Exchange rate volatility and Japan–U.S. commodity trade: An asymmetry analysis

• Published date: 01 November 2019
• Author: Huseyin Karamelikli,Mohsen Bahmani‐Oskooee
• DOI: http://doi.org/10.1111/twec.12856
• Date: 01 November 2019

World Econ. 2019;42:3287–3318. wileyonlinelibrary.com/journal/twec

|

3287

© 2019 John Wiley & Sons Ltd

Received: 19 November 2018

|

Revised: 13 May 2019

|

Accepted: 8 August 2019

DOI: 10.1111/twec.12856

ORIGINAL ARTICLE

Exchange rate volatility and Japan–U.S. commodity

trade: An asymmetry analysis

MohsenBahmani‐Oskooee1*

|

HuseyinKaramelikli2

1Department of Economics,The Center for Research on International Economics,The University of Wisconsin‐Milwaukee.

Milwaukee, WI, USA

2Department of Economics,Karabuk University, Karabuk, Turkey

KEYWORDS

asymmetry, commodity trade, exchange rate volatility, Japan, nonlinear ARDL, US

1

|

INTRODUCTION

When the international monetary system changed from a fixed to a relatively flexible exchange rate

system, many argued that the risk introduced by flexible rates could hurt the trade flows. Others ar-

gued to the contrary. Indeed, theoreticians developed theories that showed that exchange rate volatility

could have negative or positive effects on trade flows, depending on the degree of a trader's tolerance

to risk. While risk averse traders will trade less, risk tolerance traders will trade more.1 Indeed, the

empirical research reviewed by Bahmani‐Oskooee and Hegerty (2007) supports both views, regard-

less of the aggregation level of trade data. Trade data have been used between one country and the rest

of the world, between two countries at the aggregate bilateral level and between two countries at the

bilateral level but disaggregated by commodity (or industry).

As shown by the above review article, the literature is huge and each country today should be

considered and reviewed separately and our country of concern, Japan, is no exception. Bahmani‐

Oskooee and Ltaifa (1992) is a cross‐sectional study that found the negative impact of exchange rate

volatility on the exports of 86 countries that included Japan. Similar adverse effects were also reported

by Sauer and Bohara (2001) who relied upon a panel model using data from 91 countries over the

period 1973–93. Both studies suffer from aggregation bias since what is true about one cross‐sec-

tional unit may not be true for another unit. Indeed, when Sauer and Bohara (2001) estimated their

panel model using data from only 22 developed countries (DCs), the significantly negative effect of

exchange rate volatility disappeared. However, the panel model estimated for 22 DCs also suffers from

1 For a theoretical derivation of the link between trade flows and a measure of exchange rate volatility see Peree and Steinherr

(1989) and for arguments on the positive or negative effects of exchange rate volatility see De Grauwe (1988).

*Valuable comments of two anonymous referees and those of the associate editor are greatly appreciated. Remaining errors,

however, are our own.

3288

|

BAHMANI‐OSKOOEE ANd KARAMELIKLI

aggregation bias. The lack of a significant link between the exchange rate volatility and exports of 22

DCs could be due to an insignificant link between the two variables in most countries in the panel but

not in Japan. Of course, this hypothesis could be tested by estimating a time‐series model that draws

data only from Japan to which we turn to next.

Time‐series studies pertaining to Japan could be divided into three categories. The first in-

cludes those who have used Japan's trade flows with the rest of the world and mostly found the

significantly negative impact of exchange rate volatility on Japanese trade flows. The list in-

cludes Kenen and Rodrik (1986), Peree and Steinherr (1989), Asseery and Peel (1991), Qian

and Varangis (1994), Arize (1995), Arize and Shwiff (1998) and Poon, Choong, and Habibullah

(2005). Suspecting that these studies suffer from another aggregation bias, the second group has

used trade data at the bilateral level between Japan and another trading partner and have found

mixed results. While in the trade between Japan and the US. Aristotelous (2001) and Cushman

(1983) found negative effects of exchange rate volatility, De Vita and Abbott (2004) found posi-

tive effects. Cushman (1988) also found negative effects of exchange rate volatility on the export

volume of UK to Japan.

The bilateral studies do suffer from another aggregation bias in that the response of one industry

to exchange rate volatility could be different than the response of another industry. Thus, the third

group disaggregate trade flows between Japan and another partner by commodity. Maskus (1986)

considered US exports of 1‐digit SITC categories to Japan and found adverse effects of exchange

rate volatility. However, when Lee (1999) considered US imports of manufactured goods from

Japan, he found no significant effects. Bahmani‐Oskooee and Hegerty (2008) is the most compre-

hensive study that considers the response of trade flows of 117 industries that trade between Japan

and the U.S. They found that while in the short run many industries are affected by real yen–dollar

rate volatility, in the long‐run 25 Japanese export industries are affected negatively and 20 are af-

fected positively. As for the Japanese importing industries, 13 industries are affected negatively and

22 industries are affected positively. Most industries’ trade is unaffected by the volatility of real

yen–dollar rate.2 Could the lack of significant long‐run effects of yen–dollar volatility in most in-

dustries be due to assuming a linear adjustment of exchange rate volatility? Could we discover more

significant results if we introduce nonlinear adjustment of the volatility? Since using nonlinear

models amounts to testing for asymmetric response of trade flows to exchange rate volatility, we can

also ask whether trade flows respond to volatility changes in an asymmetric manner. Indeed,

Bahmani‐Oskooee and Aftab (2017) have argued and demonstrated that due to changes in traders’

expectations and information, their reaction to a decrease in volatility could be different than their

reaction to an increase in volatility.

Therefore, our main purpose in this paper was to consider industry‐level trade flows between Japan

and the US and investigate asymmetric response of exports and imports of 57 industries to a measure

of exchange rate volatility. As will be shown, when increased volatilities are separated from decreased

volatilities, a relatively more significant link between trade flows and exchange rate volatility is dis-

covered. To demonstrate this, we outline the models and explain the methods in Section 2. We then

report the empirical results in Section 3 that is followed by a summary in Section 4. Definition of

variables and sources of data are cited in the Appendix.

2 Following the same line of research and methods as Bahmani‐Oskooee and Hegerty (2008), Baek (2013) investigates the

response of trade flows of only 10 single‐digit industries that trade between Japan and Korea. The findings are similar to

those of Bahmani‐Oskooee and Hegerty (2008) in that only in one Japanese importing industry (i.e., manufactured goods)

and two Japanese exporting industries (i.e., chemical and related products, and machinery and transport equipment) exchange

rate volatility had significantly negative long‐run effects.

|

3289

BAHMANI‐OSKOOEE ANd KARAMELIKLI

2

|

THE MODELS AND METHODS

A common practice by all previous studies is to adopt an export and an import demand model in which

a measure of exchange rate volatility in addition to the real exchange rate itself as well as a scale vari-

able are the main determinants of export and import demands.3 As such, we begin with the following

long‐run specifications:

In (1), it is assumed that Japanese real export of commodity i,

XJP

i

, depends on the level of in-

come or economic activity in the US, YUS , the real industry i‐specific yen–dollar rate, REXi, and the

GARCH‐based measure of volatility of REXi , denoted by Vi . Using industry‐specific real exchange

rate and industry‐specific real exchange rate volatility is another point of departure from all previous

studies which must be attributed to the availability of industry‐specific prices in Japan and the U.S.

As for the expected estimates of the coefficients, if economic growth in the US is to boost Japanese

exports, an estimate of α 1 is expected to be positive and if a real depreciation of the yen against the

US dollar, defined by a decline in the real exchange rate (Appendix) is to also boost Japanese export

of commodity i, an estimate of α 2 is expected to be negative. Finally, since exchange rate volatility

could have negative or positive effects on exports, an estimate of α 3 could be negative or positive.

Following the same line of argument above, Equation (2) identifies the main determinants of the

Japanese import demand for commodity i,

MJP

i

. Following previous literature, Japanese own income

or economic activity denoted by YJP, the real bilateral exchange rate specific to industry i, REXi, and

volatility of the exchange rate are assumed to be the main determinants. Once again, we expect an

estimate of β1 to be positive, and if yen depreciation is to discourage Japanese imports, we expect an

estimate of β2 also to be positive. Finally, an estimate of β3 could be negative or positive.

If we were only interested in long‐run coefficient estimates, we would estimate only Equations

(1) and (2). However, to learn about the short‐run effects, we need to rewrite (1) and (2) in error‐cor-

rection formats. In order to obtain short‐run and long‐run estimates in one step, we adopt Pesaran's,

Shin, and Smith (2001) ARDL bounds testing approach and rely upon the following error‐correction

models:

Once (3) and (4) are estimated by OLS, coefficient estimates attached to first‐differenced variables

signify short‐run effects and estimates of θ1–θ3 normalised on θ0 in (3) and

𝜌1−𝜌3

normalised on

𝜌0

in

3 This section closely follows Bahmani‐Oskooee and Aftab (2017).

(1)

ln

X

JP

i,t

=𝛼

o

+𝛼1lnY

US

t

+𝛼2lnREX

i

,

t

+𝛼3lnV

i

,

t

+𝜀

t.

(2)

ln

M

JP

i,t

=𝛽o+𝛽1lnY

JP

t

+𝛽2lnREXi,t+𝛽3lnVi,t+𝜇t

.

(3)

Δ

ln XJP

i,t=a1+

n1

∑

j=1

a2jΔln XJP

i,t−j+

n2

∑

j=0

a3jΔln YUS

t−j+

n3

∑

j=0

a4jΔln REXi,t−j+

n4

∑

j=0

a5jΔln Vi,t−

j

+𝜃

0

ln XJP

i,t−1

+𝜃

1

ln YUS

t−1

+𝜃

2

ln REX

i,t−1

+𝜃

3

ln V

i,t−1

+𝜀

t

.

(4)

Δ

ln MJP

i,t=b1+

n5

∑

j=1

b2jΔln MJP

i,t−j+

n6

∑

j=0

b3jΔln YJP

t−j+

n7

∑

j=0

b4jΔln REXi,t−j+

n8

∑

j=0

b5jΔln Vi,t−

j

+𝜌0ln MJP

i,t−1

+𝜌1ln YJP

t−1

+𝜌2ln REXi,t−1+𝜌3ln Vi,t−1+𝜀t.