ISSN: 1439-2305 Number 345 – April 2018 FOREIGN INVESTMENT REGULATION AND FIRM PRODUCTIVITY: GRANULAR EVIDENCE FROM INDONESIA Robert Genthner and Krisztina Kis-Katos
ISSN: 1439-2305
Number 345 – April 2018
FOREIGN INVESTMENT REGULATION
AND FIRM PRODUCTIVITY:
GRANULAR EVIDENCE FROM
INDONESIA
Robert Genthner and Krisztina Kis-Katos
Foreign investment regulation and firm productivity:
Granular evidence from Indonesia∗
Robert Genthner† and Krisztina Kis-Katos‡
March 8, 2018
Abstract
Based on a yearly census of Indonesian manufacturing firms for 2000-2014, we
investigate the effects of a sector-specific investment policy reform on firm productiv-
ity. Hereby we exploit a protectionist foreign direct investment reform (the so-called
negative investment list) that designated certain sectors at the five-digit level to be-
come closed or only conditionally open to foreign investors. The list was first released
in 2000 and has been repeatedly revised by the Indonesian authorities since. Our
empirical analysis links the changes within this regulatory framework to variation in
firm-level productivity in a large firm panel. Controlling for an extensive set of fixed
effects as well as potential drivers of endogeneous regulation, we find robust evidence
of declining foreign capital shares in sectors subject to restrictions on foreign direct
investment, followed by a sizable decrease in firm productivity. From the different
types of conditions, sector-wide FDI bans were linked to the largest productivity de-
clines. We also document the presence of negative backward productivity spillovers
of regulation that propagate throughout the value chain.
JEL Classification: F23, L51, D24, F21, L6
Keywords: FDI, regulation, Indonesia, total factor productivity, spillovers
∗We would like to thank Rebecca Süss for excellent research assistance and Friederike Lenel, GüntherG. Schulze, Marcel Timmer and participants of seminars and workshops in Freiburg, Göttingen as well asconference participants at the ETSG in Florence, the IWB workshop in Göttingen and the FDI workshopin Mainz for helpful comments and suggestions. All remaining errors are our own.
†University of Göttingen, Germany‡University of Göttingen, Germany and IZA, Bonn
1
1 Introduction
In the course of the last two decades developing and emerging economies liberalized their
markets substantially. The process of globalization has not only lead to a successive
dismantling of trade barriers but has also facilitated the operation of multinational enter-
prises by liberalizing the inflows of foreign direct investment (FDI) world-wide. However,
this process of trade and market liberalization has not progressed uniformly and also
experienced numerous regulatory shifts and reversals, affecting FDI flows (Harding and
Javorcik 2011, Bourlès et al. 2013). While a substantial literature has documented links
between foreign participation and firm productivity (Aitken and Harrison 1999, Arnold
and Javorcik 2009),1 only few papers have studied the effects of FDI regulation on firm
productivity directly (Bourlès et al. 2013, Duggan et al. 2013, Eppinger and Ma 2017).
Foreign capital is expected to affect firm productivity through several channels. It can
substitute for domestic capital and relieve liquidity constraints if access to domestic cap-
ital is limited. Foreign investors have been shown to introduce nontangible productive
assets such as technological, managerial and marketing skills, trading contacts and rep-
utation (Aitken and Harrison 1999, Arnold and Javorcik 2009). As a result, firms with
foreign participation are typically more productive, more capital intensive and pay higher
wages (Harrison and Rodríguez-Clare 2010). Moreover, the effects of FDI can spill over
horizontally within industries or vertically along the value chain. Empirically, the evi-
dence is strongest in support of positive backward spillovers from foreign-invested firms
to their domestic suppliers, most likely taking place through targeted transfers of technol-
ogy (Javorcik 2004, Blalock and Gertler 2008, Barrios et al. 2011, Newman et al. 2015).
By contrast, studies documenting no or negative horizontal spillovers within the same
industry (Djankov and Hoekman 2000, Javorcik 2004, Blalock and Gertler 2008) and no
forward spillovers to domestic customers (Javorcik 2004, Newman et al. 2015) suggest
that foreign invested firms are successful at preventing technology leakages to their local
competitors as well as down the value chain.
Since precise data on FDI regulation is frequently unavailable, most studies rely on FDI
flows to proxy for reforms in FDI regulation. But as investment flows themselves are
influenced by a large number of different factors, this raises the fundamental problem
of unobserved heterogeneity (Harrison and Rodríguez-Clare 2010). Alternatively, newer
studies rely on aggregated indices of FDI openness (Topalova and Khandelwal 2011, Dug-
gan et al. 2013), which by construction cannot be used to capture differential effects of
regulation across more disaggregated sectors. The use of disaggregated regulation data
should help us to trace the effects of FDI regulation as well as regulatory spillovers at a
much finer grained sectoral scale.
1See Görg and Strobl (2001) and Görg and Greenaway (2004) for surveys of the earlier literature.
2
Indonesia offers a great case to study the effects of FDI on firms (e.g., Blalock and Gertler
2008) as well as the effects of FDI regulation itself. As one of the largest economies in
the world, with a wide variety of industries that rely on an abundance of both human
and natural resources, Indonesia has emerged as an attractive FDI recipient. At the same
time, the Indonesian government has been sending mixed signals to foreign investors,
among others by setting up a blacklist of sectors to be closed or only conditionally open
to FDI (Lindblad 2015). The first negative investment list (NIL) has been released in
2000 in order to protect selected domestic industries from international competition and
foreign acquisitions. It has been repeatedly revised since, with a substantial tightening
of FDI policies in 2007 that was followed by partial deregulation in later years. The
effects of this policy instrument have been hitherto unexplored and offer a particularly
interesting opportunity to investigate the effects of FDI regulation on firm performance at
a highly disaggregated level. The excellent quality of Indonesian firm data, especially as
compared to that from other developing countries (Blalock and Gertler 2008), enables us to
investigate the effects of FDI policies on a large sample of middle and large manufacturing
enterprises in an emerging economy at an unusually high granularity.
This paper exploits variation due to three revisions (in 2007, 2010 and 2014) of the NIL
that regulates sectors at the five-digit sector code level, listing each sector that will be
fully or partially closed to FDI in the future and also specifying whether all firms or only
certain types of firms are to be affected.2 The first revision of 2007 substantially tightened
the existing FDI regulation, increasing the restrictiveness towards FDI in a wide range
of sectors. While the revision of 2010 only introduced minor changes, the restrictions on
inward FDI were substantially relaxed in 2014.
We link the regulatory changes in the NIL at the fine-grained level of five digit sectors
to a firm panel of 15 years (2000 to 2014), derived from the Indonesian yearly census of
manufacturing plants. This census intends to include the full universe of manufacturing
firms with at least 20 employees and measures a wide range of plant-level outcomes.3
Our two main outcome variables are the share of foreign ownership of each firm and
an estimated measure of firm productivity. Our regulation indicators are firm-specific
and vary by year, linking information from the NIL to the firm’s main product (at five-
digit level) while also utilizing individual firm characteristics (firm size, legal status and
prior foreign investment) to identify direct exposure to regulation. In a further step, we
construct regulatory spillover variables that measure the indirect penetration of the NIL
to other potentially non-regulated sectors through the value chain (Bourlès et al. 2013).
2The KBLI (Klasifikasi Baku Lapangan Usaha) sector classification is published by BPS (IndonesianStatistical Office, Badan Pusat Statistik). It is equivalent to the United Nation’s International StandardIndustrial Classification of All Economic Activities (ISIC) at the four-digit level, but it is adjusted tofive-digit level in order to distinguish between additional Indonesian sectors of local importance.
3In what follows, we use the concepts of firm and plant interchangeably as we have no further informationon the structure of multiplant firms.
3
All our regressions are conditional on firm fixed effects and hence only consider within-
firm variation in the main economic outcomes over time. Additionally, our preferred
specifications include two-digit sector-year effects that capture all average time variation
due to global and national industry-specific shocks as well as variation in average input
prices. By that, we only focus on the differences between regulated and unregulated
firms within the same broad economic sector and year. The panel structure of 15 yearly
waves allows us to investigate the time profile of regulation in a more flexible way by also
including lags and leads of regulatory change.
The identification of causal linkages between FDI regulation and firm-level changes in
foreign capital shares and productivity requires two main conditions to be fulfilled: no
spuriously correlated further interventions and no endogenous regulation. As to the first
condition, by the mid-2000s the tariff liberalization of the Indonesian economy was already
mostly over, and although there were some further adjustments in non-tariff barriers to
trade over the decade, their sectoral range was much more limited than the sectoral
coverage of the NIL. By that, we are fairly confident that two-digit industry-year effects
can sufficiently deal with average effects of additional broad regulatory trends as well as
industry-level shocks to market prices.
The second requirement poses a larger challenge as firms operating in the least productive
sectors may be more successful at lobbying for protection against foreign entry. Thus,
a negative correlation between regulation and productivity may arise as a result of the
endogenous lobbying process (in the spirit of Grossman and Helpman 1994). The use
of firm and two-digit industry year-effects reduces these concerns to some extent as they
capture both time invariant differences in firm characteristics and general shifts in the
likelihood of regulation at the broad industry level. After the inclusion of these fixed
effects, we see no evidence of pre-reform trends in productivity differences across regu-
lated and non-regulated firms, which also supports a causal interpretation. Finally, our
preferred specifications control for the potential drivers of endogenous protection (sectoral
concentration of sales, share of blue-collar employment and share of public enterprises)
explicitly, interacting pre-reform sector characteristics at the five-digit level with a full set
of year effects. We also test for a more extensive set of sectoral characteristics as initial
conditions. These control for the residual correlation between industry characteristics that
may drive lobbying success and political decisions as well as changes in firm productivity
over time. Our results remain robust to the inclusion of these controls, which makes us
confident that our results are not driven by reverse causality.
The results document a robust negative relationship between regulation and foreign capital
shares: regulated firms shed more or acquire less foreign capital than their unregulated
counterparts. Firms seem to react already one year before the regulation comes into force,
reflecting the importance of anticipation effects. Furthermore, we find that following the
4
regulatory changes, total factor productivity of regulated firms decreases relative to that
of non-regulated firms operating within the same broad economic sectors. The regulatory
effect on productivity persists if we control for changing foreign capital shares within the
firm. Hence, the productivity declines must be driven by general changes in the decision
environment that go beyond the effects of the measured changes in foreign capital shares.
We also find that these productivity declines are concentrated in the middle of the firm
productivity distribution and among firms with global trade linkages. When comparing
the effects of different types of regulation, we see that sector-wide investment bans lead
to the largest and most significant productivity declines, whereas bans that are linked
to firm-specific conditions or regulations that do not ban FDI but only establish upper
limits on FDI shares do not matter as much for firm productivity. As such regulations may
offer themselves to a more flexible interpretation, they may not constrain firm decisions as
much as a blank prohibition. We also find a weakly significant negative effect of additional
licensing requirements on productivity.
When analyzing productivity spillovers of the NIL along the value chain, we mainly find
backward productivity spillovers from regulated firms to their local suppliers. Thereby, the
use of more disaggregated national input-output tables allows us to distinguish between
spillover effects within and across two-digit sectors. The backward spillover effect across
two-digit sectors is negative and conforms with previous literature that relies on similar
sectoral detail (Javorcik 2004, Newman et al. 2015), showing productivity losses among
supplier industries, potentially due to reduced technology transfer. By contrast, backward
spillovers within the same two-digit sectors are weakly positively linked to productivity. If
industries within the same two-digit sector provide closer substitutes as targets of foreign
investment to each other, regulation may shift foreign investment to and exert competitive
pressure on more closely related unregulated sectors.
Our contribution to the literature is twofold. First, we are among the first to exploit fine
grained variation in the regulatory framework of FDI in an emerging economy.4 By also
taking firm characteristics into account, we identify direct firm exposure to regulation
more precisely and link it to declines in firm productivity. We also investigate effect
heterogeneity of regulation across firms and across different types of rules. Second, we
add to the literature on FDI spillovers by measuring the horizontal and vertical spillover
effects of FDI regulation. Since we use detailed national input-output tables including 87
manufacturing sectors, we are able to carefully disentangle vertical spillovers within and
across two-digit sectors. We thereby show that our results are mainly driven by regulatory
spillovers across but not within two-digit industries.5
4To our knowledge, there is only one other study (Eppinger and Ma 2017) that looks at FDI regulation atthe firm level. Eppinger and Ma (2017) investigate the effect of China’s WTO accession and the relatedmassive liberalization of the FDI regime on changes in the ownership structure of firms. Their resultsshow that changing ownership structures also lead to a boost in output and labor productivity.
5Bourlès et al. (2013) also construct measures of “regulatory burden”. In their cross-country study they
5
This paper proceeds as follows. Section 2 describes the regulatory framework of the
negative investment list in Indonesia and introduces the data sources. Section 3 presents
the estimation strategy and the identification approach. Section 4 presents our results on
the direct effects of the investment reform on foreign capital share and firm productivity
as well as the indirect effects through regulatory spillovers. Section 5 concludes.
2 Regulatory background and measurement
2.1 Foreign investment regulation in Indonesia
Indonesia started to remove the first barriers to foreign direct investment already un-
der the “New Order” regime of President Suharto. The investment coordination board
(BKPM, Badan Koordinasi Penanaman Modal) was installed in 1973 in order to deal with
foreign investment approvals (Gammeltoft and Tarmidi 2013). However, due to its strong
dependence on natural resources, the Indonesian manufacturing sector was only poorly
developed until the early 1980s (Lindblad 2015). Starting in 1983, successful efforts to-
wards industrialization increased the importance of the manufacturing sector and made it
the driving force behind Indonesia’s accelerating growth (Blalock and Gertler 2008). Dur-
ing the 1990s the Indonesian government has changed its previously investment-hostile
regime by opening up the economy to investments from abroad. It quickly became “one
of the most promising host countries [for investment], combining liberal legislation with a
massive endowment of natural resources and a huge and rapidly growing domestic market
for manufactured goods” (Lindblad 2015, p. 225).
The Asian financial crisis of 1997 marks a break in Indonesia’s economic development.
Despite immediate intervention by the International Monetary Fund, the consequences of
the rapidly depreciating Rupiah spread to the real economy. This was accompanied by
social and political instability that destroyed much of the confidence in Indonesia as a
host for investment (WTO 1998). In order to regain the confidence of foreign investors,
steps towards democratization, administrative reform and further trade liberalization were
taken (Duggan et al. 2013). However, Indonesia did not immediately return to economic
growth and foreign investors remained cautious since the business and legal environment
remained rather precarious. Major reforms after 2004 introduced fiscal incentives to
foreign investors, streamlined bureaucratic procedures and prescribed non-discriminatory
treatment for foreign and domestic investors (WTO 2013). In the aftermath of these
reforms, FDI inflows have massively increased again and economic growth has recovered
strongly.
Despite the ongoing liberalization, trade and investment policy in Indonesia remains
especially focus on the impact of upstream sector regulation on productivity in downstream industries.
6
“blurred by contradictory signals” (Lindblad 2015, p. 229). In the close aftermath of
the Asian financial crisis in 2000, the president released the Presidential Decree 96/2000,
at the core of which lies the so-called negative investment list (NIL 2000), naming sec-
tors that are closed or only conditionally open to FDI. Conditions included the need to
form joint ventures between domestic and foreign entities, authorization in certain regions
and licensing requirements. Before 2000, no explicitly formulated version of the NIL was
available. There was a blacklist of sectors closed to foreign investment, but approval pro-
cedures lacked transparency and were completely in the hands of the BKPM. The NIL
2000 was the first to publish regulatory information in a transparent way but still listed
sectors verbally, without recurring to detailed sector codes.
The NIL has been revised for the first time in 2007 and the new list was released with
the Presidential Decree 77/2007. The new version replaced the old vague register from
2000 (Article 7) and listed sectors by their detailed KBLI codes at five-digit level for the
first time. In its trade policy review, the World Trade Organization highlights that a
detailed NIL brings greater transparency with respect to investment and therefore may
be beneficial (WTO 2013). However, closing or conditionally opening certain sectors to
foreign investment is likely to be associated with wasted gains from FDI. In this sense,
the revised version can be considered as a protectionist measure as it adds substantially
more sectors and involves more conditions compared to the NIL 2000. The NIL 2007
comprises manufacturing as well as agriculture and services and introduces standardized
categories of conditions for the first time. According to these conditions, some sectors were
fully closed to foreign investment, in others, FDI was only allowed in small and medium
sized firms, in form of partnerships, up to a certain limit of foreign capital ownership, in
certain locations, or required licensing by the ministry in charge. The next revision of
the NIL came with the Presidential Decree 36/2010, leaving regulation in some sectors
unchanged but also removing some sectors from and adding other sectors to the NIL.
Finally, the latest revisions (with the Presidential Decrees 39/2014 and 44/2016) removed
many sectors from the list of closed sectors, converted bans into licensing requirements,
and clearly decreased the overall extent of regulation. A comprehensive overview of all
revisions of the NIL is provided in Table A1 in the Appendix, including its representation
in the sample and the shares of regulated firms in total manufacturing output.6
6An important characteristic of all versions of the NIL is that regulation is forward looking and doesnot apply to previously approved investments. Thus, the regulation refers to new investment plansand interferes with possible future FDI inflows. See Article 8, Presidential Decree 36/2010, article 5 ofPresidential Decree 77/2007 and article 9 of Presidential Decree 39/2014.
7
2.2 Firm data
The source of firm data is the annual manufacturing census of Indonesia (Survei Industri,
SI) that surveys the universe of all registered Indonesian manufacturing firms with at least
20 employees. The census has been conducted by BPS yearly since 1975 and contains a
rich set of information at the level of manufacturing plants, including the values of inputs
and output, foreign ownership, the value of imports and exports as well as employment
and capital stocks. We follow the literature by using the share of foreign capital as a proxy
of FDI (see e.g., Amiti and Konings 2007, Blalock and Gertler 2008, Arnold and Javorcik
2009, all based on the same SI data). In the SI data, sector codes always refer to the main
product of a firm, which introduces some imprecision as multi-product firms may switch
between sectors every year. We assess the potential extent of this issue by recording all
switches in the main product from year to year. One concern is that regulation creates
incentives to switch the sector to avoid limitations to FDI. We test for the relevance of
this channel in section 4.4.
The data is cleaned for missing values and extreme outliers. As common in the literature,
data points are interpolated between the previous and the next year to avoid the loss of
too many observations, while further missing observations are dropped from the sample
(Amiti and Konings 2007).7 Our final dataset consists of an unbalanced panel of 30,508
firms with a total of 208,120 observations. The sample size decreases further in some
regressions because of missing values in some of the control variables as well as due to the
choice of lag structure: the main regressions estimating the effects of regulation on firm
productivity rely on 185,978 observations. We investigate the robustness of our results to
the choice of lag structure and to using other proxies of productivity for a larger sample
of firms in section 4.4.
We transform all input and output variables to their natural logarithms, using a Box-Cox-
transformation to deal with zeros. This not only allows for a more intuitive interpretation
of coefficients as percent changes but also makes estimations less vulnerable to remaining
outliers. We deflate all monetary values to the base year 2008 by using the yearly wholesale
price index from BPS and comment on the role of price deflators in section 4.4.
There are some concerns regarding data quality in the SI. First, doubts arise with respect
to its completeness since it claims to include all medium sized and large manufacturing
firms in Indonesia. Due to the large number of firms, it is at least possible that SI misses
firms in some years or cannot investigate cases of non-respondents, leading to non-random
selection and undercounting of smaller firms. The presence of financial incentives for the
field agents to register new firms and verify firms that do not reply immediately reduces
this problem as budgets are linked to the number of reported establishments (Blalock and
7Appendix A.1 provides more detailed information on data cleaning procedures.
8
Gertler 2008, Arnold and Javorcik 2009). However, this may also create wrong incentives
by tempting field agents to fill in values of non-reporting firms themselves. A second
issue arises because of potential misreporting by firms. Government law guarantees the
exclusive and anonymized use of information for statistical purposes. Firms may still be
concerned, however, that reported information is leaked to tax authorities or competitors
and may intentionally report wrong data (Blalock and Gertler 2008). Furthermore, if firms
do not put much effort into the correct completion of the questionnaires, numbers may
be falsely reported by accident. Hence, noise within the data is likely to be a considerable
issue. However, as long as firm selection and response behavior is not directly linked to FDI
regulation, firm and two-digit sector-year fixed effects are likely to lead to unbiased within-
estimates. We investigate one specific potential channel of misreporting more explicitly
in section 4.4 by assessing whether regulation makes firms to switch their reported main
product and whether switching firms show different productivity responses to regulation.
2.3 Combined dataset and descriptive trends
We combine the firm data with self-collected information from five revisions of the NIL
by five-digit KBLI sector. While total closure to any investment within a sector unam-
biguously affects all firms, other rules depend on firm characteristics such as firm size,
legal status or location. We thus use the survey information on sales, net assets, legal
status and location to determine firm-specific exposure to regulation. Moreover, since the
preambles of the Presidential Decrees exclude already existing foreign investments from
a new regulation, we offset regulation for those firms that exceed the later legal limits on
the share of foreign capital already before the revision of the NIL.8
Finally, although some of the NIL stipulations verbally narrow regulation to selected
subcategories (e.g., to product features) within a five-digit KBLI sector, we always assume
regulation for the whole five-digit sector. The resulting measurement error is most likely
to cause attenuation bias and thus our results will be underestimated. For a detailed
description of the coding and merging procedures, see Appendix A.2.
We combine the detailed conditions of the NIL with firm level information to generate a
single measure of exposure to regulation. The indicator variable Regulated takes one if
a firm is subject to any kind of regulation in a certain year, taking firm characteristics
relevant for the applicability of the regulation into account. For example, Regulated takes
zero if a medium sized firm operates in a sector that is open to small and medium sized
firms but requires licensing from large firms. For a large firm operating in the same sector,
8The effect of regulation on firm productivity does not crucially depend on this latter adjustment. However,the effect on FDI is much weaker without excluding firms with existing FDI from the regulation. As theforeign capital share in most of the adjusted cases is 100% or close, forward looking regulation does notlimit the direct scope of FDI in such firms, at least in the short run.
9
however, Regulated takes one since FDI is conditional on a successful licensing procedure.9
We further investigate which types of rules affect foreign share and productivity by ex-
plicitly differentiating between different types of regulation (sector-wide and firm-specific
bans and limitations and licensing requirements) in section 4.3.
Table 1 presents summary statistics for the main variables in the years 2000, 2007 and
2014.10 Despite increasing regulation, the overall share of foreign capital increases over
time, although most domestic firms receive zero FDI throughout the whole time period.
Figure 1 shows the more detailed time patterns of regulation and total factor productivity.
The black dashed line shows the average share of regulated firms over time and marks a
strong tightening of regulation in 2007 as the share of regulated firms increased from less
than 5% to more than 20%. In 2014, the scope of regulation declined somewhat, with
a marked shift from firm specific bans towards licensing requirements (cf. Table 1). The
solid lines plot the average development of total factor productivity in our sample of firms
over time. They distinguish between regulated and non-regulated firms in the respective
years and show that on average, firms that operated in sectors subject to FDI restrictions
were slightly more productive than other firms before 2007. In 2007, both regulated
and non-regulated firms faced a negative productivity shock on average, but the drop
in productivity was larger among the regulated firms. In the aftermath, productivity in
both groups recovered again. This evidence is clearly descriptive but already foreshadows
our identified results. The observed negative productivity shock precedes any potential
effects of the global financial crisis, the macro-economic effects of which did not reach
the emerging markets for other two years. It was only in 2009 that Indonesian GDP
experienced a short stagnating period with a subsequent quick recovery and hence the
2007 productivity drop is less likely to be driven by this common global market shock.
Moreover, common market shocks that affected whole industries will be factored out in
our empirical analysis by the use of two-digit sector-year effects.
Table 2 displays the share of regulated firms, the average foreign capital share and total
factor productivity by two-digit sector and year. Regulation across sectors shows a very
heterogeneous picture. Some industries are not affected by the NIL at all whereas other
sectors like wood and wood products were already strictly regulated in 2000. Our analysis
will utilize variation in regulation and productivity across the sub-sectors of the presented
two-digit sectors over time, while controlling for sector-year effects. Thus, we do not
explain changes in FDI penetration or sector-wide changes in productivity, but only focus
9We define firm size according to the Presidential Decree No. 36/2010 that refers to law 20/2008 on smalland medium sized enterprises. A firm is defined as large by this law if its annual sales are higher than50 billion IDR or its net assets (excluding land and buildings) surpass 10 billion IDR. The PresidentialDecree No. 77/2007 refers to an earlier law 9/1995 on small enterprises with very similar definitions oncethe thresholds are adjusted for inflation. We apply this rule yearly, adjusting for inflation.
10Summary statistics for the full sample can be found in Table A2 in the Appendix. Table A3 shows theyearly number of observations in the two largest estimation samples.
10
at the within-firm and within-sector-year differential relationship between FDI regulation
and firm outcomes.
3 Estimation strategy
3.1 Baseline model
We investigate the effect of the foreign investment regulation on firm outcomes by esti-
mating the equation
yijt = αREG ijτ +X′ijtβ + λi + γrt + ψst +Z
′j,2005 × φt + εijt τ = t− 1, t, (1)
where yijt measures the relevant outcomes of firm i operating in the five-digit sector j
in year t. Our two main outcomes measure the percentage of foreign equity in total
firm equity (FDI share) and the log of total factor productivity per firm. REG ijτ is the
investment restriction in sector j in year τ , conditional on the characteristics of firm i.
We test for contemporaneous as well as lagged effects of regulation, τ = t, t − 1, and in
further tests also include up to three lags and leads of regulation at the same time. All
regressions include a vector of controls X ijt to capture time-variant firm characteristics,
such as a set of indicators for firm age categories and a public enterprise indicator (if
more than half of a firm’s capital is owned by the state). We condition our results on firm
fixed effects λi, a set of year effects that vary by macro-region (island) γrt,11 and two-digit
sector-year fixed effects ψst. The residuals εijt are robustly estimated and clustered at the
firm level.
Our extensive fixed effects mitigate issues with unobserved heterogeneity and endogenous
regulation. Firm fixed effects absorb all time invariant unobservable firm characteristics,
including the firms’ average propensity to enjoy protection or be subject to regulation
(Goldberg and Pavcnik 2005). Island-year fixed effects control flexibly for all regional
influence factors that may correlate with both regional exposure to regulation and shifts
in foreign capital shares. The sector-year fixed effects control for time variant incentives
to lobby for protection at the two-digit sector level (Blalock and Gertler 2008). The time
effects also implicitly cancel out common time trends and common macroeconomic or
regulatory shocks to FDI and productivity. Adding further lags and leads of regulation
helps us to better understand the timing patterns of regulatory effects and to look for
anticipation effects or pre-trends.
In our preferred specifications, we capture a further set of determinants of the propensity
to be subject to regulation at the detailed five-digit sectoral level by interacting average
11We distinguish between Sumatra, Java, Kalimantan, Sulawesi and the rest of smaller islands.
11
characteristics of sector j from two years before the major regulatory reform with a full set
of year effects: Z′j,2005 × φt. By doing so, we explicitly control for sectoral characteristics
at the five-digit level that may explain the later success of lobbying behavior. In our
main specifications we interact the sectoral concentration of sales, the share of blue-collar
(production) workers and the share of public enterprises within the sector in 2005 with a
full set of year effects. A higher sector concentration provides firms with better incentives
to pursue co-ordinated action and more power to lobby for protection (Grossman and
Helpman 1994). Sectors with a larger share of low-skilled workers are potentially more
vulnerable and policymakers tend to protect those sectors more (Topalova and Khandelwal
2011). Finally, sectors with a high share of government-owned enterprises may be more
successful at hindering foreign entry (Chari and Gupta 2008).
In further specifications, we also test for a more extended set of initial sector traits in
2005, again interacted with a full set of year effects. These additional characteristics
include employment growth between 2000 and 2005, average capital intensity, average
foreign capital share and import penetration. Employment growth controls for declining
industries which may be treated with special care by policymakers (Grossman and Help-
man 1994). Moreover, high capital intensity, existing FDI presence or strong competitive
pressure from foreign imports have also been argued to influence the political decision for
regulation (Gawande and Krishna 2003).
3.2 Estimating total factor productivity
We estimate total factor productivity (TFP) for each firm, while simultaneously account-
ing for the correlation of the firm’s input choices with the error term (cf. Javorcik 2004,
Amiti and Konings 2007, Newman et al. 2015). If a firm adjusts its choice of inputs to
unobserved productivity shocks, disregarding this adjustment will induce a severe simul-
taneity problem and, thus, lead to biased coefficients.12
The estimation is based on a Cobb-Douglas production function in value added terms on
plant level:
VAit = Yit −Mit = AitLαL
it KαK
it , (2)
where the value added of firm i in year t, VAit, is calculated by subtracting the value
of the intermediate inputs Mit from total firm output Yit. Value added is a function of
productivity Ait, the variable input factor labor Lit and quasi-fixed capital Kit. Taking
natural logs results in:
vait = α0 + αLlit + αKkit + ωit + eit, (3)
12See van Beveren (2012) for a detailed discussion of total factor productivity estimation in the literature.
12
where small letters denote logs. The error term can be decomposed into two components,
an unobserved productivity component ωit and the independently identically distributed
error term eit. Simultaneity bias is introduced because a part of the productivity shocks
is also correlated with the choice of the variable inputs, namely labor and intermediate
goods.
In order to consistently estimate total factor productivity, we apply the approach sug-
gested by Wooldridge (2009). We estimate the log of total factor productivity for each
two-digit sector s separately, taking into account the varying importance of input factors
across industries:
ln(TFP)ist = vaist − α̂0s − α̂l
s list − α̂ks kist, (4)
where αsl and αs
k are the sector-specific input coefficients (see also Table A4).13
A more disaggregated estimation, yielding separate input coefficients on three-digit sector
level, is also feasible. However, as this results in somewhat less stable input coefficients,
we prefer to rely on two-digit input coefficient estimates for the baseline results. We
comment on the role of aggregation in section 4.4. Moreover, estimating TFP over time
may also be sensitive to the choice of price deflators. As detailed sector specific price
deflators are not available for the whole time period, our main results rely on a common
wholesale price deflator. Section 4.4 reports further results that use more detailed sectoral
input and output deflators measured at the five-digit level.
3.3 The effects of regulatory spillovers
As a further step, we investigate the effects of horizontal and vertical regulatory spillovers
that are propagated along the value chain:
yijt = αREG ijt + δ1 Horizontalkt + δ2 Backwardkt + δ3 Forwardkt+
X′ijtβ + λi + γrt + ψst +Z
′j,2005 × φt + εijt τ = t− 1, t.
(5)
Horizontalkt captures the extent of overall regulation within the three-digit sector k in
year t and is measured by the share of protected firms within a sector weighted by each
firm’s median sales Salesi (Javorcik 2004):14
Horizontalkt =
(∑
∀i∈k
REG it × Sales i
)/∑
∀i∈k
Sales i. (6)
13See Appendix A.3 for a more detailed description of the approach and Newman et al. (2015) for usingthe same approach on the two-digit sector level.
14We determine the median level of sales for each firm to make sure that results are not driven by time-varying output. Results are robust to the use of time-varying sales.
13
Horizontal spillovers arise when regulation affects foreign investment or productivity
within a whole sector beyond its direct effects on each regulated firm. Horizontal spillovers
may capture regulatory avoidance behavior that shifts foreign capital to the non-regulated
parts of the three-digit sector, or overall shifts in competitive pressure. Their productivity
effects are ex-ante ambiguous: shifting FDI may increase the productivity of non-regulated
firms, but at the same time, reductions in the overall competitive pressure within the sec-
tor or increasing perceptions of regulatory insecurity may also reduce productivity.
Backwardkt proxies for regulatory penetration in the customer industries v of the three-
digit industry k:
Backwardkt =∑
v 6=k
αkv × Horizontalvt, (7)
where αkv is the proportion of sector k’s output supplied to sector v, taken from the nation-
wide input-output (IO) table of 2005 (BPS 2005). Unlike international IO databases,15
the BPS provides highly disaggregated information that distinguishes 175 sectors in total
(out of which 87 belong to manufacturing), which is broadly comparable to the three-digit
sector level.
Due to this high level of disaggregation, we are able to differentiate between two different
types of backward spillovers separately. We define the variable Intra-sectoral Backwardkt
to capture spillovers to suppliers that occur within the same two-digit industries, and
Inter-sectoral Backwardkt to capture those that affect suppliers that are active in other
two-digit sectors. Due to the relatively low level of disaggregation, other studies usually
refer to inter-sectoral spillovers whenever they identify backward linkages (Javorcik 2004,
Blalock and Gertler 2008, Newman et al. 2015). We expect a negative impact of regulation
on firm productivity in downstream industries if both industries are sufficiently distinct
from each other (inter-sectoral backward linkages). Protection of important customer
sectors may reduce transfers of technology and other non-tangible assets to domestic
upstream firms. This can arise either because of a decrease in foreign presence or because
existing foreign-invested firms abstain from knowledge transfers to local suppliers in the
face of higher uncertainty. However, the mechanisms are not necessarily the same if the
customer and upstream industry operate within the same two-digit sector as in this case
the intra-sectoral backward effect could also pick up spillovers that are usually labeled
as horizontal and reflect avoidance behavior by foreign investors towards non-regulated
sub-sectors or types of firms. The net effect in this case is not clear ex-ante.
Finally, we construct Forwardkt that proxies for regulatory penetration in the supplying
15OECD (2005) or the world input-output database (Timmer et al. 2015).
14
industries w of three-digit industry k (Bourlès et al. 2013):
Forwardkt =∑
w 6=k
σwk × Horizontalwt, (8)
where σwk is the proportion of sector k’s intermediate inputs that it purchases from sec-
tor w. Like before, we disentangle Intra-sectoral Forwardkt and Inter-sectoral Forwardkt
spillovers to account for potentially different linkages across sectors within the same broad
industry and across industries. In all cases, variation within the horizontal and vertical
spillover measures stems from changes in regulation over time. Both the input-output
coefficients and the firm sales are fixed to make sure that we only measure effects that
originate from revisions of the NIL and not from general structural change.
4 Results
4.1 Pre-trends and anticipation effects
A preliminary investigation of time patterns in FDI and productivity around the regula-
tory change indicates some anticipation effects for FDI but no pre-trends in productivity.
This latter finding supports a causal interpretation of the effects of regulation on pro-
ductivity. Figure 2 plots estimated coefficients from fully specified firm level regressions
that include time variant controls, initial sector traits in 2005 interacted with a full set of
year effects, firm fixed effects as well as island-year and two digit sector-year fixed effects.
Moreover, they include the main regulation indicator together with its three lags and
three leads:
yijt =∑t+3
τ=t−3 ατREG ijτ +X′itβ + λi + γrt + ψst +Z
′j,2005 × φt + εijt, (9)
the estimated coefficients α̂τ from which then are plotted in Figure 2 (see also Table A5
for all coefficients and alternative dependent variables).16
Figure 2 shows that foreign capital shares and productivity follow substantially different
time patterns around the regulatory intervention. FDI shares within the regulated firms
as compared to their non-regulated counterparts decline already before the regulation
has taken effect, with a significant reduction in the year before the regulatory tightening,
followed by a further decrease in the year of regulation (see Panel A of Figure 2). FDI
stabilizes in the aftermath of regulation and we even see a short-term rebound effect in
16The inclusion of several lags and leads of regulation reduces the sample size substantially, especially aswe omit the years 2000 and 2001 from the regression. We include leading observations of regulation byrelying on information from the NIL 2016 revision. Table A3 of the Appendix provides an overview ofthe total number of observations by year in our full sample.
15
the year t + 1. This result holds not only for the intensive but also for the extensive
margin of FDI (see column 2 of Table A5).
By contrast, the time effects on total factor productivity show no pre-trends in the three
years before regulation (see Panel B of Figure 2). Regulated and non-regulated firms
are comparable in the run-up to regulatory change in terms of their productivity once
sector-year, island-year and firm fixed effects are factored out. Starting with the year
of regulation, productivity shows a marked drop in the group of regulated firms and
the negative effects persist for two further periods. While the coefficient on regulation
becomes statistically significantly different from zero only two periods after regulation has
been introduced, the Figure shows that the structural break occurs already in t.
Given these time patterns, we decided to follow the literature by linking TFP to regulatory
intervention within the past year, setting τ = t − 1 (see equation 1), but at the same
time to consider a shorter term response of FDI to contemporary regulation with τ = t.
The missing pre-trends in productivity suggest that policy-makers were not implementing
protectionist measures in sectors with declining productivity and thus reverse causality is
unlikely to drive our results. Nonetheless, the significant FDI reaction in the year before
regulation highlights the potential importance of anticipation effects.
Anecdotal evidence indeed suggests that firms already may have anticipated changes in
the regulatory framework before the introduction of the new presidential decrees. For
instance, the largest Indonesian newspaper, Kompas, has already started to cover the
topic in 2005 (June 30), two years before the actual NIL 2007 revision, reporting that
the wheat industry will not enter the new list. Coverage got more pronounced at the
beginning of 2007. On February 8, Kompas announced that the Ministry of Industry
wants sugar refineries to enter the list. When the revision finally took place, Kompas
reported some concerns from business actors who criticized the list because “existing
investment is difficult to develop even though the NIL is not retroactive” (July 16, 2007,
p. 18). Similarly, Kompas already started to report on plans to revise the NIL at the
beginning of 2013 while the Presidential Decree was only released in April 2014. In
February 2013, Kompas quoted the head of the Investment coordination board, M. Chatib
Basri, saying that “the main goal is to improve national competitiveness and to be more
investor friendly” while “there are still sectors that must be protected” (February 18, 2013,
p. 20). Another article reported on plans for relaxation of investment in the alcoholic
beverage industry on July 12, 2013. By the end of 2013, news coverage of the topic
increased substantially. For example, Reuters reported on the “ease of regulation to allow
foreign companies [...] to manage and operate airports” (November 20, 2013). Around the
same time, Kompas published a letter to the editor in which a concerned reader named
further sectors where access to foreigners is planned, among others also pharmaceuticals
(November 28, 2013).
16
Even though this evidence is only anecdotal, the fact that newspapers openly discussed
the revisions of the NIL more than one year ahead shows its relevance for the Indone-
sian economy. We further believe that industries and firms have been aware of more
detailed plans of the revisions even before they entered public media, which can explain
the observed anticipation effects and sinking foreign investment shares in selected sectors.
4.2 Baseline results
Results in Table 3 show a significant decline in FDI within regulated firms. The first five
columns report the baseline effects of regulation on FDI shares within the firms, using
equation (1) and linking FDI shares to contemporary regulation. As FDI has responded
immediately to regulation (and has even shown anticipation effects, see above), we focus
here on the immediate response that started in the year of regulation. The first column
in Table 3 includes only firm fixed effects and island-year effects whereas further columns
consecutively extend this specification, without changing the magnitude or statistical
significance of the estimated effect. Column 2 adds categories of firm age and an indicator
for state enterprises as time variant controls. Column 3 extends the model by two-digit
sector-year interactions whereas column 4 and 5 interact five-digit sector characteristics
in the pre-reform year 2005 with a full set of year effects. While column 4 uses a more
limited set of sector characteristics, including sector concentration, the share of blue-collar
workers and the share of public enterprises, column 5 extends this list even further by
adding prior employment growth, capital intensity, the share of foreign capital and import
penetration at the five-digit level.
Throughout all specifications, the estimated impact of the regulation indicator on foreign
capital share is highly significant and implies that becoming protected is associated with
a 0.8 percentage points lower foreign equity ownership share on average. This effect may
not seem very substantial but it still amounts to about 10% of the mean ownership within
the sample (which is about 7%). Column 6 shows that our result is not driven by the
intensive margin of firm ownership only. When we substitute the dependent variable by
an indicator for a multinational enterprise (MNE, defined as a firm with at least 10%
foreign ownership), the results show that regulation also reduces the likelihood of a firm
having considerable foreign capital (at the extensive margin). Overall, the coefficients
stay remarkably stable when further fixed effects and initial industry characteristic times
year fixed effects are additionally controlled for, indicating that endogenous regulation is
unlikely to drive these results.
At the same time, regulation is also linked to a statistically significant decline in total
factor productivity (see Table 4). Unlike before, we use the first lag of the regulatory
indicator in our main specifications as productivity is usually expected to adjust more
17
slowly upon regulation (and Figure 2 has shown no anticipation effects).17 The first five
columns of Table 4 report the baseline effects of regulation on TFP (based on equation
1). As before, the table starts with firm and island-year effects only and adds further
controls step-by-step. Unlike in the case of FDI shares, controls do make a difference for
our TFP estimates. Although the productivity effects of regulation have a negative sign
even without any controls, they only gain significance once two-digit sector-year effects are
included in the regressions. This makes sense as sector-year effects are often considered
as crucial controls for productivity estimates (Goldberg and Pavcnik 2005), especially as
they also pick up industry-specific variation in the price of intermediates, which will affect
TFP estimates substantially.18
Adding time effects interacted with initial five-digit sector characteristics to the regressions
in columns 4 and 5 does not change the point estimates, which shows that the changes in
productivity are not driven by characteristics of the finely grained sectors that may make
regulation more likely. The point estimate on the regulation coefficient implies that a firm
that has become regulated in the previous year experiences a 3.4 to 3.6 percent reduction
of its TFP ceteris paribus. Column 6 checks the sensitivity of our results to the specific
TFP estimation. It takes the log of value added per worker as an alternative dependent
variable which is often used as proxy for labor productivity in the literature (cf. Amiti
and Konings 2007). The coefficient of regulation barely changes and remains significant
at the five percent level.19
Our baseline TFP regressions measure the full regulatory effect, without controlling for
changes in FDI shares directly. However, alternative specifications that include the foreign
capital share as additional control result in virtually the same regulatory coefficient (see
Table A9 in the Appendix). While we also find that TFP is positively related to foreign
ownership shares (though without reaching conventional levels of statistical significance),
the regulatory effects do not seem to be driven by immediate drops in the firms’ FDI
shares and may rather reflect changing patterns of technological upgrading or changing
expectations with respect to competitive pressure that lead to adjustments in factor use.
17Table A6 shows that our results are robust to substituting contemporaneous for lagged regulation. Seealso section 4.4.
18Columns 7 to 9 of Table A7 in the Appendix show that when using five-digit sector-specific input andwholesale price deflators and estimating TFP at the three-digit level, regulation significantly reducesTFP even without the inclusion of sector-year effects. As more detailed price deflators are not availablefor the full time period and have to be partially imputed, while the three-digit TFP estimates turn outgenerally less robust, we prefer to rely on the aggregated wholesale price index, two-digit sector basedTFP estimates and two-digit sector-year effects in our main specifications. See also section 4.4.
19Alternative measures of productivity, such as total value added, value added per worker and value addedper capital yield overall comparable results (see Table A8 in the Appendix).
18
4.3 Heterogeneities in regulation and across firms
Our general regulation indicator combines a range of different provisions, not all of them
equally restrictive. In order to better understand which types of regulatory interven-
tions are more likely to affect FDI and productivity, Table 5 differentiates between major
types of regulation by contrasting licensing requirements with the more direct bans and
limitations on FDI. Licensing requirements leave the affected sectors open to FDI but
potentially increase the compliance burden by introducing costly and time-consuming
procedures. However, as licensing requirements have been more extensively used in the
later years (as part of the easing of regulation in 2010 and later), their introduction may
have also been seen as guaranteeing openness and signaling that FDI will still be possible
in these sectors. By contrast, the diverse bans and limitations on FDI are less ambiguous
in their effect as they are more likely to restrict FDI flows. We further disentangle these
latter rules into three groups: 1. limitations that introduce upper limits on FDI shares,
2. sector-wide bans that prohibit FDI in the given sectors irrespectively of firm character-
istics, and 3. firm-specific bans that restrict FDI conditional on firm characteristics like
firm size or legal status.20 Table 5 shows that licensing requirements do not curb for-
eign investment whereas bans and limitations generally tend to reduce FDI (column 1).
However, both types of regulation are similarly costly in productivity terms (column 4).
Further disentangling the restrictions by their type is somewhat limited by power issues,
but the results show that while all bans and limitations tend to reduce FDI (columns 2
and 3), in terms of productivity losses, the closing of full sectors to FDI seems to be the
most detrimental (column 6), suggesting the importance of general reductions in perceived
competitive pressure.
In Table 6, we investigate which types of firms were the most affected by regulation and
distinguish between firms based on their foreign market linkages, their place in the rela-
tive productivity distribution, and their size. Firms with linkages to international markets
may be more directly (and hence also more strongly) affected by protectionist measures.
We thus check for differential effects of regulation among those firms that have engaged
in either exports or imports, or had an above ten percent foreign capital participation
at least in one year of our sample period.21 The results show that FDI decreased upon
regulation not only in firms with international linkages but also on average (column 1).
The regulatory effects on FDI did not differ in firms that engage in foreign trade, but
were naturally concentrated among MNEs (defined as firms with above 10% of foreign
ownership at any point over the whole time period). The interaction coefficient on foreign
ownership is highly significant and shows a total decline of about 4.1 percentage points
20See Appendix A.2 for a detailed overview of the types of regulation and how categories are aggregated.21There are 12,691 firms with some import or export activity and 3,379 firms with foreign participation.
2,907 firms show both trade and FDI linkages.
19
in FDI shares in MNEs. This reflects a nearly-mechanical scale effect as the scope of
adjustment is highest among such firms by definition. By contrast, the decomposition of
productivity effects by foreign market exposure does not yield significant coefficients (col-
umn 4), although the total effect for domestic trading firms is substantial (−5.6 percent)
and tests significant at the 1 percent level. These two sets of results show that regula-
tion did not act directly by reducing productivity in the MNEs only, rather, it especially
affected domestic firms with global trade linkages.
As a second test of across-firm heterogeneities, we split firms into three groups according
to their position within the productivity distribution. The low (high) productivity group
includes those firms that have been in the lowest (highest) decile of the two-digit sectoral
TFP distribution in at least one year. The group of medium productivity comprises
firms that have never been in the lowest or highest decile of the sectoral productivity
distributions, or, in some cases, appeared in both the upper and lower decile, mostly
due to sector switches across two-digit industries.22 The results show that regulation
reduced FDI shares in firms of all three productivity categories significantly (column 2),
but high productivity firms suffered substantially larger declines in FDI (of 1.5 instead of
0.3 percentage points on average). The subsequent productivity losses have been especially
concentrated in medium productivity firms. Finally, we distinguish between medium-sized
and large enterprises following Indonesian legal definitions.23 We find no statistically
significant differences in FDI or productivity effects of regulation by firm size.
4.4 Further robustness issues
The baseline models fixed the time structure of the regulatory response by considering
contemporaneous regulation for FDI and lagged regulation for TFP. Figure 2 and Table A5
depict a more complex timing structure of regulatory effects, but the lag structure comes
at the cost of using a substantially shorter time frame and showing reduced significance
of the yearly effects of regulation on TFP. As an alternative test of time patterns around
the regulatory intervention, Table A6 repeats the baseline regressions for FDI and TFP
by shifting the time lag of the regulatory treatment by a year in each direction. Unlike
the results based on three lags and leads (see Figure 2 and Table A5), these shifts in the
lag-structure reduce the sample by one year only. We report the results pairwise using
regulation in t and t − 1 (or t and t + 1), relying on the same sample for each pair. For
FDI, there is hardly any difference between the coefficient of regulation in t and t+1, with
22According to this definition, there are 7,960 firms in the lowest category (with an average foreign capitalshare of 1.3%), 15,774 firms in the mid category (with an average foreign capital share of 4.6%) and 7,291firms are identified as highly productive (with an average foreign participation of 15.8%).
23We distinguish between medium-sized and large enterprises according to the definition in law law 20/2008on micro, small and medium enterprises (see also Appendix A.2). There are 29,236 (1,789) medium-sized(large) firms in our sample with an average foreign capital share of 4.9% (29.3%).
20
both of them being significant, whereas regulation in t − 1 does not reach conventional
levels of significance. By contrast, while there is no evidence in favor of any anticipation
effects on TFP, the lagged coefficient comes out somewhat larger in magnitude and more
significant than the contemporaneous effect.
Although our results are identified within the same firms and hence are less likely to be
driven by shifts of firm composition, it is still useful to understand whether the compo-
sition of firms endogenously adjusts in response to the revision of the NIL. Protection
of a sector may increase the incentives for new firms to enter the market or reduce the
exit rate of firms by allowing non-competitive firms to stay in the market. Conversely,
regulation may also impact firms negatively, forcing them to leave the market or keeping
new entrants out of the market. It is not clear in which direction the effect will go ex
ante, but the resulting shifts in firm composition may affect average firm productivity.
Column 1 of Table 7 documents that regulation within a particular sector indeed reduces
the probability of market entry by new firms in the same period. Furthermore, regulation
also increases the likelihood of market exit in the next period. However, the nature of the
data does not allow for a closer analysis of market entry and exit dynamics as the exact
year of entry or exit may be mismeasured due to firm (non-)response behavior. We thus
concentrate instead on whether there is a differential response to regulation among those
firms that either enter or exit the market over the sample period. Columns 3 and 5 show
that new entrants do not differentially respond to regulation in terms of their FDI or TFP:
while interaction coefficients are positive, they are not significant. For firms exiting over
the sample period, the interaction coefficients with both FDI and TFP become positive
and significant in column 4 and 6. Taken together, these results show that entering and
exiting firms do not react more negatively (and potentially react somewhat more weakly)
to regulation. This makes it unlikely that even the average results would be driven by
sample selection issues.
A different, and potentially more serious, concern is that firms will endogenously decide on
whether they want to operate in a regulated sector or switch to a non-regulated product.
Such sector switching may bias our estimates, but the direction of the bias is not clear ex-
ante. It is both possible that firms select into newly protected sectors or choose to operate
in non-regulated sectors. Moreover, as SI firms (more accurately, plants) may produce
multiple products but only report their main product, regulation may also simply lead to
firms reporting to belong to a non-regulated sector as a form of avoidance behavior. This
second channel is unlikely to play a central role though as the firm census is not used by
the authorities to explicitly monitor firms (Blalock and Gertler 2008).
Table 8 addresses the product switching behavior by firms. Our dependent variable in
the first two columns is an indicator for a sector switch that takes one if a firm changes
its reported five-digit sector code in year t as compared to its previously observed sector.
21
Column 1 shows no evidence for a sector switch occurring in year t as a response to
contemporaneous regulation (in year t), hence firms do not switch into protected sectors.
Column 2 looks at the response to regulation in year t − 1 instead, testing for whether
firms actively select out of regulated sectors, yielding again an insignificant coefficient.
As a next step, we test the differential effects of regulation on FDI and TFP among
switching firms. We therefore distinguish between four different transition types: firms
switching from non-regulated to regulated, from regulated to non-regulated and within
the regulated or non-regulated sectors. Columns 3 and 4 of Table 8 find that, beyond the
robust negative effect of regulation on FDI, no form of sector switching is significantly
associated with a decline in FDI. Column 5 shows more pronounced differences for TFP:
firms that have recently switched sectors experience a drop in TFP by 1.9 percent in the
next period. This is intuitively plausible as switches may require changes in the production
process that come at the cost of initial productivity losses. As before, the direction of
the switch matters (see column 6). While the average regulatory effect remains robust,
firms switching into or out of regulated sectors do not experience changes in productivity,
only for firms switching within non-regulated sectors turns the coefficient to significantly
negative. This makes it clear that it is very unlikely that switching behavior across sectors
would drive our findings.
Due to a large number of missing observations on capital, our sample size shrinks sub-
stantially during the data cleaning process (see Appendix A.1 for more detail). Table A10
performs an important robustness check showing that our findings also hold when using
the full sample size, also including firms with missing capital observations. Naturally, we
cannot use TFP as a measure for productivity in this fuller sample as capital is a key
ingredient into the productivity estimates. But value added per worker (see also Table
4) can be used as an alternative productivity measure for the larger sample without any
further limitations. The effect of regulation on FDI is robust across specifications and
even larger in magnitude as compared to Table 3. Becoming subject to the NIL decreases
the share of foreign capital by about 1 percentage point in this larger sample. Just like
for the TFP results in Table 4, the effect of regulation on valued added per worker only
becomes significant when controlling for sector-year interactions in column 5. The esti-
mated coefficient turns out slightly smaller in magnitude as compared to the results in
column 6 of Table 4, but overall the baseline results can be fully confirmed.
As a last robustness check, Table A7 assesses the sensitivity of baseline results to our
TFP-estimation procedures. The first three columns of the table repeat the baseline
TFP-estimates, calculated at the two-digit level with and without sector-year fixed effects
and pre-reform sector traits and year interactions. As a comparison, the further columns
report results that are based on a more disaggregated TFP estimation. Here the TFP
regressions are separately estimated for each three-digit sector, even though some sectors
22
have to be combined because of insufficient observations. This is especially the case in
high technology sectors such as railways, aircraft and ships or optical medical instruments.
Columns 4 to 6 of Table A7 fully replicate our main results. Columns 7 to 9 of Table A7
add further detail to the three-digit TFP estimates by exchanging the common wholesale
price deflators used for the baseline results with five-digit sector specific price deflators.
These include a five-digit wholesale price index used to deflate firm sales, a five-digit
input price index used to deflate intermediate inputs, and a machinery price index, used
to deflate the capital stock and net assets (for identifying large firms).
The results show that our preferred specification yields almost the same regulatory effects
on TFP irrespective of the sectoral aggregation for TFP estimation and sectoral detail in
price deflators. Moreover, Table A7 also yields a further insight. In our baseline model, the
inclusion of sector-year effects is needed in order to detect a significant effect of regulation
on TFP. By contrast, using more detailed sectoral price deflators together with more
detailed TFP estimates renders the regulation coefficient significant even before state-
year effects are included (see column 7 of Table A7). This suggests that sector specific
variation in input and output prices over time makes the inclusion of sector-year effects
necessary when using two-digit TFP estimates. At the same time, two-digit sector-year
effects also seem sufficient for dealing with the issue of sectoral price and productivity
variation at a finer scale.
All in all, we prefer to use the two-digit TFP estimates (together with the aggregated
wholesale price deflator) as our baseline as the higher level of detail in TFP sectors and
price deflators comes at the cost of a loss in precision. Due to the relatively lower number
of firms operating in some three-digit sectors, input coefficients estimated at the three-
digit level tend to be more unstable and some of them even turn negative, which does not
happen at the two-digit level (cf. Table A4). Moreover, five-digit sector-specific output
and input price indices as well as the machinery price index are only available to us until
2012 and have to be imputed for the following years by assuming proportionate sectoral
price variation. As our fully specified results do not change when using more detailed
TFP estimates or more detailed price deflators, we interpret this as a support of our more
aggregated TFP estimation approach.
4.5 Regulatory spillover effects
The effects of regulation are very likely to be transmitted along the value chain, leading
not only to a significant productivity decline in regulated firms, but also affecting other
non-regulated firms. Regulatory spillovers could potentially help to explain at least a
part of the substantial productivity drop observed among non-regulated firms (cf. Figure
1). Classical FDI spillovers are expected to mainly act through the channel of technology
23
diffusion and hence be less frequent horizontally among competing firms (Blalock and
Gertler 2008, Newman et al. 2015) but rather take place vertically and especially be
directed from buyers to their suppliers as backward spillovers (Javorcik 2004, Barrios
et al. 2011). By contrast, we expect FDI regulation to generate not only vertical but also
horizontal spillovers. When regulated firms face harder constraints with respect to foreign
investment (up to a complete ban), FDI may shift to similar non-regulated firms or sub-
sectors within the same sector, reducing the productivity of regulated and increasing the
productivity of non-regulated competing firms. At the same time, a second, countervailing
force may also be at play as with a larger penetration of regulation within any sector, a
decrease in average productivity may reduce the overall competitive pressure, dampening
the productivity even of the non-regulated firms (Caves 1974, Kokko et al. 1996). Which
of the horizontal forces outweighs is a priori unclear. When considering vertical spillovers,
the use of detailed input-output tables allows the distinction between vertical spillovers
that take place across and within two-digit sectors. While we expect vertical regulatory
spillovers across two-digit sectors to conform with the usual findings on FDI spillovers,
the vertical spillovers within two-digit sectors may still be closer to horizontal spillovers
in their nature.
Table 9 shows the results of regulation and its spillovers on foreign capital share and pro-
ductivity, measuring the underlying input-output linkages on three-digit level. Columns
1 and 4 introduce the general measures of horizontal, backward and forward regulatory
spillovers. Columns 2 and 5 split each type of vertical spillovers into those working within
the same two-digit sector and those working across two-digit sectors. Finally, columns
3 and 6 test alternative dependent variables. Over all specifications, the direct effects
of regulation remain robust and significantly negative: regulation directly reduces the
foreign capital share by 1 percentage point and productivity by up to 3.8 percent. Hori-
zontal spillovers within the same three-digit sector turn out to significantly increase FDI
but not to affect productivity. A one standard deviation larger increase of horizontal
penetration of regulation increases FDI shares by about a quarter of a percentage point
(0.23×−0.011 = −0.0025), potentially reflecting FDI shifting out of regulated firms into
other firms within the same three-digit sector. Nonetheless, the horizontal productivity
spillovers do not turn out significant. This is not surprising as we expect productivity to
increase due to FDI shifts and to decline due to reducing competition, and hence these
two effects may well cancel out.
Vertical regulatory spillovers do not have a strong effect on FDI shares or productivity on
average (see columns 1 and 4). Backward spillovers from regulated buyers do not seem
to affect foreign capital shares, neither productivity. Forward spillovers from regulated
suppliers seem to slightly decrease foreign capital shares but the effect is of negligible
size: a one standard deviation larger penetration in forward spillovers reduces FDI by 0.1
24
percentage points (0.08×−0.017 = −0.001).
However, splitting up vertical spillovers into those that act within and across two-digit sec-
tor reveals a somewhat more nuanced picture, especially in terms of TFP. For FDI shares,
only one of the four vertical spillover measures is significant (increasing FDI shares due
to backward spillovers across two-digit sectoral borders) but it does not turn out con-
sistently significant across the intensive and extensive margin (columns 2 and 3) and
lies below 0.1 percentage points per standard deviation (0.01 × 0.089 = 0.0009). For
spillovers in TFP, the coefficient of regulatory intra-sectoral backward spillovers remains
insignificantly positive in column 5. However, inter-sectoral backward spillovers turn out
marginally significant and of a substantial and economically relevant size: a one stan-
dard deviation larger regulatory penetration in customer industries reduces productivity
across two-digit sectoral lines by 1 percent (0.02 × −0.513 = −0.01). Compared with
the direct productivity effect of regulation of about 3.5%, a 1% productivity decrease is
still sizable. Using an alternative proxy of labor productivity in column 6 shows practi-
cally the same negative backward productivity spillovers across two-digit industries and
marginally significant positive productivity spillovers within two-digit industries. The
positive within-two-digit sector spillovers show a more modest effect of about 0.4% per
standard deviation (0.03 × 0.140 = 0.04). Regulatory forward linkages are smaller in
magnitude and do not reach conventional levels of significance.
Our results demonstrate that the level of aggregation matters when looking at linkages
along the value chain. Our highly disaggregated input-output table allows us to identify
vertical linkages that would have been coded as horizontal spillovers on less granular level.
We show that intra-sectoral spillovers behave differently from inter-sectoral spillovers and
that the TFP spillovers in column 4 mask a composition effect of two opposing mecha-
nisms. For example, take two vertically linked firms operating in the food and beverages
industry (two-digit) where one of them is protected by the NIL while the other not. A
foreign investor who initially was interested in investing into the protected firm may easily
adjust her investment plans towards the non-regulated firm. In that sense, regulation may
put competitive pressure on other sectors within the same two-digit industry due to the
easier substitution of investment, leading to increased productivity in the non-regulated
firms. The same mechanism may not be relevant for across industry linkages as an in-
vestor cannot easily switch its investment target from a firm in the food and beverages
sector to an IT enterprise because of missing know-how and expertise. Thus, the competi-
tion channel is muted. The dominating vertical spillovers of regulation work through less
technology and managerial skill transfers, lower quality of intermediate inputs, or lower
demand for and supply of domestic products.
25
5 Conclusion
This paper contributes to the literature on the effects of sector specific regulation of
foreign investment. Despite an increasingly open FDI regime, the Indonesian government
uses the instrument of negative investment lists that restrict future foreign investment in
particular industries. For our analysis, we especially exploit the revisions of the NIL in
2007, 2010 and 2014. Our identification strategy is based on an extensive set of fixed effects
at the firm, region-year and two-digit sector-year level. Moreover, we allow for yearly
productivity changes being proportionate to a set of five-digit industry characteristics
in order to control for classical political economy factors that could drive industry-level
variation in regulatory action. An examination of the time pattern of productivity changes
helps to exclude that the effects of regulation merely reflect differences in pre-trends.
We find robust evidence that shows a substantial effect of FDI restrictions on foreign
ownership shares within the affected firms, once time variant controls and an extensive
set of fixed effects are controlled for. Analyzing the relationship between regulation by the
NIL and firm level productivity, we find evidence in favor of a causal negative influence
of investment regulation on TFP and value added per worker. FDI restrictions are asso-
ciated with an average decrease of total factor productivity of about 3.5 percent in the
year following the regulatory change. The productivity declines cannot be mechanically
explained by a drop in foreign capital shares. Instead, we see productivity declines to
be concentrated among firms in the middle of the productivity distribution and among
firms with global trade linkages. From the different types of regulation, blanket sector
level bans turn out to reduce productivity the most but licensing requirements are also
negatively linked to productivity. We also see some evidence that regulation indirectly
affects non-protected firms through the value chain. Using a disaggregated input-output
table, we show that regulatory backward spillovers across the two-digit sectors have a
sizable productivity reducing effect. By contrast, backward spillovers across firms op-
erating within the same two digit sector turn out weakly positive, potentially reflecting
countervailing effects of competitive pressure.
Our empirical results raise substantial doubts about the economic benefits of a protec-
tionist policy by showing that regulatory tightening of the FDI regime in Indonesia had
on average detrimental effects on firm productivity. Naturally, policy makers do follow
multidimensional objectives when deciding on such regulatory interventions. The NIL
may have brought short-run political benefits domestically but also may have affected
other economic outcomes beyond productivity more favorably. Investigating its effects
from a broader perspective would complement our study and offer an interesting avenue
for further research.
26
References
Aitken, B. J. and Harrison, A. E. (1999). Do domestic firms benefit from direct foreign
investment? Evidence from Venezuela. American Economic Review, 89(3):605–618.
Amiti, M. and Konings, J. (2007). Trade liberalization, intermediate inputs, and produc-
tivity: Evidence from Indonesia. American Economic Review, 97(5):1611–1638.
Arnold, J. M. and Javorcik, B. S. (2009). Gifted kids or pushy parents? Foreign direct
investment and plant productivity in Indonesia. Journal of International Economics,
79(1):42–53.
Barrios, S., Görg, H., and Strobl, E. (2011). Spillovers through backward linkages from
multinationals: Measurement matters! European Economic Review, 55:862–875.
Blalock, G. and Gertler, P. J. (2008). Welfare gains from foreign direct investment through
technology transfer to local suppliers. Journal of International Economics, 74(2):402–
421.
Bourlès, R., Cette, G., Lopez, J., Mairesse, J., and Nicoletti, G. (2013). Do product
market regulations in upstream sectors curb productivity growth? Panel data evidence
for OECD countries. Review of Economics and Statistics, 95(5):1750–1768.
BPS (2005). 2005 Indonesian input output table: Volume I-III. Technical report, Statistics
Indonesia, Jakarta.
Caves, R. E. (1974). Multinational Firms, Competition, and Productivity in Host-Country
Markets. Economica, 41(162):176–193.
Chari, A. and Gupta, N. (2008). Incumbents and protectionism: The political economy
of foreign entry liberalization. Journal of Financial Economics, 88(3):633–656.
CompNet Task Force (2014). Micro-based evidence of EU competitiveness: The CompNet
Database. ECB Working Paper Series 1634, European Central Bank, Frankfurt.
Djankov, S. and Hoekman, B. (2000). Foreign Investment and Productivity Growth in
Czech Enterprises. World Bank Economic Review, 14(1):49–64.
Duggan, V., Rahardja, S., and Varela, G. (2013). Service sector reform and manufacturing
productivity: Evidence from Indonesia. Policy Research Working Paper 6349, World
Bank, Washington, DC.
Eppinger, P. and Ma, H. (2017). Optimal ownership and firm performance: Theory and
evidence from China’s FDI liberalization. Job market paper, University of Tübingen,
Tübingen.
27
Gammeltoft, P. and Tarmidi, L. T. (2013). Chinese foreign direct investment in Indonesia:
Trends, drivers and impacts. International Journal of Technological Learning, Innova-
tion and Development, 6(1,2):136–160.
Gawande, K. and Krishna, P. (2003). The political economy of trade policy: Empir-
ical approaches. In Choi, E. K., editor, Handbook of International Trade, Blackwell
Handbooks in Economics, pages 212–250. Blackwell, Malden, Mass. [u.a.].
Goldberg, P. K. and Pavcnik, N. (2005). Trade, wages, and the political economy of
trade protection: Evidence from the Colombian trade reforms. Journal of International
Economics, 66(1):75–105.
Görg, H. and Greenaway, D. (2004). Much ado about nothing? Do domestic firms really
benefit from foreign direct investment? The World Bank Research Observer, 19(2):171–
197.
Görg, H. and Strobl, E. (2001). Multinational companies and productivity spillovers: A
Meta-analysis. Economic Journal, 111(475):F723–39.
Grossman, G. M. and Helpman, E. (1994). Protection for sale. American Economic
Review, 84(4):833–850.
Harding, T. and Javorcik, B. S. (2011). Roll out the red carpet and they will come:
Investment promotion and FDI inflows. Economic Journal, 121(557):1445–1476.
Harrison, A. and Rodríguez-Clare, A. (2010). Trade, foreign investment, and industrial
policy for developing countries. Handbook of Development Economics, 5:4039–4214.
Javorcik, B. S. (2004). Does foreign direct investment increase the productivity of domestic
firms? In search of spillovers through backward linkages. American Economic Review,
94(2):605–627.
Kokko, A., Tansini, R., and Zejan, M. C. (1996). Local technological capability and
productivity spillovers from FDI in the Uruguayan manufacturing sector. Journal of
Development Studies, 32(4):602–611.
Levinsohn, J. and Petrin, A. (2003). Estimating production functions using inputs to
control for unobservables. Review of Economic Studies, 70(2):317–341.
Lindblad, J. T. (2015). Foreign direct investment in Indonesia: Fifty years of discourse.
Bulletin of Indonesian Economic Studies, 51(2):217–237.
Newman, C., Rand, J., Talbot, T., and Tarp, F. (2015). Technology transfers, foreign
investment and productivity spillovers. European Economic Review, 76:168–187.
28
OECD (2005). OECD STAN: Bilateral trade database. Technical report, Organization
for economic co-operation and development, Paris.
Olley, G. S. and Pakes, A. (1996). The dynamics of productivity in the telecommunication
equipment industry. Econometrica, 64(6):1263–1297.
Timmer, M. P., Dietzenbacher, E., Los, B., Stehrer, R., and de Vries, G. J. (2015).
An illustrated user guide to the World Input-Output Database: The case of global
automotive production. Review of International Economics, 23(3):575–605.
Topalova, P. and Khandelwal, A. (2011). Trade liberalization and firm productivity: The
case of India. Review of Economics and Statistics, 93(3):995–1009.
van Beveren, I. (2012). Total factor productivity estimation: A practical review. Journal
of Economic Surveys, 26(1):98–128.
Wooldridge, J. M. (2009). On estimating firm-level production functions using proxy
variables to control for unobservables. Economic Letters, 104(3):112–114.
WTO (1998). Trade policy review Indonesia. TPR, World Trade Organization, Geneva.
WTO (2013). Trade policy review Indonesia. TPR, World Trade Organization, Geneva.
29
Figures
Figure 1: Average regulation and total factor productivity
���
���
���
����
�
���
����
�����
����
�����
���
���
����
���
����
�����
���
���� ���� ���� ��������
������������� ��������� ������������������������
Note: The graph plots the share of protected firms over the sample period (right scale), as well as the average log oftotal factor productivity among regulated and non-regulated firms in the respective year (left scale).
30
Figure 2: Timing effect on FDI and ln(TFP)
����
�����
���
����
�
��� ��� ��� � ��� ��� ���
����������������������������������������������������������������������
����
����
����
���
���
�
��� ��� ��� � ��� ��� ���
��������������������������������������������������������������������������
Note: Plotted coefficients are estimated controlling for categories of firm age, a public enterprise indicator and inter-actions of five-digit sector traits with year dummies, as well as firm, island-year and industry-year FEs. Bars aroundthe point estimates denote ninety percent confidence intervals.
31
Tables
Table 1: Summary statistics in 2000, 2007 and 2014
2000 2007 2014
Mean SD Mean SD Mean SD
Regulated 0.03 0.18 0.22 0.42 0.15 0.36Licensing requirements 0.01 0.10 0.03 0.16 0.07 0.25FDI limitations + bans 0.02 0.15 0.22 0.41 0.12 0.33Sector-wide limits 0.00 0.00 0.01 0.11 0.03 0.18FDI bans 0.02 0.15 0.21 0.41 0.10 0.30Sector-wide bans 0.02 0.14 0.01 0.10 0.01 0.11Firm-specific bans 0.00 0.05 0.20 0.40 0.09 0.28
Reg. horizontal 0.04 0.15 0.19 0.29 0.19 0.25Reg. backward 0.01 0.02 0.06 0.17 0.07 0.14
Reg. backward (within 2 digit) 0.01 0.02 0.05 0.17 0.06 0.14Reg. backward (across 2 digit) 0.00 0.01 0.01 0.02 0.01 0.03
Reg. forward 0.04 0.08 0.04 0.06 0.06 0.07Reg. forward (within 2 digit) 0.01 0.05 0.03 0.06 0.04 0.07Reg. forward (across 2 digit) 0.03 0.07 0.02 0.02 0.02 0.03
FDI share 0.05 0.21 0.06 0.22 0.08 0.26MNE 0.07 0.25 0.06 0.25 0.09 0.29ln(TFP) 10.13 1.57 10.05 1.53 10.97 1.60ln(VAD/L) 9.91 1.24 9.88 1.23 10.74 1.28ln(K) 14.29 2.07 13.79 2.07 14.25 2.12ln(L) 4.21 1.16 4.03 1.06 4.18 1.16Firm age below 5 years 0.15 0.36 0.16 0.37 0.21 0.41Firm age between 5-15 years 0.45 0.50 0.39 0.49 0.20 0.40Firm age between 15-25 years 0.24 0.43 0.27 0.44 0.30 0.46Firm age above 25 years 0.15 0.36 0.19 0.39 0.29 0.45Switch in t 0.16 0.37 0.19 0.40 0.12 0.33
Switch into regulated 0.00 0.07 0.03 0.16 0.01 0.11Switch into nonregulated 0.00 0.00 0.01 0.09 0.01 0.11Switch within regulated 0.00 0.00 0.00 0.05 0.01 0.08Switch within nonregulated 0.16 0.36 0.16 0.36 0.09 0.29
Exit in t 0.00 0.00 0.02 0.15 0.00 0.00Entry in t 0.06 0.23 0.04 0.19 0.00 0.00
Note: Number of observations in 2000: 12,762; 2007: 14,126; 2014: 13,227.
32
Table 2: Summary statistics by sectors in 2000, 2007 and 2014
Regulated FDI ln(TFP)
2000 2007 2014 2000 2007 2014 2000 2007 2014
Food products and beverages 0.01 0.28 0.16 0.02 0.03 0.04 9.42 9.48 10.13Tobacco products 0.00 0.27 0.78 0.00 0.00 0.00 8.44 8.42 10.35Textiles 0.00 0.11 0.10 0.05 0.04 0.05 10.68 10.47 11.22Wearing apparel 0.00 0.00 0.03 0.05 0.05 0.09 9.25 9.17 10.10Leather and leather products 0.00 0.00 0.00 0.06 0.07 0.09 11.05 10.85 11.67Wood and wood products, except furniture 0.33 0.31 0.30 0.04 0.04 0.06 10.13 9.85 10.67Pulp, paper and paper products 0.01 0.03 0.01 0.05 0.07 0.10 10.64 10.60 11.36Publishing, printing and recorded media 0.00 0.18 0.00 0.01 0.01 0.02 10.83 11.20 11.56Coke, refined petroleum products and nuclear fuel 0.00 0.00 0.00 0.29 0.00 0.11 10.75 10.94 11.06Chemicals and chemical products 0.09 0.36 0.32 0.14 0.16 0.19 12.49 12.59 13.34Rubber and plastics products 0.00 0.00 0.10 0.06 0.10 0.11 11.63 11.78 12.45Other non-metallic mineral products 0.00 0.71 0.11 0.02 0.01 0.03 9.58 9.51 10.35Basic metals 0.00 0.15 0.27 0.22 0.18 0.21 11.55 11.59 12.05Fabricated metal products 0.00 0.06 0.05 0.09 0.13 0.12 10.45 10.69 11.50Machinery and equipment 0.02 0.20 0.16 0.13 0.20 0.24 10.71 11.06 11.77Electrical equipment, office machinery, computers 0.00 0.00 0.00 0.26 0.32 0.24 12.32 12.30 13.43Radio, television and communication equipment 0.00 0.00 0.00 0.54 0.61 0.53 12.84 12.39 12.78Medical, precision and optical instruments 0.00 0.00 0.00 0.19 0.25 0.19 11.54 11.10 11.58Motor vehicles 0.00 0.00 0.00 0.13 0.20 0.22 11.94 12.05 13.13Other transport equipment 0.00 0.64 0.48 0.08 0.12 0.18 10.63 11.35 12.06Furniture and n.e.c. 0.00 0.14 0.09 0.05 0.06 0.10 10.03 10.04 10.95
Total 0.03 0.22 0.15 0.05 0.06 0.08 10.13 10.05 10.97
Note: Average share of regulated firms, average foreign capital share and average log productivity within sectors. Number ofobservations in 2000: 12,762; 2007: 14,126; 2014: 13,227.
33
Table 3: Baseline results: Impact on FDI
Dependent variable: FDI share MNE
(1) (2) (3) (4) (5) (6)
Regulated −0.007*** −0.007*** −0.007*** −0.008*** −0.008*** −0.008***(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
Controls Yes Yes Yes Yes YesSector-year interactions Yes Yes Yes YesSector traits in 2005 × Year Yes YesExtended traits in 2005 × Year YesTreatment in year t t t t t t
Observations 208,120 208,120 208,120 208,120 207,105 208,120Firms 30,508 30,508 30,508 30,508 30,444 30,508R-squared 0.875 0.875 0.876 0.876 0.876 0.873
Note: The dependent variables measure the share of foreign capital within each firm and a multinationalenterprise indicator turning 1 if the foreign capital share is above 10%. Time-variant controls includecategories of firm age and a public enterprise indicator. Five-digit sector traits in 2005 include sectorconcentration of sales, the share of blue collar workers and the share of public enterprises. The extended setof sector traits in 2005 adds employment growth, capital intensity, the share of foreign capital and importpenetration at the sectoral level, all interacted with a full set of year effects. All regression include firm andisland-year FE. Robust standard errors are clustered on firm level and reported in parentheses. Significanceat or below 1 percent (***), 5 percent (**) and 10 percent (*).
34
Table 4: Baseline results: Impact on productivity
Dependent variable: ln(TFP) ln(VAD/L)
(1) (2) (3) (4) (5) (6)
Regulated −0.009 −0.009 −0.036*** −0.036*** −0.034*** −0.034***(0.012) (0.012) (0.013) (0.013) (0.013) (0.013)
Controls Yes Yes Yes Yes YesSector-year interactions Yes Yes Yes YesSector traits in 2005 × Year Yes YesExtended traits in 2005 × Year YesTreatment in year t− 1 t− 1 t− 1 t− 1 t− 1 t− 1
Observations 185,978 185,978 185,978 185,978 184,999 185,978Firms 28,307 28,307 28,307 28,307 28,232 28,307R-squared 0.822 0.822 0.835 0.835 0.835 0.746
Note: The dependent variable is log of total factor productivity or log of value added per worker. Time-variant controls include categories of firm age and a public enterprise indicator. Five-digit sector traits in 2005include sector concentration of sales, the share of blue collar workers and the share of public enterprises. Theextended set adds employment growth, capital intensity, share of foreign capital and import penetration. Allregression include firm and island-year FE. Robust standard errors are clustered on firm level and reportedin parentheses. Significance at or below 1 percent (***), 5 percent (**) and 10 percent (*).
35
Table 5: Distinguishing between types of regulation
Dependent variable: FDI share ln(TFP)
(1) (2) (3) (4) (5) (6)
Licensing requirements 0.006* 0.006 0.006 −0.046* −0.046 −0.048(0.003) (0.004) (0.004) (0.026) (0.032) (0.032)
FDI limitations + bans −0.010*** −0.028**(0.002) (0.014)
FDI limitations −0.007 −0.008 −0.018 −0.020(0.006) (0.006) (0.052) (0.052)
FDI bans −0.010*** −0.013(0.002) (0.013)
Sector-wide bans −0.013*** −0.066**(0.004) (0.031)
Firm-specific bans −0.009*** −0.003(0.002) (0.015)
Controls Yes Yes Yes Yes Yes YesSector-year interactions Yes Yes Yes Yes Yes YesSector traits in 2005 × Year Yes Yes Yes Yes Yes YesTreatment in year t t t t− 1 t− 1 t− 1
Observations 208,120 208,120 208,120 185,978 185,978 185,978Firms 30,508 30,508 30,508 28,307 28,307 28,307R-squared 0.876 0.876 0.876 0.835 0.835 0.835
Note: The dependent variable is the foreign capital share within each firm or log total factor productivity.Time-variant controls include categories of firm age and a public enterprise indicator. Five-digit sectortraits in 2005 include sector concentration of sales, the share of blue collar workers and the share ofpublic enterprises. All regressions include firm and island-year FE. Robust standard errors are clusteredon firm level and reported in parentheses. Significance at or below 1 percent (***), 5 percent (**) and10 percent (*).
36
Table 6: Analysis of effect heterogeneity on FDI and TFP across firms
Dependent variable: FDI share ln(TFP)
(1) (2) (3) (4) (5) (6)
Regulated −0.004*** −0.003** −0.006*** −0.019 −0.009 −0.039***(0.001) (0.001) (0.002) (0.017) (0.021) (0.013)
Trading firm × regulated 0.000 −0.037(0.002) (0.025)
Foreign firm × regulated −0.037** 0.012(0.018) (0.048)
Mid productivity × regulated −0.002 −0.081***(0.002) (0.025)
High productivity × regulated −0.012** 0.027(0.005) (0.036)
Large firm × regulated −0.014 0.026(0.010) (0.048)
Controls Yes Yes Yes Yes Yes YesSector-year interactions Yes Yes Yes Yes Yes YesSector traits in 2005 × Year Yes Yes Yes Yes Yes YesTreatment in year t t t t− 1 t− 1 t− 1
Observations 208,120 208,120 208,120 185,978 185,978 185,978Firms 30,508 30,508 30,508 28,307 28,307 28,307R-squared 0.876 0.876 0.876 0.835 0.835 0.835
Note: The dependent variable is the share of foreign capital within each firm or log of total factor produc-tivity. Time-variant controls include categories of firm age and a public enterprise indicator. Five-digitsector traits in 2005 include sector concentration of sales, the share of blue collar workers and the shareof public enterprises. All regression include firm and island-year FE. Robust standard errors are clusteredon firm level and reported in parentheses. Significance at or below 1 percent (***), 5 percent (**) and 10percent (*).
37
Table 7: Robustness: Exit and entry of firms
Dependent variable: Entry Exit FDI share ln(TFP)
(1) (2) (3) (4) (5) (6)
Regulated −0.009*** 0.006** −0.009*** −0.009*** −0.047*** −0.047***(0.002) (0.003) (0.002) (0.002) (0.017) (0.014)
Entry firm × regulated 0.004 0.027(0.003) (0.023)
Exit firm × regulated 0.006** 0.053**(0.003) (0.025)
Controls Yes Yes Yes Yes Yes YesSector-year interactions Yes Yes Yes Yes Yes YesSector traits in 2005 × Year Yes Yes Yes Yes Yes YesTreatment in year t t− 1 t t t− 1 t− 1
Observations 194,515 185,978 194,515 194,515 185,978 185,978Firms 29,665 28,307 29,665 29,665 28,307 28,307R-squared 0.283 0.325 0.878 0.878 0.835 0.835
Note: The dependent variable is an entry (exit) indicator turning 1 if a firm enters (leaves) the sample int, the foreign capital share within each firm or log total factor productivity. Time-variant controls includecategories of firm age and a public enterprise indicator. Five-digit sector traits in 2005 include sectorconcentration of sales, the share of blue collar workers and the share of public enterprises. All regressioninclude firm and island-year FE. Robust standard errors are clustered on firm level and reported inparentheses. Significance at or below 1 percent (***), 5 percent (**) and 10 percent (*).
38
Table 8: Robustness: Sector switching behavior
Dependent variable: Switch FDI share ln(TFP)
(1) (2) (3) (4) (5) (6)
Regulated −0.005 0.001 −0.007*** −0.008*** −0.037*** −0.041***(0.005) (0.005) (0.002) (0.002) (0.013) (0.015)
Sector switch −0.001 −0.019***(0.001) (0.006)
Switch into regulated sector 0.000 0.001(0.002) (0.020)
Switch into nonregulated sector −0.003 −0.006(0.003) (0.021)
Switch within regulated sectors −0.003 0.011(0.006) (0.043)
Switch within nonregulated sectors −0.001 −0.023***(0.001) (0.007)
Controls Yes Yes Yes Yes Yes YesSector-year interactions Yes Yes Yes Yes Yes YesSector traits in 2005 × Year Yes Yes Yes Yes Yes YesTreatment in year t t− 1 t t t− 1 t− 1
Observations 194,515 185,978 194,515 194,515 185,978 185,978Firms 29,665 28,307 29,665 29,665 28,307 28,307R-squared 0.373 0.382 0.878 0.878 0.835 0.835
Note: The dependent variable is a sector switch indicator turning 1 if a firm switches its operating sector in t, theforeign capital share within each firm or log total factor productivity. Time-variant controls include categoriesof firm age and a public enterprise indicator. Five-digit sector traits in 2005 include sector concentration ofsales, the share of blue collar workers and the share of public enterprises. All regression include firm andisland-year FE. Robust standard errors are clustered on firm level and reported in parentheses. Significance ator below 1 percent (***), 5 percent (**) and 10 percent (*).
39
Table 9: Spillover effects on FDI and TFP
Dependent variable: FDI share MNE ln(TFP) ln(VAD/L)
(1) (2) (3) (4) (5) (6)
Regulated −0.011*** −0.011*** −0.014*** −0.038*** −0.036** −0.025*(0.002) (0.002) (0.003) (0.015) (0.015) (0.015)
Reg. horizontal 0.011*** 0.012*** 0.018*** 0.009 0.007 −0.018(0.003) (0.003) (0.004) (0.025) (0.025) (0.025)
Reg. backward 0.005 0.096(0.005) (0.078)
Reg. backward (within 2 digit) 0.002 0.003 0.117 0.140*(0.005) (0.006) (0.081) (0.078)
Reg. backward (across 2 digit) 0.089* 0.093 −0.513* −0.543*(0.053) (0.071) (0.307) (0.303)
Reg. forward −0.017* 0.021(0.009) (0.065)
Reg. forward (within 2 digit) −0.011 −0.013 −0.037 −0.059(0.009) (0.011) (0.073) (0.072)
Reg. forward (across 2 digit) −0.033 −0.032 0.179 0.130(0.026) (0.029) (0.110) (0.109)
Controls Yes Yes Yes Yes Yes YesSector-year interactions Yes Yes Yes Yes Yes YesSector traits in 2005 × Year Yes Yes Yes Yes Yes YesTreatment in year t t t t− 1 t− 1 t− 1
Observations 208,120 208,120 208,120 185,975 185,975 185,975Firms 30.508 30.508 30.508 28.307 28.307 28.307R-squared 0.876 0.876 0.873 0.835 0.835 0.746
Note: The dependent variable is the share of foreign capital within each firm, a multinational enterpriseindicator turning 1 if the foreign capital share is above 10%, log of total factor productivity or log of valueadded per worker. Time-variant controls include categories of firm age and a public enterprise indicator. Five-digit sector traits in 2005 include sector concentration of sales, the share of blue collar workers and the shareof public enterprises. All regression include firm and island-year FE. Robust standard errors are clustered onfirm level and reported in parentheses. Significance at or below 1 percent (***), 5 percent (**) and 10 percent(*).
40
A Appendix
A.1 Cleaning the firm data
Matching the yearly firm panel and the NIL regulatory data requires that five digit sector
codes are unambiguously available for all observations of each firm. We first deal with
missing or incomplete sector codes, which is especially relevant in the years 2001 and
2003. We consider sector codes as incomplete if they have less than five-digits, e.g., ‘151’
instead of ‘15111’. Whenever plausible and unambiguously possible, we impute the same
code as in the year before or in the next year.
We exclude all observations for which sector codes are still missing or incomplete after this
adjustment. Additionally, sector codes are converted to the common standard of KBLI
2000 coding based on conversion tables provided by BPS. We drop all observations with
ambiguous conversion results.
We start with 351,366 observations over the 15 years, of which 1,229 have to be removed
due to missing or incomplete coding. 472 additional observations are lost because of
ambiguous conversion results between the years. These may occur because of a split or
unification of sector codes between different versions.
In order to estimate total factor productivity, TFP, we rely on information on the capital
stock, employment, the value of intermediate inputs, and value added. Out of these
four core variables, capital stock is the one missing most frequently. As common in the
literature (Amiti and Konings 2007), we interpolate the capital stock if values are missing
in a unique year. This is especially relevant in the year 2006, where information on the
capital stock is missing for a large part of the firm sample. We are able to interpolate
21,304 observations within 17,010 firms. Even after interpolation, we have to drop 120,561
observations within 30,361 firms due to missing information on the capital stock. Thereby
we lose 10,307 firms completely. We investigate the sensitivity of our results to these
missing observations by also repeating our estimates for a larger sample using value added
per worker as a proxy for productivity, the results on which are comparable.
In a next step, we exclude extreme outliers by dropping all observations for which inputs
or output are not within the threefold of the inter-quartile range above and below the
25 and 75 percentiles. We deal with extraordinary spikes in the data by also dropping
all observations with plant-level input growth (labor, intermediate inputs and capital) as
well as output growth that is outside the first and ninety-ninth percentile range of each
variable’s distribution. These steps reduce our sample size by further 12,642 observations
within 8,026 different firms.
Finally, we drop all firms with only one observation within the sample period. We end
up with a final dataset of 209,048 observations pertaining to 31,025 firms.
41
A.2 Merging the NIL conditions to product codes
Both the Negative Investment List (NIL) and the main products of the firm (KBLI) are
encoded at the same five-digit level that we use to determine a sectoral match between
firms and regulated sectors.
Unlike in later years, NIL 2000 does not yet provide KBLI codes, but only states the
names of the included sectors. Thus, in this one year we match the verbally stated sector
names to the corresponding KBLI sector codes. Furthermore, as the NIL 2007 slightly
changes in 2008 by an amendment to the existing regulation, we use the content of the
first draft of NIL for 2007 and the amendment for the years starting with 2008. We
convert the changing KBLI sector codes between the years and adjust the coding of the
NIL 2010, NIL 2014 and NIL 2016 to the KBLI 2000 standard. The regulatory and firm
data are merged according to the five-digit KBLI 2000 sector codes and the relevant year.
As several of the regulatory instruments are conditional on firm characteristics (see Table
A1 for a more detailed representation), we encode them conditional on firm attributes:
• Closed [closed to new investment in general] applies to all kinds of investment, both
domestic and foreign. We set this regulatory measure to zero for firms that have
already existing foreign involvement as the regulation is forward looking and cannot
restrict foreign participation anymore.24 The average FDI share among these firms
is 85% with the majority of firms reporting full foreign ownership of 100%. Thus,
these firms are not limited by forward looking regulation since a further increase is
not feasible for them anyway.
• Condition a [opened to small and medium sized firms] is conditional on firm size as
regulated in law 20/2008 on micro, small and medium enterprises (see Presidential
Decree 36/2010). According to Law 20/2008, firms should be considered as large if
they have annual revenues from sales above 50 billion Rupiah and assets (excluding
land and buildings) equal or above 10 billion Rupiah. The earlier Presidential
Decree 77/2007 refers to the law 9/1995 on small enterprises, which establishes
similar thresholds in real terms. When applying the firm size thresholds over time,
we adjust for inflation. Accordingly, we generate an indicator variable that encodes
large firms based on their annual sales and assets, thereby deviating from the most
commonly used definition in literature which relies on the number of workers. Due
to high volatility in the data, we use the median sales and median net assets of each
firm in order to circumvent wrong coding in cases of outliers. Hence, we consider
the classification into large and small enterprises to be time invariant.
24We use offsets in this and other categories for a total of 799 firms: 755 firms in closed sectors (defined byconditions closed, a, b, d, f, i) and further 44 firms that fall under FDI limitations (c, h).
42
• Condition b [opened to partnerships] depends on the legal status of a firm. We ex-
ploit information on the firm’s legal status given by the SI as regulation in condition
b only applies to firms that do not have the legal status of a partnership. Unfor-
tunately, the SI does not give any useful extra information on neither the exact
structure of the partnership nor the partner’s identity. Additionally, the variable on
legal status suffers from plenty of missing values. In these cases, we assume no reg-
ulation as the default. Therefore, we suspect that we may undercount firms subject
to condition b. We checked the robustness of our results to setting condition b to
apply sector-wide instead: TFP results stay practically the same also if we consider
this condition to apply to all firms within a five-digit sector, while the results for
FDI reduce in size and significance.
• Condition c [upper limit to foreign capital] sets a maximum share of capital that
can be owned by foreign investors. In nine out of ten cases the upper limit to foreign
capital is set to be 95 percent of total capital. We set this regulatory measure to
zero for firms that have already reached a foreign capital share above the threshold
as the regulation is forward looking and cannot restrict their foreign capital shares
anymore.
• Condition d [limited to certain locations] and condition g [upper limits of foreign
capital ownership in a certain location] are easily implemented by matching the
regulation with plant location. Regulation is applied if a firm is located outside the
authorized province.
• Condition e [licensing requirement] and condition h [upper limit to foreign capital
ownership and license] allow for (limited) FDI under the prerequisite of a valid
license issued by the appropriate authorities.
• Condition f [investment open to domestic capital] and condition i [investment open
to domestic capital and license] ban FDI in the affected sectors entirely.
To investigate the impact of these conditions on our outcome variables, we introduce two
major categories of regulation. First, we collect all conditions that are related to Licensing
requirements since they should impose costs to investment but are different from full FDI
bans. This includes conditions e, h and i.
Second, we generate the category FDI limitations and bans. We collect all remaining con-
ditions as well as conditions h and i, since they also impose hard limitations on FDI. To
account for heterogeneity within the different types of limits and bans, we subdivide the
category into four further groups. We first split the former category into FDI limitations
(including conditions c and h) and FDI bans (including conditions closed, d, f and i).
Furthermore, we distinguish between two last subcategories Sector-wide bans (including
43
conditions closed, d, f and i) and Firm-specific bans (including conditions a and b). Both
FDI limitations and Sector-wide bans apply to whole sectors irrespectively of firm char-
acteristics, whereas Firm-specific bans heavily rely on individual firm characteristics such
as size, legal status or location.
A.3 The Wooldridge approach for productivity estimation
Wooldridge (2009) suggests an alternative and more efficient way of estimating total
factor productivity compared to the well-known procedures by Olley and Pakes (1996)
or Levinsohn and Petrin (2003). Hereby, estimation of total factor productivity needs to
account for potential simultaneity bias due to correlation of input choices with the error
term.25
By using small letters to denote logs, total output can be decomposed as:
vat = yt −mt = α0 + αllt + αkkt + ωt + et, (A1)
where vat denotes value added, yt stands for total output, lt for labor, kt for capital, and
mt for other intermediate inputs. The error term combines an unobserved productivity
shock component, ωt, which is correlated with the input choices, and the independently
identically distributed error term component, et. For the i.i.d. component it must hold
that
E(et|lt, kt,mt, lt−1, kt−1,mt−1, ..., l1, k1,m1) = 0. (A2)
At the same time, assume that the dynamics of productivity shocks are restricted to
E(ωt|kt, lt−1, kt−1,mt−1, ..., l1, k1,m1) = E(ωt|ωt−1)
= j(ωt−1),(A3)
where ωt−1 = g(kt−1,mt−1).
By introducing productivity innovations at, the error components turns to
ωt = j(ωt−1 + at), (A4)
under the assumption that
E(at|kt, lt−1, kt−1,mt−1, ..., l1, k1,m1) = 0. (A5)
Consequently, only the contemporaneous choice variables lt and mt are correlated with
innovations at, while kt and all past values of inputs are uncorrelated with at. The
25See CompNet Task Force (2014) for a more detailed description of the approach. Our notation followsthat of CompNet Task Force (2014).
44
production function becomes:
vat = α0 + αllt + αkkt + j(g(kt−1,mt−1)) + ut, (A6)
where ut = at + et and E(ut|kt, lt−1, kt−1,mt−1, ..., l1, k1,m1) = 0.
Assuming that the productivity process is a random walk with drift ωt = τ + ωt−1 + at
(cf. CompNet Task Force 2014) and the function g(.) takes the polynomial form of order
three, we can identify the coefficients of input factors αK and αL. Then, equation (A6)
becomes:
vat = (α0 + τ) + αllt + αkkt + g(kt−1,mt−1) + ut. (A7)
We estimate equation (A7) using a pooled instrumental variable approach, instrumenting
labor by the one period lag of labor input. The estimation relies on a two-step efficient
generalized method of moments (GMM) approach.
45
Table A1: Conditions of the NIL over time: affected sectors and regulated firms in the sample
Industry division closed a b c d e f h iRegulatedfirms insample
% share ofregulated
firmswithin
industry
% share ofregulatedfirms intotal
output
Panel A: NIL 2000
Food and beverages 3 0 0 0 0 1 0 0 0 26 0.79 0.85
Wood products 0 0 0 0 2 3 0 0 0 406 34.73 2.93
Pulp and paper 0 0 0 0 0 1 0 0 0 2 0.78 0.57
Publishing and printing media 0 0 0 0 0 1 0 0 0 0 0 0
Chemicals 2 0 2 0 0 1 0 0 0 55 9.34 0.65
Machinery and equipment 1 0 0 0 0 0 0 0 0 3 1.66 0.00
Regulated firms in sample 41 0 28 0 261 162 0 0 0 492 3.44 5.00
Panel B: NIL 2007
Food and beverages 3 14 7 7 0 0 0 1 0 1118 28.90 8.44
Tobacco products 0 1 3 0 0 3 0 0 0 280 30.34 9.20
Textiles 0 3 1 0 0 0 0 0 0 182 10.98 0.05
Wood products 0 7 5 0 0 4 0 0 0 288 30.80 1.24
Pulp and paper 0 0 0 0 0 2 0 0 0 7 2.56 1.57
Publishing and printing media 0 0 0 0 0 1 2 0 0 69 18.75 0.47
Chemicals 3 1 1 3 0 2 1 2 0 179 34.76 2.50
Rubber and plastic 0 1 0 0 0 0 0 0 0 0 0 0
Other non-metallic mineral prod. 0 1 11 0 0 0 0 0 0 923 70.46 4.75
Basic metals 1 0 0 0 0 1 0 0 0 18 17.14 0.82
Fabricated metal products 0 4 1 0 0 0 0 0 0 29 6.94 0.19
Machinery and equipment 0 0 3 0 0 0 0 0 1 32 18.10 0.06
Other transport equipment 0 0 4 0 0 0 0 0 0 83 63.85 1.37
Furniture 0 1 6 0 0 0 0 0 0 245 14.69 0.64
Regulated firms in sample 67 130 3150 87 0 333 103 123 1 3453 22.26 31.30
Panel C: NIL 2010
Food and beverages 3 16 9 0 0 0 0 11 0 1254 31.75 12.20
Tobacco products 0 1 1 0 0 3 0 1 0 685 88.73 4.52
Textiles 0 5 1 0 0 0 0 0 0 271 13.53 0.68
Wearing apparel 0 1 0 0 0 0 0 0 0 26 2.17 0.39
Wood products 0 7 5 0 0 5 0 0 0 313 35.45 1.23
Pulp and paper 0 0 0 0 0 2 0 0 0 3 1.01 1.77
Publishing and printing media 0 0 0 0 0 1 2 0 0 7 2.58 0.00
Chemicals 3 1 1 2 0 3 1 3 0 201 31.96 1.50
Rubber and plastic 0 3 0 0 0 1 0 3 0 39 3.95 2.22
Other non-metallic mineral prod. 0 1 6 0 0 0 0 0 0 143 11.33 0.03
Basic metals 0 0 0 0 0 1 0 0 0 31 19.75 0.85
Fabricated metal products 0 4 2 0 0 0 0 0 0 68 12.76 0.27
Machinery and equipment 0 0 3 0 0 0 0 0 1 35 18.32 0.34
Other transport equipment 0 0 4 0 0 0 0 0 0 98 51.90 0.79
Furniture 0 1 5 0 0 1 0 0 0 194 13.23 0.27
Regulated firms in sample 54 297 2712 84 0 515 53 469 2 3368 21.45 27.06
Panel D: NIL 2014
Food and beverages 3 16 7 4 0 0 0 11 0 702 17.31 16.08
Tobacco products 0 1 1 0 0 3 0 1 0 486 79.41 3.97
Textiles 0 5 1 0 0 0 0 0 0 167 9.77 0.97
Wearing apparel 0 1 0 0 0 0 0 0 0 44 3.91 0.78
Wood products 0 7 3 0 0 5 0 0 0 216 29.92 1.48
Pulp and paper 0 0 0 0 0 2 0 0 0 4 1.43 1.56
Publishing and printing media 0 0 0 0 0 1 2 0 0 7 2.34 0.22
Chemicals 3 1 1 2 0 3 1 3 0 197 30.26 1.46
Rubber and plastic 0 2 0 0 0 0 0 2 1 109 11.19 2.27
Other non-metallic mineral prod. 0 1 6 0 0 0 0 0 0 127 10.84 0.05
Basic metals 0 0 0 0 0 1 0 0 0 44 25.29 0.92
Fabricated metal products 0 4 1 0 0 0 0 0 1 28 5.34 0.18
Machinery and equipment 0 0 3 0 0 0 0 0 1 33 14.47 0.46
Other transport equipment 0 0 4 0 0 0 0 0 1 98 47.57 0.68
Furniture 0 1 5 0 0 1 0 0 0 121 9.70 0.33
Regulated firms in sample 42 297 1672 98 0 513 48 465 113 2383 16.00 31.41
Note: Panels A to D outline the sectoral incidence of various forms of regulation in the NIL 2000, 2007, 2010 and 2014. In eachpanel, two-digit sectors are displayed in rows and the various conditions of the NIL in columns (closed and a to i). The figuresdisplay the number of five-digit sectors that are subject to the specific form of regulation in the respective year whereas the lastcolumns display sectoral penetration of regulation. The specific conditions include: a - Reserved for micro, small and mediumenterprises and cooperatives. b - Reserved for partnerships. c - Upper limit to foreign capital ownership. d - Limited to certainlocations. e - Special license required. f - 100% local capital. g - Upper limit to foreign capital ownership and limited location. h- Special license and upper limit to foreign capital ownership. i - 100% local capital and special license.
47
Table A2: Summary statistics of the main variables
Mean SD Minimum Maximum Observations
Regulated 0.12 0.32 0.00 1.00 208,120Licensing requirements 0.03 0.17 0.00 1.00 208,120FDI limitations + bans 0.10 0.31 0.00 1.00 208,120Sector-wide limits 0.01 0.12 0.00 1.00 208,120FDI bans 0.10 0.29 0.00 1.00 208,120Sector-wide bans 0.02 0.12 0.00 1.00 208,120Firm-specific bans 0.08 0.27 0.00 1.00 208,120
Reg. horizontal 0.12 0.23 0.00 1.00 208,120Reg. backward 0.04 0.12 0.00 0.88 208,120
Reg. backward (within 2 digit) 0.03 0.12 0.00 0.88 208,120Reg. backward (across 2 digit) 0.01 0.02 0.00 0.19 208,120
Reg. forward 0.05 0.07 0.00 0.38 208,120Reg. forward (within 2 digit) 0.02 0.06 0.00 0.35 208,120Reg. forward (across 2 digit) 0.02 0.05 0.00 0.35 208,120
FDI share 0.07 0.24 0.00 1.00 208,120MNE 0.08 0.27 0.00 1.00 208,120ln(TFP) 10.40 1.63 0.51 20.39 208,120ln(VAD/L) 10.18 1.31 1.06 19.99 208,120ln(K) 14.12 2.11 4.58 23.56 208,120ln(L) 4.15 1.14 3.04 9.26 208,120Firm age below 5 years 0.15 0.35 0.00 1.00 208,120Firm age between 5-15 years 0.37 0.48 0.00 1.00 208,120Firm age between 15-25 years 0.28 0.45 0.00 1.00 208,120Firm age above 25 years 0.21 0.41 0.00 1.00 208,120Switch in t 0.15 0.35 0.00 1.00 208,120
Switch into regulated 0.01 0.11 0.00 1.00 208,120Switch into nonregulated 0.01 0.10 0.00 1.00 208,120Switch within regulated 0.00 0.06 0.00 1.00 208,120Switch within nonregulated 0.12 0.33 0.00 1.00 208,120
Exit in t 0.04 0.21 0.00 1.00 208,120Entry in t 0.03 0.18 0.00 1.00 208,120
48
Table A3: Number of observations in the full sample by year
Year FDI regressions TFP regressions
2000 12,7622001 13,619 12,0762002 13,744 13,0922003 13,737 13,0732004 13,897 12,9302005 13,799 12,9142006 11,487 11,4152007 14,126 13,4502008 15,079 14,4632009 15,077 14,6592010 14,576 14,0282011 14,556 13,8962012 14,288 13,6852013 14,146 13,5732014 13,227 12,724
Total 208,120 185,978
49
Table A4: Production function coefficients by two-digit sector
ln(TFP)
Sector Labor Capital Observations
Food products and beverages 15 0.567*** 0.149*** 43,607Tobacco products 16 0.643*** 0.114*** 7,885Textiles 17 0.546*** 0.081*** 18,300Wearing apparel 18 0.795*** 0.093*** 14,886Leather and leather products 19 0.690*** 0.027 4,269Wood and wood products, except furniture 20 0.597*** 0.115*** 10,527Pulp, paper and paper products 21 0.574*** 0.112*** 3,088Publishing, printing and recorded media 22 0.679*** 0.036 4,165Coke, refined petroleum products and nuclear fuel 23 0.514*** 0.146* 361Chemicals and chemical products 24 0.443*** 0.059*** 7,185Rubber and plastics products 25 0.510*** 0.061*** 10,752Other non-metallic mineral products 26 0.440*** 0.142*** 15,444Basic metals 27 0.535*** 0.132** 1,633Fabricated metal products 28 0.659*** 0.084*** 6,304Machinery and equipment 29 0.617*** 0.089*** 2,624Electrical equipment, office machinery, computers 31 0.669*** 0.021 1,544Radio, television and communication equipment 32 0.598*** 0.044 1,102Medical, precision and optical instruments 33 0.543*** 0.096 413Motor vehicles 34 0.595*** 0.049 1,897Other transport equipment 35 0.513*** 0.121*** 2,030Furniture and n.e.c. 36 0.718*** 0.066*** 16,980
Note: The production function is estimated by GMM according to Wooldridge (2009).
50
Table A5: The time pattern of regulation
Dependent variable: FDI share MNE ln(TFP) ln(VAD/L)
(1) (2) (3) (4)
Regulated in t+ 3 0.002 0.005** 0.016 0.018(0.002) (0.002) (0.014) (0.014)
Regulated in t+ 2 −0.001 −0.002 0.009 0.007(0.002) (0.002) (0.013) (0.013)
Regulated in t+ 1 −0.006*** −0.007*** 0.013 0.018(0.002) (0.002) (0.013) (0.013)
Regulated in t −0.008*** −0.010*** −0.021 −0.016(0.002) (0.003) (0.014) (0.014)
Regulated in t− 1 0.004** 0.005** −0.020 −0.022(0.002) (0.002) (0.014) (0.014)
Regulated in t− 2 −0.001 −0.000 −0.035*** −0.033**(0.002) (0.002) (0.013) (0.013)
Regulated in t− 3 0.002 0.003 −0.009 −0.012(0.002) (0.002) (0.014) (0.014)
Controls Yes Yes Yes YesSector-year interactions Yes Yes Yes YesSector traits in 2005 × Year Yes Yes Yes Yes
Observations 116,555 116,555 116,555 116,555Firms 18,207 18,207 18,207 18,207R-squared 0.886 0.882 0.840 0.748
Note: The dependent variable is the share of foreign capital within each firm, amultinational enterprise indicator turning 1 if the foreign capital share is above10%, log of total factor productivity or log value added per worker. Time-variantcontrols include categories of firm age and a public enterprise indicator. Five-digit sector traits in 2005 include sector concentration of sales, the share of bluecollar workers and the share of public enterprises. All regression include firm andisland-year FE. Robust standard errors are clustered on firm level and reportedin parentheses. Significance at or below 1 percent (***), 5 percent (**) and 10percent (*).
51
Table A6: Shifting the time window of regulation
Dependent variable: FDI share ln(TFP)
(1) (2) (3) (4) (5) (6) (7) (8)
Regulated −0.008*** −0.007*** −0.008*** −0.002 −0.025* −0.001 −0.030** −0.036***(0.002) (0.002) (0.002) (0.002) (0.013) (0.012) (0.013) (0.013)
Controls Yes Yes Yes Yes Yes Yes Yes YesSector-year interactions Yes Yes Yes Yes Yes Yes Yes YesSector traits in 2005 × Year Yes Yes Yes Yes Yes Yes Yes YesTreatment in year t t+ 1 t t− 1 t t+ 1 t t− 1
Observations 196,049 196,049 185,978 185,978 196,049 196,049 185,978 185,978Firms 28,674 28,674 28,307 28,307 28,674 28,674 28,307 28,307R-squared 0.875 0.875 0.877 0.877 0.829 0.829 0.835 0.835
Note: The dependent variable is the foreign capital share within each firm or log total factor productivity. Time-variant controlsinclude categories of firm age and a public enterprise indicator. Five-digit sector traits in 2005 include sector concentration ofsales, the share of blue collar workers and the share of public enterprises. All regression include firm and island-year FE. Robuststandard errors are clustered on firm level and reported in parentheses. Significance at or below 1 percent (***), 5 percent (**)and 10 percent (*).
52
Table
A7:
Impact
on
pro
duct
ivity
-Lev
els
ofT
FP
esti
mati
on
and
sect
or-
spec
ific
defl
ato
rs
Dep
enden
tvari
able
:ln
(TFP
)ln
(TFP
),3-d
igit
ln(T
FP
),3-d
igit
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
Reg
ula
ted
−0.0
09
−0.0
36***
−0.0
36***
−0.0
14
−0.0
31**
−0.0
31**
−0.0
27**
−0.0
32**
−0.0
32**
(0.0
12)
(0.0
13)
(0.0
13)
(0.0
12)
(0.0
13)
(0.0
13)
(0.0
12)
(0.0
13)
(0.0
13)
Contr
ols
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Sec
tor-
yea
rin
tera
ctio
ns
Yes
Yes
Yes
Yes
Yes
Yes
Sec
tor
traits
in2005×
Yea
rY
esY
esY
esTre
atm
ent
inyea
rt−
1t−
1t−
1t−
1t−
1t−
1t−
1t−
1t−
1
5-d
igit
sect
or
spec
ific
defl
ato
rsYes
Yes
Yes
Obse
rvations
185,9
78
185,9
78
185,9
78
185,9
78
185,9
78
185,9
78
185,6
44
185,6
44
185,6
44
Fir
ms
28,3
07
28,3
07
28,3
07
28,3
07
28,3
07
28,3
07
28,2
69
28,2
69
28,2
69
R-s
quare
d0.8
22
0.8
35
0.8
35
0.8
24
0.8
35
0.8
35
0.8
36
0.8
44
0.8
44
Note
:T
he
dep
enden
tvariable
islo
gto
talfa
ctor
pro
duct
ivity
as
estim
ate
don
the
two
dig
it(c
olu
mn
1to
3)
and
thre
edig
itse
ctor
level
(colu
mn
4to
9).
Colu
mn
7to
9additio
nally
use
five-
dig
itse
ctor-
spec
ific
input
and
whole
sale
pri
cedefl
ato
rs.
Tim
e-vari
ant
contr
ols
incl
ude
cate
gori
esof
firm
age
and
apublic
ente
rpri
sein
dic
ato
r.Fiv
e-dig
itse
ctor
traits
in2005
incl
ude
sect
or
conce
ntr
ation
ofsa
les,
the
share
ofblu
eco
llar
work
ers
and
the
share
of
public
ente
rpri
ses.
All
regre
ssio
nin
clude
firm
and
isla
nd-y
ear
FE
.R
obust
standard
erro
rsare
clust
ered
on
firm
level
and
report
edin
pare
nth
eses
.Sig
nifi
cance
at
or
bel
ow
1per
cent
(***),
5per
cent
(**)
and
10
per
cent
(*).
53
Table A8: Alternative TFP measures
Dependent variable: ln(TFP) ln(VAD) ln(VAD/L) ln(VAD/K)
(1) (2) (3) (4) (5) (6)
Regulated in t− 1 −0.036*** −0.026* −0.034*** −0.039*** −0.045*** −0.051***(0.013) (0.014) (0.013) (0.013) (0.012) (0.013)
Control for ln(K/L) Yes Yes
Regulated in t −0.029** −0.015 −0.023* −0.025* −0.046*** −0.027***(0.013) (0.014) (0.013) (0.013) (0.013) (0.009)
Control for ln(K/L) Yes Yes
Note: All entries report separate estimation results of regulation in t − 1 or t. The dependentvariable is log total factor productivity, log value added or log value added per worker (or percapital). All regressions include time-variant controls (categories of firm age, a public enterpriseindicator) and interactions of five-digit sector traits in 2005 (sector concentration of sales, shareof blue collar workers, share of public enterprises) with year dummies, as well as firm, island-year and sector-year FE. The number of observations is 208,120 when using regulation in t and185,978 for regulation in t− 1. Robust standard errors are clustered on firm level and reportedin parentheses. Significance at or below 1 percent (***), 5 percent (**) and 10 percent (*).
54
Table A9: Impact on productivity while controlling for FDI
Dependent variable: ln(TFP) ln(VAD/L)
(1) (2) (3) (4) (5) (6)
Regulated −0.009 −0.008 −0.036*** −0.036*** −0.034*** −0.034***(0.012) (0.012) (0.013) (0.013) (0.013) (0.013)
FDI share 0.012 0.028 0.027 0.029 0.024(0.028) (0.026) (0.026) (0.026) (0.026)
Controls Yes Yes Yes Yes YesSector-year interactions Yes Yes Yes YesSector traits in 2005 × Year Yes YesExtended traits in 2005 × Year YesTreatment in year t− 1 t− 1 t− 1 t− 1 t− 1 t− 1
Observations 185,978 185,978 185,978 185,978 184,999 185,978Firms 28,307 28,307 28,307 28,307 28,232 28,307R-squared 0.822 0.822 0.835 0.835 0.835 0.746
Note: The dependent variable is log total factor productivity or log value added per worker. Time-variantcontrols include categories of firm age and a public enterprise indicator. Five-digit sector traits in 2005include sector concentration of sales, the share of blue collar workers and the share of public enterprises. Allregression include firm and island-year FE. Robust standard errors are clustered on firm level and reportedin parentheses. Significance at or below 1 percent (***), 5 percent (**) and 10 percent (*).
55
Table A10: Full sample including observations with missing capital
Dependent variable: FDI share ln(VAD/L)
(1) (2) (3) (4) (5) (6)
Regulated −0.010*** −0.010*** −0.011*** 0.001 −0.024** −0.024**(0.001) (0.002) (0.002) (0.010) (0.010) (0.010)
Controls Yes Yes Yes Yes Yes YesSector-year interactions Yes Yes Yes YesSector traits in 2005 × Year Yes YesFull sample (incl. missing capital) Yes Yes Yes Yes Yes YesTreatment in year t t t t− 1 t− 1 t− 1
Observations 298,487 298,487 298,487 267,582 267,582 267,582Firms 37,641 37,641 37,641 35,126 35,126 35,126R-squared 0.864 0.865 0.865 0.728 0.732 0.732
Note: The dependent variable is the foreign capital share within each firm or log value added per worker.Time-variant controls include categories of firm age and a public enterprise indicator. Five-digit sector traits in2005 include sector concentration of sales, the share of blue collar workers and the share of public enterprises.All regression include firm and island-year FE. The sample is larger compared to the baseline sample sizedue to non-omission of missing or zero capital values. Robust standard errors are clustered on firm level andreported in parentheses. Significance at or below 1 percent (***), 5 percent (**) and 10 percent (*).
56