Japan & The World Economy 55 (2020) 101024

Contents lists available at ScienceDirect

Japan & The World Economy

journal homepage: www.elsevier.com/locate/jwe

Do domestic producers benefit from safeguards? The case of a Japanese T

safeguard on Chinese vegetable imports in 2001⋆
Kazutaka Takechi
Faculty of Economics, Hosei University, 4342 Aihara-machi, Machida, Tokyo 194-0298, Japan

ARTICLE INFO ABSTRACT

JEL classification: This study examines the effects of a safeguard policy imposed by Japan in 2001 using detailed product-level
F13 transaction data from domestic markets. The market prices of imported and domestic goods are almost always
F14 higher during the safeguard period compared with those in the previous year. However, the safeguard measure
decreases the margins for imported goods, but does not affect the margins for domestic goods. As temporary
Keywords: import restrictions are expected to enable structural changes in the domestic industry, we also estimate the long-
Safeguard
term effect on margins. We find that five years after the safeguard period, the margins remain similar for do-
Markups
mestic goods and are smaller for imported goods. These results suggest that the temporary import restrictions
were both harmful to imported goods producers and consumers and unbeneficial to domestic producers.

1. Introduction beneficial for domestic (and foreign) suppliers.

The World Trade Organization acknowledges that transparent and
temporary trade restriction measures can be required to facilitate free-
trade negotiations by easing pressure from imports competing with
domestic industries. A safeguard is one such import restriction mea-
sure.1 If there is a rapid increase in imports that damages the domestic
industry, temporary restrictions can be imposed as a form of safeguard.
Not only are such policies important from a political perspective, their
economic consequences are also of strong interest. Hence, evaluating
the impact of temporary import restrictions is a key task. We examine a
case in which the Japanese government imposed a provisional safe-
guard measure on Chinese vegetable imports in 2001.
The temporary import quotas limit the foreign supply; therefore,
market prices will rise as a natural consequence. However, whether
such changes affect imported goods or domestic goods, or both, is not
obvious. Furthermore, because of the increased trade costs, the benefits
for producers of increased prices may not be substantial. If trade re-
strictions do not raise domestic goods prices substantially and, thus, if
the level of profits remains at the pretrade policy level, the purpose for
which the policy was implemented will not be achieved. Hence, this
study examines the level of margins before, during, and after the
safeguard policy period and addresses whether import restrictions are
The effectiveness of temporary trade restrictions also hinges on
whether the import-competing sector is able to conduct structural
changes; for example, if the sector can innovate to develop new pro-
ducts or to find a more efficient production process, a temporary
safeguard measure may succeed in rescuing the domestic industry in a
transparent way. One way of creating innovation is by building a brand.
If the fact that goods are produced locally is attractive to consumers
(e.g., as exemplified by a “local production for local consumption”
slogan), then a temporary restriction can offer the industry a chance to
build awareness of the benefits of the domestically produced goods. In
the case of vegetable imports in Japan, there were concerns about the
level of pesticides remaining in imported vegetables. These concerns
offered domestic producers an effective way to boost the domestic in-
dustry by creating a reputation for domestic vegetables as being safer
and fresher. To determine whether this occurred requires an empirical
study of the safeguard. In this paper, using detailed domestic market
data enables us to undertake a comprehensive evaluation of the impact
of the trade policy on domestic suppliers. This study contributes to the
literature by showing how domestic market transactions changed in
response to the safeguard and, in particular, how the profits reacted to
the imposition of the safeguard in the long run.
We use domestic vegetable market data based on 55 geographically

⋆
I would like to thank two anonymous referees for useful comments, which greatly improve the paper. I am also grateful to Taiju Kitano, Kenmeni Tsubota, and
seminar participants at the ICES conference and RIETI for helpful discussions. The usual disclaimer applies. This study was conducted as a part of the Analyses of
Offshoring project undertaken at RIETI. I would also like to thank the hospitality of the Sauder School of Business at the University of British Columbia, where part of
this work was done during my visit. I gratefully acknowledge the financial support of the Japan Society for the Promotion of Science (Grant No. 18K01624).
E-mail address: ktakechi@hosei.ac.jp.
1
Theoretically, Bagwell and Staiger (1990) and Ethier (2002) show that a safeguard measure can facilitate trade liberalization negotiation.

https://doi.org/10.1016/j.japwor.2020.101024
Received 22 January 2020; Received in revised form 29 June 2020; Accepted 29 June 2020
Available online 14 July 2020
0922-1425/ © 2020 Elsevier B.V. All rights reserved.
K. Takechi
separated wholesale markets across Japan. Imported vegetables are also
traded in wholesale markets. We empirically investigate market out-
comes by using a hedonic price approach and demand function esti-
mations. By using hedonic approaches, we investigate the price effect of
trade policy in a difference-in-difference framework. This enables us to
isolate the effect of trade policy on imported goods exclusively. Because
market supply is limited, there may be spillover effects on domestic
goods, which may disguise the impact on imported goods. In addition,
it is important to observe the direct effect of trade policy and the
changes in price caused by the import quota. Furthermore, by esti-
mating hedonic price equations, the coefficients of the product char-
acteristics are the shadow prices of the product attributes. This enables
us to evaluate the changes in demand-side perceptions associated with
the safeguard measure.
Then, we estimate a structural model to detect the margins that
producers enjoy. We employ a demand estimation for differential goods
and assume that the supply side involves monopolistic competition. As
discussed later, the market for vegetables that we investigate is a dif-
ferential good market and the number of producers is large, yet limited;
therefore, suppliers have market power to some extent. These char-
acteristics fit the properties of monopolistic competition. By obtaining
demand elasticity, we can compute margins for each product. We
analyze whether the margins change in association with the imposition
of import quotas. We also estimate the same framework using data for
the period five years after the end of the import restrictions. Whether
the temporary restrictions successfully induced structural changes in
the domestic industry can be determined by examining the margins;
that is, if these were higher five years later, it would indicate that the
import restrictions had positive effects on the profits of domestic pro-
ducers in the long run.
We consider the case of Welsh onions (Allium fistulosum), to which
the provisional safeguard measure against China applied from April 23,
2001 to November 8, 2001. The data utilized are daily traded wholesale
market data, which enables us to perform a detailed examination of the
effect of the trade policy on domestic markets. The empirical results of
the hedonic analysis show that the safeguard increased the market price
for Welsh onions. Moreover, the price increases resulting from the
safeguard influenced imported goods more than it influenced domestic
goods. Hence, the higher prices were largely achieved for the imported
goods. Then, our empirical analysis of the structural model demon-
strates that the margins of imported goods are low during the safeguard
period compared with pre- or postsafeguard periods, whereas the do-
mestic margins are virtually the same for all periods. Hence, although
consumers suffer from higher prices and foreign producers lose their
higher margins, the import restriction does not provide benefits for
domestic producers.
Considering the long-run effect, temporary import restrictions are
expected to lead to structural changes, such as attempts by the domestic
industry to build brand images that highlight the superiority of the
domestic product. However, we find that the demand parameters did
not change drastically five years after the provision of the safeguard.
This suggests that the temporary import restrictions did not result in
domestic producers taking advantage of the opportunity to gain a
competitive edge.
The economic impacts of safeguards have been studied empirically
in the literature. Several works confirm that safeguards have virtually
no impact on domestic markets. For example, Kitano and Ohashi (2009)
demonstrate that the US safeguard policy for motorcycles had little to
do with the survival of Harley–Davidson in the US in the face of com-
petition from imported motorcycles. Instead, the key factor was that
there was low substitutability between Harley–Davidson and the im-
ported motorcycles, such as Honda and Kawasaki. Hence, the safeguard
policy itself had little impact on increasing Harley–Davidson's profits.
Similarly, Chung et al. (2016) examine the effect of the US safeguard
against Chinese tire imports and find no effect on the US labor market
because, even though Chinese imports were restricted, there were
2
Japan & The World Economy 55 (2020) 101024
increases in imports from other countries. While the presence of a third
country may have important consequences for trade policy, this is not
the case in relation to Japan's vegetable imports from China. Here, we
can focus on the imports from China and still find no empirical evidence
of successful temporary trade restrictions for domestic agricultural
producers. Because under the WTO, approximately 21% of safeguards
have been initiated in agricultural sectors (79 cases out of a total of
377), we contribute to the literature by assessing the effects of safe-
guard policy in this frequently used sector.
2. Background
Due to a surge of vegetable imports from China, Japanese farmers
sought to take action against imported vegetables. On December 22,
2000, the Japanese government initiated an investigation into a pro-
visional safeguard measure. There are three requirements to initiate a
safeguard: there is a surge of imports, the domestic industry is seriously
damaged, and there is causality between increases in imports and do-
mestic damage. If these conditions are met, under the WTO (GATT
Article 19), temporary import restrictions can be imposed. If such an
action is needed to be taken urgently, a provisional measure can be
undertaken, following which a definitive measure would be imposed. In
our case, after a four-month investigation process, a provisional safe-
guard measure was implemented on April 23, 2001 (for more details,
see Mulgan, 2005 and Kuno, 2006).
The vegetables included in the safeguard measure were raw shiitake
mushrooms, Welsh onions, and igusa (rush grass). The actual import
restriction regime imposed was a quota tariff scheme: up to a certain
volume of imports, the existing tariff was levied; beyond that import
level, a higher import tariff rate was imposed. In this study, we focus on
Welsh onions. This is partly because of data availability for igusa, for
which no market transaction data are available. With respect to shiitake
mushrooms, because the actual import volume was below the quota
level, it was difficult to ascertain the impact of the safeguard. As shown
below, because the import volume of Welsh onions was above the quota
level, this trade policy is believed to affect market transactions. The
regular tariff rate for Welsh onions was 3%. However, between April 23
and November 8 (200 days), a 225-yen tariff rate per kg was levied,
which corresponds to a 256% tariff rate, for imports beyond 5383 tons.
Based on the government investigation report, the average price of
domestic Welsh onions was 337 yen per kg and that of imported Welsh
onions was 112 yen. Thus, the new tariff rate was designed to ap-
proximately bridge the gap between the domestic and imported goods.
As will be shown later, the volume of imports declined during the
safeguard period.
After Japan imposed the provisional safeguard, China retaliated by
increasing its import tariff on cell phones and automotive parts from
Japan. Consequently, there was a series of negotiation meetings held
between Japanese and Chinese government officials, which culminated
in an agreement to form a “Japan–China Trade Council on Agricultural
Products.” This is a private forum intended to promote information
exchange, which nonetheless may induce voluntary export restraints
(Kagitani and Hariyama, 2015). Under the agreement, Japan agreed not
to implement its safeguard measure, and China agreed to stop imposing
a higher rate of tariffs on Japanese imports. Thus, the safeguard against
Chinese vegetables in Japan did not last more than 200 days. That is,
only the provisional safeguard was imposed. This paper attempts to
study how market outcomes changed before and after the safeguard and
whether the temporary import restrictions affected economic outcomes
and, in particular, whether it influenced market prices and markups.
3. Data
To conduct this investigation, we use detailed domestic market
transaction data. The data set used in this study is the Daily Wholesale
Market Information on Fresh Fruit and Vegetables (Seikabutsu
K. Takechi
Hinmokubetsu Shikyo Joho in Japanese). This data set covers daily
transactions in Japanese agricultural wholesale markets, including
those for Welsh onions. Imported vegetables are also traded in whole-
sale markets and, hence, the changes in market outcomes that resulted
from the trade policy are reflected in these data. There are 47 pre-
fectures in Japan and each prefecture has at least one wholesale market,
ensuring that our data are spatially diversified. We use the data to ex-
amine the impact of the safeguard over four periods: 2000 (before the
safeguard measure), 2001 (when the safeguard was imposed between
April 23 and November 8), 2002 (after the safeguard measure was re-
moved), and 2007 (five years after the safeguard was removed) to de-
termine whether the policy had any long-run effects. The data record
daily transaction data, for example, there were 274 market open days in
2007.
This data set includes information on market price, quantity traded,
grade and size of goods, and the production place (prefecture or im-
ported country). That is, we can analyze a substitution pattern between
47 differentiated domestic products (based on origin prefectures) and
one imported (Chinese) product. We are able to control for product
characteristics, which enables us to abstract from the influence of dif-
ferences between the characteristics of the domestic and imported
goods. We consider that a product's characteristics are a combination of
brand (e.g., white Welsh onion or green Welsh onion), grade (e.g.,
“excellent” or “good”), size (e.g., “L” or “M”) and producing place
(“Tokyo” or “China”). Because there is a great deal of variety in terms of
grade and size categories in the data, there were 1145, 1124, 1151, and
1115 distinct Welsh onion products in 2000, 2001, 2002, and 2007,
respectively. This high level of categorization allows us to obtain the
price of individual goods and avoid aggregation biases when analyzing
prices.
Because we use data for the periods before and after the imposition
of the safeguard measure, we can compare the overall data pattern for
these years. Table 1 reports the descriptive statistics for each year. The
average price per kg is around 300–350 yen (approximately USD 3.30).
During this period, as there was no inflation in Japan, we can compare
the nominal prices. In the year when the safeguard measure was ap-
plied, the average price was relatively higher than in the previous and
following years, probably because of natural environmental conditions;
however, the 2007 price was almost the same as that in the safeguard
year. The average delivery distance was approximately 250 km from
2000 to 2001, whereas the distance was much longer in 2007. If a
certain producing prefecture suffers from a low level of production due
to natural disaster, other prefectures may supply to markets that they
have not reached before. This results in a higher price and longer de-
livery distance.
Table 1 also shows the ratio of the particular product characteristics
of high quality and large size. While there are many unidentified ca-
tegory names for quality and size, we take a conservative approach,
such that we use the identified names of categories. We can assume that
a grade name including syu (excellent) means the product is high
quality and that a size category including the letter “L” implies a large
size. High-quality products account for approximately 30% of the total
and large-sized products account for 60 percent. The share of imported
goods is also reported in Table 1. It was lower in 2001 compared with
Table 1
Summary statistics.
Year 2000 Year 2001 Year 2002 Year 2007
Average price 302.234 347.812 328.497 349.756
S.D. of price 228.348 243.494 212.088 245.486
Average delivery distance 266.755 255.177 266.271 334.154
Percentage of high quality 30.043 29.628 29.213 25.539
Percentage of large size 55.673 56.639 58.968 60.036
Percentage of imported goods 5.704 5.241 6.317 9.056
Num. of obs. 36,465 36,270 35,792 37,609
3
Japan & The World Economy 55 (2020) 101024
2000 and 2002, which is likely to be the effect of the safeguard imposed
during 2001. By 2007, the share of imported Welsh onions had risen to
9 percent.
Although the share of imported goods in the total is relatively small,
the presence of individual imported goods in each wholesale market is
sometimes significant. Because there are fragmented delivery patterns
in daily trade, we consider a time window of a month for each market,
which ensures stable delivery and market share patterns. In other
words, we assume that a market in a prefecture is defined on a monthly
basis because, if markets are defined on a daily basis, often only one
product is supplied in a market: delivery patterns are volatile daily.
Furthermore, the pattern of demand and supply changes because of
seasonality and such a variation does not drastically occur daily or
weekly.
Specifically, consider the following example. On January 5, 2001, in
Tokyo market, Welsh onions from Chiba and Saitama prefectures and
China were traded. On the other hand, on January 19, Welsh onions
from Chiba, Saitama, and this time, Gunma prefecture were traded. If
we consider that markets are defined on a daily basis, the markets on
these two dates are separated, that is, Welsh onions from China were a
substitute for Chiba and Saitama, but not for Gunma, and vice versa.
However, it is natural to consider that Welsh onions produced in
Gunma prefecture faced competition from Welsh onions from China.
Therefore, we assume that Chinese Welsh onions traded on the 5th were
a substitute for those from Gunma traded on the 19th, because these are
traded in the same month. This allows us to conduct a stable empirical
analysis using a market defined monthly. Furthermore, we consider that
the product from Saitama traded on the 5th and that traded on the 19th
are different products and are treated as substitutable. Even if the
product characteristics are the same, because the trading days are dif-
ferent, we assume that these goods are different. The share of individual
goods is calculated by the amount of vegetables traded on each day
divided by the total market size.
The total market size is defined by the aggregate total volume in
market n , Xn . The total market size corresponds to the total consump-
tion volume in each prefecture. This is calculated using two data series:
the Food Balance Sheet issued by the Ministry of Agriculture, Forestry
and Fisheries, which reports average consumption amounts of stem
vegetables (including cabbage, spinach, and Welsh onions) in each
prefecture; and the population numbers in each prefecture, reported by
the Census. The share of each product, j , is shown by snj = qnj/ Xn , where
qnj is the quantity of each good.
Fig. 1 shows the share of individual goods in each market. We take
the average of the market share, which is defined as the quantity sup-
plied divided by the market size in each month in each market. The
share of imported goods seems large based on the volume of imports.
The reason for this is as follows. Consider a market in which there are
two domestic goods, with market shares of 0.1 and 0.2, and one
Fig. 1. Share (individual goods).
K. Takechi Japan & The World Economy 55 (2020) 101024

imported good, with a market share of 0.1, then the average domestic
goods share is 0.15 and the share of imports is 0.1. Although domestic
goods account for 30% of the total market share, at the individual goods
level, the shares are more comparable.
By focusing on 2001, we can calculate how the average prices vary
between the periods with and without the safeguard in place. The
safeguard measure was in place, roughly, from May to October of 2001.
The average price per kg of domestic goods without the safeguard was
305.553 yen, whereas it was 388.996 yen under the safeguard. Hence,
the import restrictions imposed under the safeguard policy raised the
domestic goods price. However, the imported goods price was 157.031
yen without the safeguard and 257.343 yen under the safeguard. The
quota level was set to 5383 tons and the actual imported volume was
6251.884 tons, indicating that, although the quota was binding, the
volume of imports exceeding the quota limit was not large. Hence, the
increase in the imported goods price was not merely caused by the tariff Fig. 2. Imports of Welsh onions.
(of 225 yen per kg), but by other factors, such as changes in demand-
side behavior. Considering other variables that influence price, the distance be-
Although our focus is on using domestic transaction data, it is in- tween market and source is an important factor. We define the inter-
formative to observe the trade patterns for Welsh onions from China by prefectural distance as the direct distance between the prefectural head
examining the trade customs data in Table 2. These data are only offices in the prefectural capital cities.2 We set the internal distance to
provided on a monthly basis, but they indicate import patterns and, in 10 km, which is slightly less than the minimum interprefectural dis-
particular, show that a decrease of imports occurred during the provi- tance of 10.4 km (Kyoto–Shiga). Considering the imports from China,
sional safeguard period. Often, when quotas are imposed, they do not we do not directly observe the distance between production places in
end up being binding. However, in the case of Welsh onion imports China and markets in Japan. Because the vegetables are shipped by sea
from China, it appears that the quota was binding. Our primary interest and the three major ports in Japan are Tokyo and Osaka in the east and
is in investigating how this restricted supply affected domestic market west, respectively, and Fukuoka in the Kyushu area, we assume that the
outcomes. distance between markets and the source in China is the distance be-
Fig. 2 plots the monthly value of imports from 2000 to 2001. Be- tween Beijing and each prefecture where the port is located. Although
cause the safeguard measure was imposed from April 23, 2001 to No- this is a simplified assumption, it approximates actual shipment pat-
vember 8, 2001, it was mainly effective from May to October 2001. As terns.
the figure shows, the value of imports was lower in these months than Finally, we plot the delivery pattern of Welsh onions in Fig. 3. This
in other years, which means that the safeguard measure was effective. is a contour plot of the frequency of supply from an origin to a market.
To understand the effects of the safeguard on market outcome in The horizontal axis shows an index of origin prefecture and the vertical
more detail, we calculate monthly average prices of imported and do- axis is an index of markets. Number 1 represents the northernmost
mestic goods (Table 3). As a reference, the 2000 and 2002 prices are prefecture, Hokkaido. The higher the number, the more south geo-
also reported. During the safeguard period, the imported goods prices graphically is the corresponding prefecture; thus, number 47 is the
were higher than they were during the other months of 2001 or during southernmost prefecture, Okinawa. The final number, 48, corresponds
other years. Hence, by imposing higher tariffs and limiting the supply of to China. Each point (contour line) represents the frequency of de-
imported goods, the safeguard resulted in increased imported goods liveries from source prefectures to consuming markets, and the number
prices. Similarly, domestic goods prices were higher under the safe- of deliveries is large when the color of the point is bright. As shown, the
guard than during the other months of 2001. Compared with the 2000 bright diagonal line reflects a large number of local deliveries; that is,
and 2002 prices, the prices under the safeguard are higher in almost all Welsh onions produced in a prefecture are supplied to that prefecture's
months. Thus, the safeguard policy appears to have successfully in- market. The right-hand side of the origin axis shows China. Welsh
creased prices so that domestic producers were able to increase their onions produced in China are shipped across Japan on a nationwide
profits compared with the case without safeguards. However, com- basis. Thus, restricting Chinese imports affects every market in Japan.
paring the prices in the reference year indicates that although the prices
of imported goods almost doubled, those of domestic goods increased 4. Model
only by up to 20%.
To investigate the impact of the safeguard measure, we employ two
types of models: (1) a hedonic price model with a difference-in-differ-
Table 2 ence framework; and (2) a discrete-choice model to estimate demand
Import volume (million yen) of Welsh onions: * = safeguard period.
for differentiated goods by Berry (1994). The first approach is used to
Month Year 2000 Year 2001 Year 2002 show the direct consequences of trade restrictions. The second frame-
work is used to derive producers’ margins and to analyze the impact of
Jan 290.6 288.4 367.1
trade policy.
Feb 208.3 427.3 235.2
Mar 254.5 240.2 138.9 A hedonic price model is based on a consumer's utility-maximizing
Apr 178.8 165.2 130.8 behavior. Assume that a consumer in region n has the following utility
May 184.7 68.9* 236.8 when consuming good i with characteristic k :
Jun 306.7 78.4* 210.2
Jul 320.4 89.2* 189.2 un = k ln x i, k + c,
Aug 293.32 97.6* 157.7
Sep 379.9 119* 193.6 where k is a taste parameter, x i, k is good i 's characteristic k , and c is the
Oct 517.9 135.8* 347.6
Nov 331 432.6 440.9
2
Dec 406.2 554.6 525.9 This information is available from the Geospatial Information Authority of
Japan, see http://www.gsi.go.jp/KOKUJYOHO/kenchokan.html.

4
K. Takechi Japan & The World Economy 55 (2020) 101024

Table 3
Average monthly price: imported and domestic goods.
2000 2001 2002 Ratio (2001/2002)

Import Domestic Import Domestic Import Domestic Import Domestic

January 60.156 236.853 180.439 352.705 147.602 294.605 122.25% 119.72%

February 75.614 266.986 162.548 321.438 127.2 263.823 127.79% 121.84%
March 105.418 271.71 97.29 253.77 112.969 229.742 86.12% 110.46%
April 117.073 272.087 142.727 259.285 161.114 279.428 88.59% 92.79%
May 120.537 272.253 230.595 391.844 214.518 401.731 107.49% 97.54%
June 121.972 282.793 301.821 391.285 115.063 334.367 262.31% 117.02%
July 137.816 288.3 261.69 365.65 128.314 316.205 203.95% 115.64%
August 139.133 299.782 310.597 495.05 135.825 411.525 228.67% 120.30%
September 153.034 324.321 354.648 465.88 157.942 429.033 224.54% 108.59%
October 155.317 381.762 279.271 346.62 158.083 379.074 176.66% 91.44%
November 197.467 390.954 233.86 319.997 157.161 384.822 148.80% 83.15%
December 206.24 441.695 137.456 288.339 130.554 343.376 105.29% 83.97%

Fig. 3. Delivery pattern in 2001.

consumption of outside goods. The budget constraint is
y = c + 0 + x and y is income. Then, from the utility-maximization
problem, we have the following relationship:
k
= k.
x i, k
Thus, the shadow price is the marginal utility of characteristic k . This
fundamental parameter can be obtained using a hedonic price equation.
The hedonic price equation is expressed by a function of product
characteristics:
pnjt = f (grade, size, origin), (1)
where n is the market, j is the place of origin, and t is the month. Then,
we assume linearity in this price function and introduce product type
dummies, monthly dummies (month t ), and the dummy for policy
timing:
ln pnjt = a + b safeguard jt + c montht + d import j + e gradenjt + f size njt ,
(2)
where safeguard jt takes a value of one for the imported goods when the
safeguard measure was in place, import j takes a value of one for the
imported goods, grade njt takes a value of one if the quality of goods is
equal to or higher than excellent, and size njt takes the value of one if the
size of goods is equal to or larger than size L. The indicator of imported
goods from China reveals how consumers evaluate imported goods.
This perception could be changed by the domestic producers if they
used the safeguard period to build a domestic brand based on local
products being superior. Thus, for the period when the safeguard
measure was imposed, we add a safeguard dummy to the import index.
If the magnitude of this import index is different before, during, and
after the safeguard period, it indicates that the domestic industry suc-
cessfully changed consumers’ perceptions of products produced do-
mestically. Thus, the coefficient for the safeguard dummy reveals the
true impact of the trade policy on market prices.
We also investigate the effects of the safeguard measure by esti-
mating a demand equation. We assume that wholesalers purchase a
particular vegetable on each day from an origin prefecture, which
yields the highest benefit. Hence, we employ a discrete-choice approach
for demand. In particular, we adopt a simple demand estimation pro-
cedure for differentiated goods by Berry (1994).
A discrete-choice model of demand estimation is employed to in-
vestigate the effects of trade policy (Berry et al., 1999; Kitano and
Ohashi, 2009). This approach is also adopted to estimate the demand
for agricultural goods (Allender and Richards, 2010) and there is a
study of the food industry regarding WTO policy related to Japan (Doi
and Ohashi, 2017). However, mainly due to data availability, the

5
K. Takechi Japan & The World Economy 55 (2020) 101024

demand estimation of differentiated agricultural goods has not been goods. Then, the mean utility level in Eq. (4) is expressed by the shares,
conducted extensively. As shown in the Data section, our unique as in the logit model:
wholesale market data and the properties of vegetables in Japan (the ˆj (=q apj + j ) = ln sj ln sj / g ln s0, j = 1, …, J ,
vegetables can be regarded as differentiated products) enable us to j

employ the discrete-choice approach. We contribute to the literature by where sj / g is the share within the group. This provides us with a demand
estimating structural parameters and conduct a rigorous assessment of function formula that we estimate as follows:
trade policy on the agricultural sector.
Consider utility in a discrete-choice model, in which a consumer in ln sj ln s0 = qj apj + j + ln sj / g .
region n chooses product j :
Because prices and within-group shares are endogenous variables, we
u nj = qj apj + j + nj, (3) need to employ an IV approach. The candidates are exogenous shocks
and variables of geographical distance, in the sense that these are
where qj is product characteristics, and a are the parameters, j is correlated with market prices through competition and cost, but are not
unobservable product characteristics, and nj is the error term. The
correlated with own product characteristics.
mean utility level, j , is shown by the terms in Eq. (3),
The demand elasticity, j , is obtained by the following:
j = qj apj + j . By making a logit assumption on the error term in
Eq. (3), the share of each good is simply expressed by: sj pj
j = = a ( 1/(1 ) + sj / g /(1 ) + sj ) pj .
exp( j ) pj sj (5)
sj = J
,
j =1
exp( j ) This is used to derive producers’ margins of each product, j .
Considering the supply side, we assume that producers compete in a
where the share of the outside good, s0 , is s0 = 1 j j . From the data,
s
monopolistically competitive fashion. In general, the main suppliers are
the actual share, ŝj , can be calculated. Hence, the estimated mean utility not farmers but agricultural cooperatives, who engage in marketing
level, ˆj , is obtained by solving the following system of equations: activities with wholesale markets and have shipping facilities, where
exp( j ) they collect vegetables from farmers. The number of agricultural co-
ŝj = , j = 1, ...J . operatives is large (there are approximately 900), but because of the
exp( j )
j
local nature of competition, it is not perceived as infinite.3 Further-
Using the definition of the mean utility yields: more, because wholesale prices are different depending on the source
prefecture, goods are considered to be differentiated based on the place
ˆj = q apj + j, j = 1, …, J .
j of origin. For example, Table 4 shows the average prices in Tokyo
market depending on origin and the average prices are dispersed based
Because we can express the mean utility level by the shares
on where the products are made. This also implies that a discrete-choice
( j = ln sj ln s0 ) and normalize the utility from outside goods to zero
( 0 = 0 ), the above equation is converted to the demand function: model for differentiated goods demand is an appropriate framework to
ln sj ln s0 = qj apj + j . As j is an unobserved product character- describe demand-side behavior.
istic, it is likely to be correlated with price pj . Thus, an endogeneity Thus, these features fit with the properties of monopolistic compe-
problem should be addressed. Applying an instrumental variable (IV) tition. Because their goods are differentiated, the cooperatives have
approach can provide us with a consistent estimator. some market power and set a markup price. The profit for product j
Because logit models are subject to restricted substitution patterns from market i is:
of goods demanded, we consider a more unrestricted framework, the ij = pj Msj ij cj Msj,
nested logit framework. Assume that there are two categories of goods,
domestic and imported goods. The utility from product j is expressed where M is the total market size and sj is the product share, as before.
by: The trade cost is shown by ij and marginal cost is cj . From the profit-
maximizing behavior, the optimal price will be:
u nj = qj apj + j + g + (1 ) nj, j = 1, …, J , (4) sj
pij = ij cj .
where g is an index of the group of goods and g is group-specific sj / pj
characteristics. The random component, nj, has an extreme value dis-
tribution. The parameter captures the correlation within a group of Then, the margins are:
pj ij cj 1
Table 4 = .
pj j (6)
Average prices in Tokyo market.
Origin prefecture Tokyo market Year 2000–2002 Combined with the demand elasticity estimates, we can derive the
Name Mean Price Observations margins over time. Hence, we can demonstrate how these margins
change in association with the changes in trade policy. Note that be-
Hokkaido 303.619 201
cause we use the data in Japanese wholesale markets, the margin for
Aomori 338.11 379
Iwate 315.004 48 imported goods may not be the margin of Chinese producers, but that of
Akita 282.765 100 importers (or intermediaries).
Yamagata 225.4 75
Fukushima 221.992 39
5. Results
Ibaraki 275.315 1966
Tochigi 312.371 141
Gunma 152.761 400 The result of a hedonic price regression is reported in Table 5. As the
Saitama 242.847 2119 import coefficient is negative, the imported goods are perceived nega-
Chiba 261.535 3041 tively from the demand side. Other characteristics, such as high quality
Tokyo 242.591 58
and large size, are positively evaluated by consumers. The inclusion of
Niigata 277.931 164
Shimane 332.533 5 the safeguard dummy controls for the direct effect on prices and, as we
Fukuoka 740 2
Imported 114.117 388
3
http://www.maff.go.jp/j/tokei/kouhyou/noukyo (in Japanese).

6
K. Takechi Japan & The World Economy 55 (2020) 101024

Table 5 Table 6
Hedonic regressions (difference-in-difference). Demand estimation.
Safeguard 0.468 0.464 OLS Logit Nested logit
(0.027) (0.027)
Imported 0.758 0.702 Price 0.0004 0.023 0.01
(0.009) (0.009) (0.0002) (0.001) (0.0001)
High quality 0.269 0.21
(0.004) (0.012)
Large size 0.056 High quality 0.407 1.003 0.177
(0.004) (0.007) (0.07) (0.02)
Constant 5.482 5.359 Large size 0.034 0.787 0.328
(0.008) (0.008) (0.007) (0.047) (0.016)
Month/year dm Yes Yes Imported 0.592 4.839 2.398
R square 0.148 0.185 (0.014) (0.195) (0.047)
Num. of obs. 108,527 108,527 Constant 7.885 1.139 4.461
(0.014) (0.298) (0.075)
Month/year dm Yes Yes Yes
R2 /Sargan Test (P value) 0.084 3.81 (0.05) 0.61 (0.44)
expected, it has a positive impact on price. The safeguard dummy is the
Num. of obs. 97,935 97,935 97,935
interaction term of the time dummy during the safeguard period and
the imported goods dummy. This interaction term captures the treat- “OLS” reports the results of simple regression results.
ment effect on the demand for imports as in the difference-in-difference “Logit” and “Nested logit” show the estimation results of the discrete choice
estimation. The safeguard dummy term is positively significant (0.464), model of demand.
which implies that those on the demand side valued imported goods
highly during the safeguard period. One of the purposes of temporary
import restrictions is to raise the market prices of imported goods. In
this sense, the safeguard policy accomplishes this policy object.
Table 6 reports the results of the demand estimations.4 We employ a
simple ordinary least square (OLS), a logit demand, and a nested logit
model estimation by Berry (1994). We use the volume of goods from
other sources, the number of other high-quality goods, and the log of
geographical distance to market from other sources as instruments.
Although geographical distance does not correlate with the product
characteristics of the good itself, it affects the product price through
competition and cost spillover effects. Thus, the geographical distance
attribute facilitates identification.
In addition to the nested logit result, we report the simple regression
and logit demand results. The coefficient of price is significantly ne-
gative in all estimations. However, because of endogeneity, the price
coefficient is the lowest in OLS. There is also a lower demand for large Fig. 4. Share-weighted margins.
size goods, which is probably due to the high price. The demanded
quantity for imports is lower than the domestic quantity. The Sargan Table 7
test statistics show that the overidentifying restriction tests are not re- Margin regressions.
jected; hence, the validity of our instruments is not rejected.
Domestic Import
Using the estimated results enables us to compute the margins. The
margins are derived from the estimates of the nested logit model and Safeguard 0.004 0.375
the margin formula from Eq. (6). Fig. 4 plots the share-weighted (0.015) (0.162)
average margins of imported and domestic goods in each month. It High quality 0.161 0.018
(0.004) (0.068)
appears that during the safeguard period, the margins of imported
Large size 0.128 0.147
goods dropped, whereas those of domestic goods remained the same. (0.01) (0.083)
This is because the share of imported goods decreased and thus the Constant 0.584 1.061
demand became more elastic. In such a case, the optimal margin is also (0.012) (0.165)
lower for imported products. Month/year dm Yes Yes
R square 0.014 0.013
To understand whether the safeguard measure influenced the mar- Num. of obs. 91,921 6014
gins significantly, we regress the margins of domestic and imported
goods on the safeguard dummy and product characteristics. The em-
pirical results in Table 7 show that there is a negative effect on im- had successfully introduced structural changes, these producers would
ported goods margins (at the 5% significance level), whereas it has no have reaped higher profits, as indicated by higher margins. Table 8
impact on domestic margins. Thus, although the safeguard was suc- reports the estimation results five years after the provisional safeguard
cessful in terms of restricting imports, the domestic industry failed to measure was terminated. The coefficients of the parameters do not
exploit this opportunity to earn larger profits. differ drastically from those before or during the safeguard.
To examine the long-run effect of safeguards, we calculate the Based on the demand estimation, we derive the average margins of
margins five years after the safeguard period. If the domestic industry domestic and imported goods in 2007. Fig. 5 plots the average margins
of domestic and imported goods in 2007. These margins are similar for
domestic and imported goods, whereas foreign products have slightly
4
The number of observations is different from the hedonic analysis because of higher margins.
data availability. Although quantity information is sometimes lacking from our One of the purposes of a temporary import restriction is to give
daily transaction data, price and product characteristics data are consistently domestic import-competing industries a chance to generate structural
available.

7
K. Takechi Japan & The World Economy 55 (2020) 101024

Table 8 of the perishable nature of agricultural products, it is not clear how

Demand estimation (long-run effect). such behavior would have affected the market outcomes; thus, the
OLS Logit Nested logit static framework may replicate actual demand-side behavior well.

Price 0.0004 0.038 0.003

(0.0002) (0.006) (0.0003)
6. Robustness
0.387
(0.018) In this section, we employ robustness checks: controlling for geo-
High quality 0.476 3.866 0.091 graphical dispersion across markets, hedonic price analysis with un-
(0.014) (0.703) (0.03)
observed product characteristics controlled for, and a random coeffi-
Large size 0.188 3.979 0.159
(0.013) (0.674) (0.03) cient demand model.
Imported 0.571 9.719 1.413 Although the analysis in the previous sections sheds light on critical
(0.022) (1.473) (0.137) aspects of the safeguard measure, it has several shortcomings. The
Constant 8.153 4.391 6.295
impact of goods attributes is captured by the hedonic regressions.
(0.024) (2.014) (0.102)
Month/year dm Yes Yes Yes
However, because we treat each unit of the sample as the same, dif-
R2 /Sargan Test (P value) 0.086 0.46 (0.5) 0.003 (0.96) ferences in geographical or other market characteristics may contribute
Num. of obs. 33,301 33,301 33,301 to the product attribute effects. That is, spatial differences in markets
may drive different reactions to the safeguard. We focus on two large
“OLS” reports the results of simple regression results. markets, Osaka and Tokyo, and conduct the same regression. Table 9
“Logit” and “Nested logit” show the estimation results of the discrete choice reports that the estimation result is qualitatively similar to those in the
model of demand.
previous section, but with a larger safeguard effect.
We conduct another robustness check for the price effect of the
safeguard. As noted, because our sample is dispersed geographically, it
is crucial to control for consumer heterogeneity. Furthermore, because
size and quality measures may be insufficient to fully capture the pro-
duct characteristics, it is also important to consider the unobservable
characteristics. An approach devised by Bajari and Benkard (2005)
enables us to identify and control for consumer heterogeneity and un-
observed characteristics by implementing a nonparametric estimation
of prices. Consider an equilibrium price function of pj = p (xj , j ) , where
x j denotes the product characteristics and j is the unobserved char-
acteristic. Using the distribution function of price, under the conditions
that x j and j are independent and the price function is strictly in-
creasing in x j (Bajari and Benkard, 2005), the unobserved characteristic
is identified given the observed product characteristics:

j = F (pj |xj ),

Fig. 5. Share-weighted margins. where F is a distribution function of pj . Estimating this non-

changes, such as efficiency improvements or quality upgrades. If do-
mestic producers were able to use the safeguard period to create a
brand image conveying the high quality of domestic goods, then the
characteristic of imported goods would have a large negative effect on
demand. However, our empirical results do not support such an idea.
The estimation results using data for 2007, when the safeguard measure
had been terminated for five years, are similar to those during the
safeguard period. In 2001, when the Ministry of Agriculture, Forestry
and Fisheries initiated a policy to structurally reform the vegetable
sector to make vegetable production and distribution more efficient
(see, e.g., Mulgan, 2005), it also adopted a “positive list” policy because
of safety concerns regarding imported vegetables. This basically in-
volved a tightening of environmental standards.5 Hence, our empirical
results imply that the demand side of the market may have appreciated
these strong safety restrictions, while structural changes may not have
occurred in the domestic industry.
A remark should be made regarding the demand-side behavior. In
our setting, because the demand side is modeled as a static framework,
it is not possible to examine dynamic behavior. In our case, the demand
side consisted of wholesalers who repeatedly purchased vegetables and
possessed strong knowledge of how to trade vegetables; it is thus pos-
sible that they may have acted in response to foreseeing the con-
sequences of the imported goods restriction. At the same time, because
5
www.mhlw.go.jp/english/topics/foodsafety/residue.
parametrically yields the unobserved factor.
Then, to estimate the demand parameter , we utilize the fact that
consumers maximize their utility by choosing a particular good j :
u (xj , pj ; ) u (xk , pk ; ) for all k j , subject to the budget constraint
B = {j|pj + c y} . We adopt a Bayesian approach to estimate the dis-
tribution of the parameter. Let p ( |Cj, x , p) be a posterior distribution,
given that consumer n 's choice of product j is Cj . Then, by using the
prior distribution ( ) and the likelihood function, such that:
1 u (xj , pj ; ) u (xk , pk ; ) for all k j
L (j|x , ) =
0 otherwise, (7)
the posterior is expressed by p ( |Cj, x , p) ( ) L (j|x , ). This enables
us to obtain taste parameters while controlling for the unobserved
factor.
As in Bajari and Benkard (2005), we first regress the prices on the
characteristics of products and obtain the residuals. Due to the com-
putational issue, we aggregate the variables used for the estimation at
the monthly level. The 47 markets are also aggregated into seven re-
gions. The regressors include the distance from origin to market
(distance jt ), monthly dummies (montht ), the index of imported goods
(import j ), and the index of high quality and large size (gradenjt and
sizenjt ): ln pnjt = a + b distance jt + c montht + d import j + e gradenjt .
+ f sizenjt
The residuals provide us with the prices conditional on product char-
acteristics. We estimate the unobserved heterogeneity by using a kernel
estimator with a normal kernel.

8
K. Takechi Japan & The World Economy 55 (2020) 101024

Table 9
Hedonic regressions (Tokyo and Osaka).
Large markets

Safeguard 0.694
(0.041)
Imported 0.848
(0.015)
High quality 0.295
(0.008)
Large size 0.008
(0.009)
Constant 5.373
(0.016)
Month/year dm Yes
R square 0.318
Num. of obs. 18,771

Fig. 6. Safeguard effect.

We assume the linear utility form and incorporate the safeguard
dummy (safeguard jt) and unobserved heterogeneity ( njt ) as the elements
of utility in market n , in month t , and from product j : u njt = y
pj + 1 safeguard jt + 2 njt . The prior distribution of the parameters is Table 10
assumed to be uniform and the posterior distribution is derived by the Random utility model.
rule described in Eq. (7). For example, when product j is chosen over k Random coefficient
and safeguard jt > safeguardkt , the bound of the safeguard parameter is:
Price 0.022
yn pnjt + 1 safeguard jt + 2 njt > yn pnkt + 1 safeguardkt + 2 nkt (0.0002)
1 > ( 2 ( nkt njt ) pnjt + pnkt )/(safeguard jt safeguardkt) .6 S.D. (price) 0.007
We generate a sequence of parameters, 1, …, s , …, S from the (0.0002)
High quality 0.257
posterior distribution, p ( |Cj, x , p) . We first draw 1 from the prior
(0.018)
distribution and then find the conditional distributions, Large size 0.099
p ( 1 |Cj, x , p , 21) and p ( 2 |Cj, x , p, 11) . Then, we draw a sequence of the (0.016)
parameters from these conditional distributions; for example, the s th Import 2.486
draw, s , is drawn from p ( 1 |Cj, x , p , 2s 1) and p ( 2 |Cj, x , p, 1s 1) . (0.041)
Because the prior distribution and the likelihood are uniform distribu- Constant 3.74
(0.075)
tions, the conditional distributions are uniform on the interval Month/year dm Yes
[ i,min, i,max ]. These boundaries associated with the effect of the safe- Num. of obs. 97,935
guard in each market n are updated by the following rule:
s s 1
1,min = max[min 1 B , max[( 2 ( kt jt ) pjt + p kt )/(safeguard jt safeguardkt ),
if safeguard jt > safeguardkt]]
where n = + µn . Thus, the coefficient of the price has a random
component, which is heterogeneous across consumers. This randomly
s s 1
1,max = min[max 1 B , min[( 2 ( kt jt ) pjt + p kt )/(safeguard jt safeguardkt),
if safeguard jt < safeguardkt]]. distributed component enables us to obtain a flexible substitution pat-
tern. Assuming that the distribution function of µn is G (µn ) yields the
We discard the initial draws and then use the subsequent 1000 random following share of good j :
coefficient draws for each region (in total 7000) to derive the dis-
tribution of the parameter.7 Fig. 6 plots the estimated kernel density of j + µ nj
sj = dG(µn ),
the coefficient of the safeguard of all markets. The distribution shows J
exp( k + µ nk )
k=0
the dispersed effect of the safeguard. There are twin peaks and the
larger one mainly corresponds to the areas including the two major where j = qj pj + j and µ nj = µn pj . We conduct the demand esti-
markets, Osaka and Tokyo, which suggests that large markets experi- mation by assuming that the price coefficient is the only random
enced a larger impact of the safeguard. This is consistent with the re- variable and using the same instrumental variables as in the nested logit
gression result of Osaka and Tokyo markets. Overall, the safeguard estimation. Table 10 reports that the price coefficient, , is 0.022 (the
policy increases market prices even after controlling for unobserved standard deviation is 0.007 ), which is similar to that of nested logit
product characteristics. demand. Thus, the discussion about how the safeguard affects the
With respect to demand estimations, in our empirical analysis, we margins remains valid.
employ a nested logit model. There is another way of obtaining a
flexible substitution structure, which is the random coefficient model
7. Conclusions
developed by Berry et al. (1995). The utility of consumer n from con-
suming good j is:
How temporary trade restrictions change market outcomes is a
u nj = qj n pj + j + j,
primary concern for policymakers. We empirically analyze daily traded
markets for vegetables to investigate the effects of a safeguard measure.
The safeguard increased the price for imported goods. However, it de-
6
As the parameter is identified in the set defined by the inequalities, these creased the margins of imported goods and furthermore it did not affect
parameters are set-identified. See Takechi and Higashida (2012) for another domestic goods margins. Although temporary import restrictions are
set-identified model applied to the agricultural sector. expected to lead to structural changes in the domestic market, the
7
For the detailed Gibbs sampling procedure, see the NBER working paper
version of Bajari and Benkard (2005).

9
K. Takechi
demand parameters did not change drastically, which suggests that
structural changes did not occur. While the effects may be different
depending on each origin prefecture, overall effects on domestic pro-
ducers are not prominent. Thus, the restrictions were simply harmful
for foreign producers.
It is often argued that vegetable import increases are caused by
“develop-and-import schemes” (Mulgan, 2005; Kuno, 2006). Under
such schemes, intermediaries develop a link with foreign farmers to
produce, say, Welsh onions, to supply Japanese markets. Hence, re-
stricting imports causes a conflict between domestic producers and
intermediaries. In addition, it should be noted that there may be win-
ners who are able to export and losers who cannot export under the low
tariff quota. Based on our data, we cannot distinguish between imports
incurring a low tariff and those with a high tariff. That is, although
some foreign producers can export without incurring high tariffs, yet
benefit from the resulting higher prices, others, who are not assigned a
quota, must pay high tariffs. As outsourcing is currently spreading in
the international economy, the same scenario may occur in other in-
dustries. Hence, safeguards cause tension between producers who
continue to produce domestically and those who shift abroad, and
tension between producers who can export without incurring high
tariffs and those who cannot.
10
Japan & The World Economy 55 (2020) 101024
References
Allender, W.J., Richards, T.J., 2010. Market diversion and market power: California eggs.
Rev. Ind. Organ. 36, 37–58.
Bagwell, K., Staiger, R.W., 1990. A theory of managed trade. Am. Econ. Rev. 42, 779–795.
Bajari, P., Benkard, C.L., 2005. Demand estimation with heterogeneous consumers and
unobserved product characteristics: a hedonic approach. J. Polit. Econ. 113,
1239–1276.
Berry, S.T., 1994. Estimating discrete-choice models of product differentiation. Rand J.
Econ. 25, 242–262.
Berry, S., Levinsohn, J., Pakes, A., 1995. Automobile prices in market equilibrium.
Econometrica 63, 841–890.
Berry, S., Levinsohn, J., Pakes, A., 1999. Voluntary export restraints on automobiles:
evaluating a trade policy. Am. Econ. Rev. 89, 400–430.
Chung, S., Lee, J., Osang, T., 2016. Did China tire safeguard save U.S. workers? Eur. Econ.
Rev. 85, 22–38.
Doi, N., Ohashi, H., 2017. Empirical analysis of the national treatment obligation under
the WTO: the case of Japanese shochu. J. Jpn. Int. Econ. 46, 43–52.
Ethier, W.J., 2002. Unilateralism in a multilateral world. Econ. J. 112, 266–292.
Kagitani, K., Hariyama, K., 2015. Safeguards and voluntary export restraints under the
World Trade Organization: the case of Japan’s vegetable trade. Jpn. World Econ. 36,
29–41.
Kitano, T., Ohashi, H., 2009. Did U.S. safeguards resuscitate Harley–Davidson in the
1980s? J. Int. Econ. 79, 186–197.
Kuno, A., 2006. An evaluation of Japan’s first safeguards actions. World Econ. 29,
763–782.
Mulgan, A.G., 2005. Japan's Agricultural Policy Regime. Routledge.
Takechi, K., Higashida, K., 2012. Firm organizational heterogeneity and market structure:
evidence from the Japanese pesticide market. Int. J. Ind. Organ. 30, 193–203.