Japanese domestic firms. Furthermore, the Japanese affiliates in Mexico that
export to the US and Canada tend to source inputs from the US and Canada, not from Japan.
On the other hand, the Japanese affiliates in Mexico that mainly sell their products in Mexico
source locally as well as from the US, Canada, and Japan. These estimation results suggest that
export-platform Japanese affiliates in Mexico have strong connections with the US and Canada
in production networks within the NAFTA, and that the inputs imported from Japan are not
necessarily embedded for their output. These empirical findings provide important implications
for trade policies.
We note some limitations of this study. This study simply investigates input–output linkages
of Japanese affiliates in Mexico without looking at tariff changes across countries because of
the complex institutional changes.
Another limitation is that this study does not focus on
heterogeneity across industries owing to small sample size. For example, Fujita and Gokan
(2005) emphasize differences in trade costs between transportation and electric industries. Further
research is expected to solve these limitations.
References
[1]
Baldwin, Richard and Toshihiro Okubo (2014) “Networked FDI: Sales and sourcing patterns
of Japanese foreign affiliates,” World Economy 37(8), pp. 1051–1080.
[2]
Bergin, Paul R., Robert C. Feenstra, and Gordon H. Hanson (2009) “Offshoring and volatility:
Evidence from Mexico’s maquiladora industry,” American Economic Review 99(4), pp. 16641671.
[3]
Bombarda, Pamela and Elisa Gamberoni (2013) “Firm heterogeneity, rules of origin, and
rules of cumulation,” International Economic Review 54(1), pp. 307–328.
[4]
Conconi, Paola, Manuel Garcı́a-Santana, Laura Puccio, and Roberto Venturini (2015) “From
final goods to inputs: The cascade effect of preferential rules of origin.” Mimeo.
(URL: http://www.ecares.org/ecare/personal/conconi$/web/RoO.pdf).
14
[5]
Ekholm, Karolina, Rikard Forslid, and James R. Markusen (2007) “Export-platform foreign
direct investment,” Journal of the European Economic Association 5(4), pp. 776–795.
[6]
Feenstra, Robert C. (1998) “Integration of trade and disintegration of production in the
global economy,” Journal of Economic Perspectives 12(4), pp. 31–50.
[7]
Feng, Ling, Zhiyuan Li, and Deborah L. Swenson (2012) “The connection between imported
intermediate inputs and exports: Evidence from Chinese firms.” NBER working paper No.
18260.
[8]
Fujita, Masahisa and Toshitaka Gokan (2005) “On the evolution of the spatial economy with
multi-unit · multi-plant firms: The impact of IT development,” Portuguese Economic Journal
4(2), pp. 73–105.
[9]
Fujita, Masahisa and Jacques-François Thisse (2006) “Globalization and the evolution of the
supply chain: Who gains and who loses?” International Economic Review 47(3), pp. 811–836.
[10]
Greenaway, David and Richard Kneller (2007) “Firm heterogeneity, exporting and foreign
direct investment,” Economic Journal 117(517), pp. F134–F161.
[11]
Helpman, Elhanan, Marc J. Melitz, and Stephen R. Yeaple (2004) “Export versus FDI with
heterogeneous firms,” American Economic Review 94(1), pp. 300–316.
[12]
Hoshino, Taeko (2014) Supply Chain of Automotive Industry in Mexico, Chiba: IDE-JETRO. (in
Japanese).
[13]
Ito, Tadashi (2013) “Export-platform foreign direct investment: Theory and evidence,” World
Economy 36(5), pp. 563–581.
[14]
Johnson, Robert C. and Guillermo Noguera (2012) “Accounting for intermediates:
Production sharing and trade in value added,” Journal of International Economics 86(2), pp.
224–236.
[15]
Krishna, Kala and Anne O. Krueger (1995) “Implementing free trade areas: Rules of origin
and hidden protection,” in Levinsohn, James, Jim Levinsohn, and Alan V. Deardorff eds.
New Directions in Trade Theory, Ann Arbor: University of Michigan Press, Chap. 6, pp.
149–187.
[16]
Levinsohn, James and Amil Petrin (2003) “Estimating production functions using inputs to
control for unobservables,” Review of Economic Studies 70(2), pp. 317–341.
[17]
Neary, J. Peter (2009) “Trade costs and foreign direct investment,” International Review of
Economics & Finance 18(2), pp. 207–218.
15
[18]
Olley, G Steven and Ariel Pakes (1996) “The dynamics of productivity in the telecommunications equipment industry,” Econometrica 64(6), pp. 1263–1297.
[19]
Petrin, Amil, Brian P. Poi, and James Levinsohn (2004) “Production function estimation in
Stata using inputs to control for unobservables,” Stata Journal 4(2), pp. 113–123.
[20]
Reyes, Melba Falck (2012) “Japanese foreign direct investment in Mexico and the impact of
the global crisis,” Globalización, Competitividad y Gobernabilidad 6(1), pp. 36–54.
[21]
Timmer, Marcel P., Erik Dietzenbacher, Bart Los, Robert Stehrer, and Gaaitzen J. de Vries
(2015) “An illustrated user guide to the world input-output database: The case of global
automotive production,” Review of International Economics 23(3), pp. 575–605.
[22]
Yeaple, Stephen Ross (2003) “The complex integration strategies of multinationals and cross
country dependencies in the structure of foreign direct investment,” Journal of International
Economics 60(2), pp. 293–314.
Appendix A
TFP Estimation
This study estimates the firm level TFP using the method proposed by Levinsohn and Petrin (2003).
The value-added case of the Cobb-Douglas production technology is considered as follows:
vit = β0 + βl lit + βk kit + ωit + uit
where vit is the logarithm of value-added, lit is the logarithm of labor, and kit is the fixed capital.
The error term is assumed to consist of two components: the temporally transmitted shock ωit
that affects firm’s investment decision and the i.i.d. idiosyncratic shock uit , which has no impact
on the firm’s investment decision.
An estimation issue is that the OLS estimator of βk is inconsistent owing to the omitted variable
bias because kit is correlated with productivity shock ωit . Therefore, Levinsohn and Petrin (2003)
proposed a method to estimate the consistent estimator of βk, which is a modified version originally
proposed by Olley and Pakes (1996). After obtaining consistent estimates β̂l and β̂k by the method
of Levinsohn and Petrin (2003), the logarithm of TFP is available as follows:
η̂it = vt − β̂l lit − β̂k kit .
16
As a proxy variable of productivity shock, purchases of inputs are used.13
To consider
heterogeneity in production technology across industries, firm-level TFP by industry is estimated.
To make TFP comparable across industries, industry-year effects φ jt (industry j and year t) are
removed by running the following regression:
η̂ijt = φ jt + eijt ,
where eijt is the error term. Taking the residuals by subtracting the estimated φ̂ jt from the estimated
η̂ijt , the firm-level TFP comparable across industries is obtained as follows:
it ≡ êijt = η̂ijt − φ̂ jt
TFP
(3)
In the analysis of FDI into NAFTA and Mexico, it is assumed that firms made FDI (FDIik = 1)
if firms appear in the BSOBA at least once between 1995 and 2011. To keep consistency between
two datasets, average TFP across years is calculated as follows:
Ti
1 TFPit ,
TFPi =
Ti t
where Ti denotes the number of years for firm i.
In the TFP estimation, the BSJBSA is used to calculate the value-added vit , labor lit , and capital
kit . The net value-added is calculated as the sum of operating profit, wage bill, and taxes and
dues. Labor lit is calculated as total hours worked of regular and part-time workers. The average
hours worked by industry is obtained from the JIP 2014. The average hours worked for part-time
workers are calculated by the average hours worked of JIP 2014 and the ratio of hours worked
between regular and part-time workers. The ratio of hours worked is taken from the Monthly
Labor Survey of the Ministry of Health, Labour and Welfare. The capital kit is book value of
tangible assets deflated by the price level of investment goods (2000=100) calculated from the JIP
2014.14
13
The firm TFP is estimated by the Stata command levpet (Petrin et al., 2004).
14
The JIP 2014 offers deflators for gross output, intermediate inputs, and investment goods by industry (2000=100),
respectively. The deflators are available until 2011. This study ends up with 1995–2011 datasets of BSJBSA for the FDI
analysis.
17
Appendix B
Deflator for Sales and Purchases by Industry
The deflators of gross output and intermediate inputs in Mexico are obtained from the World
Input–Output Database (WIOD) (see Timmer et al., 2015). The socio-economic accounts of the
WIOD have the deflators of gross output and intermediate inputs by industry between 1995 and
2009 (1995=100). To complement deflators until 2012, the deflators of the National Institute of
Statistics and Geography (INEGI) are used. The INEGI offers monthly producer price indices
by industry for final and intermediate goods from 2007. First, annual average producer price
indices by industry are calculated. Second, industry classification between the WIOD and INEGI
is matched. Then, the deflators of the WIOD are extended until 2012 by percentage changes
calculated from the INEGI’s deflators.
The industrial classification of the WIOD differs from that of the BSOBA. First, the industrial
classification is organized to keep their consistency across years. Then, the industrial classification
of the WIOD is matched with that of the BSOBA. Finally, the real sales and purchases (1995=100) are
calculated as nominal sales and purchases deflated by price levels of gross output and intermediate
goods by industry, respectively.
18
Table 1: Descriptive Statistics for FDI and TFP Analysis
Variable
Dummy for FDI into NAFTA
Dummy for FDI into Mexico
Average Logarithm of TFP
Average Logarithm of Employment
Average Ratio of Foreign Capital
Average Firm Age
Obs.
Mean
S.D.
Median
30936
29431
30936
30936
30936
30936
0.052
0.004
1.301
5.098
1.006
34.203
0.223
0.062
0.475
0.935
7.636
16.701
0.000
0.000
1.301
4.864
0.000
34.400
Notes: Variables are averaged across years observed. The FDI dummy takes a value of 1 if a firm has at least one
affiliate in the corresponding country during the study period 1995–2011 and 0 otherwise. Each 0.25% upper and lower
observations for value-added, labor, capital and purchases of intermediate inputs are excluded as outliers. In addition,
firms that appear only once in the BSJBSA are excluded.
19
Table 2: Descriptive Statistics for Input–Output Linkage Analysis
Variable
Sales in Mexico
Exports to Japan
Exports to Third Countries
Exports to the US and Canada
Exports to Other Countries
Purchases in Mexico
Imports from Japan
Imports from Third Countries
Imports from the US and Canada
Imports from Other Countries
Employment
Operating Years
Ratio of Sales in Mexico to Total Sales
Ratio of Exports to Japan to Total Sales
Ratio of Exports to Third Countries to Total Sales
Ratio of Exports to the US and Canada to Total Sales
Ratio of Purchases in Mexico to Total Purchases
Ratio of Imports from Japan to Total Purchases
Ratio of Imports from Third Countries to Total Purchases
Ratio of Imports from the US and Canada to Total Purchases
Mean
S.D.
Median
1380.006
18.489
397.177
348.210
48.968
557.560
375.368
319.427
229.914
89.513
369.909
13.052
2676.238
123.340
1024.183
987.021
193.099
1324.786
807.134
845.744
727.376
357.733
819.345
11.050
407.122
0.000
2.760
0.000
0.000
52.689
60.352
39.929
2.457
0.000
96.000
10.000
0.743
0.017
0.241
0.214
0.377
0.318
0.305
0.233
0.377
0.091
0.369
0.357
0.397
0.353
0.379
0.358
0.990
0.000
0.004
0.000
0.200
0.174
0.100
0.009
Notes: The number of observations is 1,013. The unit of real values for sales, exports, purchases, and imports is
million JPY. The values of sales and exports are deflated by the price indices of gross output obtained from the World
Input–Output Database. In addition, the values of purchases and imports are also deflated by the price indices of
intermediate inputs obtained from the World Input–Output Database (1995=100). The upper 1% of distributions of
each variable, except operating years, are excluded from the sample. Inconsistent observations, in which the sum of
export values to Japan, the US and Canada, and other countries does not equal export values to the third countries, are
dropped from the sample.
20
Table 3: Probit Estimation Results for FDI and TFP
Dependent Variable: Dummy (1=FDI Firms; 0=Domestic Firms)
FDI into NAFTA
Explanatory Variables
TFP
FDI into Mexico
(1)
(2)
(3)
(4)
0.592***
(0.026)
0.274***
(0.061)
−3.035***
(0.092)
0.034
(0.082)
0.464***
(0.032)
0.012***
(0.002)
0.022***
(0.002)
−6.300***
(0.250)
29431
0.011
1492.164
29431
0.263
1118.443
Constant
−2.456***
(0.041)
0.433***
(0.033)
0.492***
(0.012)
0.011***
(0.001)
0.023***
(0.001)
−5.872***
(0.090)
Number of Observations
Pseudo R2
AIC
30936
0.043
12166.993
30936
0.265
9350.280
log(Employment)
Ratio of Foreign Capital
Firm Age
Notes: Heteroskedasticity-consistent standard errors are in parentheses. * denotes statistical significance at the 10%
level, ** at the 5% level, and *** at the 1% level.
21
Table 4: Results of Input–Output Linkage Analysis for Exports to the US and Canada
Dependent Variable: log(Exports to the US and Canada)
All Sectors
Explanatory Variables
log(Imports from the US and Canada)
log(Purchases in Mexico)
log(Imports from Japan)
log(Imports from Other Countries)
log(Employment)
Operating Years
Operating Years (/100)
Industry × Year Dummy
Control of Affiliate Fixed Effects
Number of Observations
Number of Affiliates
Adjusted/Within R2
Manufacturing Sector
(1)
(2)
(3)
(4)
0.058
(0.060)
−0.014
(0.049)
0.007
(0.054)
0.171**
(0.072)
0.300***
(0.069)
0.015
(0.018)
−0.071*
(0.042)
Yes
No
0.154***
(0.047)
−0.083
(0.057)
−0.028
(0.073)
0.157***
(0.053)
0.405***
(0.135)
0.124**
(0.058)
−0.089
(0.076)
−0.083
(0.096)
0.280***
(0.089)
0.468***
(0.139)
Yes
Yes
0.049
(0.069)
−0.066
(0.058)
0.066
(0.072)
0.215*
(0.122)
0.314***
(0.094)
0.069**
(0.027)
−0.225***
(0.078)
Yes
No
1013
209
0.906
1013
209
0.867
653
138
0.920
653
138
0.915
Yes
Yes
Notes: Heteroskedasticity-consistent standard errors clustered at the affiliate level are in parentheses. The constant
term is not reported. * denotes statistical significance at the 10% level, ** at the 5% level, and *** at the 1% level.
22
Table 5: Results of Input–Output Linkage Analysis for Sales in Mexico
Dependent Variable: log(Sales in Mexico)
All Sectors
Explanatory Variables
log(Imports from the US and Canada)
log(Purchases in Mexico)
log(Imports from Japan)
log(Imports from Other Countries)
log(Employment)
Operating Years
Operating Years (/100)
Industry × Year Dummy
Control of Affiliate Fixed Effects
Number of Observations
Number of Affiliates
Adjusted/Within R2
Manufacturing Sector
(1)
(2)
(3)
(4)
0.182***
(0.050)
0.362***
(0.056)
0.403***
(0.055)
0.098**
(0.046)
0.133**
(0.061)
0.006
(0.017)
0.004
(0.032)
Yes
No
0.115***
(0.034)
0.199***
(0.039)
0.214***
(0.048)
0.006
(0.035)
0.367***
(0.115)
0.089**
(0.043)
0.237***
(0.066)
0.192***
(0.072)
−0.023
(0.049)
0.422**
(0.180)
Yes
Yes
0.152**
(0.069)
0.435***
(0.066)
0.299***
(0.075)
0.135
(0.088)
0.120
(0.074)
−0.013
(0.031)
0.070
(0.077)
Yes
No
1013
209
0.897
1013
209
0.888
653
138
0.914
653
138
0.907
Yes
Yes
Notes: Heteroskedasticity-consistent standard errors clustered at the affiliate level are in parentheses. The constant
term is not reported. * denotes statistical significance at the 10% level, ** at the 5% level, and *** at the 1% level.
23
Table 6: Results of Input–Output Linkage Analysis in All Sectors
Dependent Variable: Sales Share to Country k
s=N
Explanatory Variables
Import Share from the US and Canada
log(Employment)
Operating Years
Operating Years (/100)
Industry × Year Dummy
Control of Affiliate Fixed Effects
Number of Observations
Number of Affiliates
Adjusted/Within R2
s=M
(1)
(2)
(3)
(4)
0.126**
(0.054)
0.040**
(0.019)
−0.002
(0.006)
−0.002
(0.010)
Yes
No
0.098*
(0.052)
−0.012
(0.014)
−0.101*
(0.053)
0.031
(0.032)
Yes
Yes
−0.110*
(0.056)
−0.041**
(0.020)
0.002
(0.006)
0.000
(0.011)
Yes
No
1013
209
0.240
1013
209
0.284
1013
209
0.224
1013
209
0.299
Yes
Yes
Notes: Heteroskedasticity-consistent standard errors clustered at the affiliate level are in parentheses. The constant
term is not reported. * denotes statistical significance at the 10% level, ** at the 5% level, and *** at the 1% level. N=the
US and Canada, M=Mexico.
24
Table 7: Results of Input–Output Linkage Analysis in the Manufacturing Sector
Dependent Variable: Sales Share to Country k
s=N
Explanatory Variables
Import Share from the US and Canada
log(Employment)
Operating Years
Operating Years (/100)
Industry × Year Dummy
Control of Affiliate Fixed Effects
Number of Observations
Number of Affiliates
Adjusted/Within R2
s=M
(1)
(2)
(3)
(4)
0.213**
(0.085)
0.048*
(0.027)
0.010
(0.011)
−0.038
(0.025)
Yes
No
0.130
(0.079)
−0.015
(0.022)
−0.136*
(0.079)
−0.002
(0.024)
Yes
Yes
−0.206**
(0.086)
−0.059**
(0.027)
−0.007
(0.011)
0.031
(0.026)
Yes
No
653
138
0.113
653
138
0.279
653
138
0.112
653
138
0.286
Yes
Yes
Notes: Heteroskedasticity-consistent standard errors clustered at the affiliate level are in parentheses. The constant is
not reported. * denotes statistical significance at the 10% level, ** at the 5% level, and *** at the 1% level. N=the US and
Canada, M=Mexico.
25
90000
12000
Final Goods
Intermediate Goods
Export Values (Million of Dollars)
Export Values (Million of Dollars)
100000
80000
70000
60000
50000
40000
30000
20000
1985
1990
1995
2000
Year
2005
2010
2015
10000
Final Goods
Intermediate Goods
8000
6000
4000
2000
0
1985
1990
(a) Export to US
1995
2005
(b) Export to Mexico
Figure 1: Japan’s Exports to the US and Mexico
Source: Data are from the RIETI-TID 2013 (http://www.rieti-tid.com/).
2000
Year
2010
2015
10
7000
8
6500
6
6000
4
5500
2
5000
2006
2007
2008
2009
2010
Year
2011
(a) FDI into US
2012
2013
0
2014
850
800
32
Growht Rate of Number of Affiliates
Total Number of Affiliates in Mexico
28
750
24
700
20
650
16
600
12
550
8
500
4
450
0
400
-4
350
2006
2007
2008
2009
2010
Year
2011
2012
2013
-8
2014
Growth Rate of Number of Affiliates (%)
7500
12
Growht Rate of Number of Affiliates
Total Number of Affiliates in US
Total Number of Affiliates
Total Number of Affiliates
8000
Growth Rate of Number of Affiliates (%)
26
(b) FDI into Mexico
Figure 2: Japan’s FDI to the US and Mexico
Source: Data are from the Annual Report of Statistics on Japanese Nationals Overseas (Ministry of Foreign Affairs of
Japan).
27
1.2
Domestic Firms
FDI Firms into NAFTA
1
Density Estimate
Density Estimate
1.2
.8
.6
.4
.2
Domestic Firms
FDI Firms into Mexico
1
.8
.6
.4
.2
0
0
−2
0
2
4
6
−2
0
TFP
(a) FDI Firms into NAFTA vs. Domestic Firms
2
4
6
TFP
(b) FDI Firms into Mexico vs. Domestic Firms
Figure 3: Comparison of TFP Distributions
Notes: The null hypothesis of equal distribution functions by the Kolomogorov-Smirnov test is rejected at the 1% level
in both Panes (a) and (b). TFP is the firm average value between 1995 and 2011. See Appendix A for more details of
TFP estimation. Domestic firms are those that do not export and had no establishment in foreign countries between
1995 and 2011, and number 29,316. FDI firms denote firms with affiliates in NAFTA countries or Mexico.
1.0
US and Canada
Mexico
Japan
0.8
0.6
0.4
0.2
0.0
1994 1996 1998 2000 2002 2004 2006 2008 2010 2012
Year
(a) Export Share by Country
Share of Sourcing by Country
Share of Sales by Country
28
1.0
US and Canada
Mexico
Japan
0.8
0.6
0.4
0.2
0.0
1994 1996 1998 2000 2002 2004 2006 2008 2010 2012
Year
(b) Import Share by Country
Figure 4: Regional Share of Sales between 1995 and 2012
Notes: The sample is the same as in Table 2. The average shares across Japanese affiliates in Mexico are shown by year.
Share of Sales in Mexico (%)
1.0
20
21
4
0.8
7
0.6
13
17
16
15
0.4
19
14
0.2
0.0
0.0
0.2
0.4
0.6
0.8
1.0
Share of Purchase in Mexico (%)
(a) Sales and Purchases in Mexico
Share of Sales to the US and Canada (%)
29
0.6
14
0.4
16
15
17
13
0.2
7
19
4
20
21
0.0
0.0
0.2
0.4
0.6
Share of Purchase from the US and Canada (%)
(b) Export to and Import from the US and Canada
Figure 5: Sales-Purchases Relationships by Industry
Notes: The sample is the same as in Table 2. The shares by industry are averaged across 1995–2012. The numbers denote
industrial classification: 4 is Food; 7 is Chemistry; 13 is General Machinery; 14 is Electrical Machinery; 15 is Information
and Communications Machinery; 16 is Transportation Machinery; 17 is Other Manufacturing; 19 is Transportation; 20
is Wholesale and Retail; 21 is Other Services. To maintain confidentiality, some industries are not shown if there are
less than five Japanese affiliates in each industry. The circle size indicates the sample size.