An asymmetric analysis of the J‐curve effect in the commodity trade between China and the US

• Date: 01 October 2019
• DOI: http://doi.org/10.1111/twec.12829
• Published date: 01 October 2019
• Author: Mohsen Bahmani‐Oskooee,Niloy Bose,Yun Zhang

2854

|

wileyonlinelibrary.com/journal/twec World Econ. 2019;42:2854–2899.

© 2019 John Wiley & Sons Ltd

Received: 26 January 2018

|

Revised: 23 October 2018

|

Accepted: 21 May 2019

DOI: 10.1111/twec.12829

INVITED REVIEW

An asymmetric analysis of the J‐curve effect in the

commodity trade between China and the US

MohsenBahmani‐Oskooee1

|

NiloyBose2

|

YunZhang3

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

Milwaukee, Wisconsin

2Department of Economics,Virginia Tech, Blacksburg, Virginia

3School of Finance,Shanghai Lixin University of Accounting and Finance, Shanghai, China

Funding information

Shanghai Pujiang Program and the “Shuguang Program”; Shanghai Education Development Foundation; Shanghai Municipal

Education Commission

KEYWORDS

asymmetric analysis, China, commodity trade, exchange rate, the US

1

|

INTRODUCTION

Recognising that there exist various adjustment lags and a slow pass‐through of exchange rates to prices

of the traded goods, Junz and Rhomberg (1973) as well as Magee (1973) pointed out that a country could

in fact experience deterioration in the trade balance following a currency devaluation before showing

any sign for improvements. While the first study identified adjustment lags such as recognition lags,

decision lags, production lags and delivery lags, the second study emphasised trade contracts. At the

early stage of a devaluation, goods in transit are at old prices and old exchange rates. Therefore, if trade

balance was deteriorating before a devaluation, it will keep deteriorating until new prices at new ex-

change rates are kicked in and improvement in the trade balance is realised. Many novel approaches have

been used to account for these J‐curve relationships in the data. Some have introduced lag structures on

exchange rates in reduced form models where the evidence of the J‐curve was interpreted by the pres-

ence of negative coefficients of the exchange rate at lower lags followed by positive coefficients at

higher lags. Others have relied on the error‐correction and co‐integration framework and have inter-

preted J‐curve as being present when devaluation is associated with long‐run improvements in the trade

balance, whereas the short‐run dynamics points to either insignificant or adverse trade balance effects.1

The evidence in support of the J‐curve is mixed at best. In this paper, we take the position that

there are two pervasive approaches in the existing literature which are partly responsible for the mixed

set of results. One approach relates to the methodology, and the other relates to the manner in which

data are used in the estimation. The bulk of the existing literature imposes a restrictive assumption

1 Please refer to Bahmani‐Oskooee and Hegerty (2010) for a detailed review of the literature.

|

2855

BAHMANI‐OSKOOEE Et Al.

that there exists a symmetric effect of appreciation versus depreciation of a currency as represented

by a single estimate of the elasticity of trade balance with respect to the exchange rate. Clearly, this

a priori assumption can mask the true nature of the relationship and there are good reasons to chal-

lenge this assumption. For example, Bahmani‐Oskooee and Fariditavana (2015) argue that traders’

expectations and responses are asymmetric during the episodes of appreciations versus depreciations.

Bussiere (2013) suggests that import and export prices react to exchange rate movements in an asym-

metric fashion. If responses of traded goods prices to exchange rate changes are asymmetric, then it

is natural to expect that trade balance will react to exchange rate changes in an asymmetric manner.

Arguments such as these have gained tractions, and recent research has increasingly turned attention

to the non‐linear ARDL framework as proposed by Pesaran, Shin, and Smith (2001) and Shin, Yu, and

Greenwood‐Nimmo (2014) as the main platform of deciphering the exchange rate–trade balance rela-

tionships. A recent example includes Bahmani‐Oskooee and Fariditavana (2015) who have analysed

the relationship between exchange rates and trade balances for four countries (i.e., Canada, China,

Japan and the US). In the case of China, they found short‐run and long‐run asymmetric effects. In the

long run, while yuan depreciations had no significant effect on China's trade balance, yuan apprecia-

tions were found to have adverse effects on China's trade balance.

We also suspect that the mixed results in the literature are due to the bias arising from the use of

the data that are aggregated at various levels. Selective examples of research that have used aggregate

data include Bahmani‐Oskooee (1985, 1989), Himarios (1985), Meade (1988), Moffet (1989),

Felmingham (1988), Noland (1989), Lal and Lowinger (2002), Hacker and Hatemi‐J (2003), Moura

and Da Silva (2005), Halicioglu (2007, 2008), and Bahmani‐Oskooee and Gelan (2012). In the context

of China, there are also examples of studies that have used aggregate data and have analysed the ef-

fects of exchange rate movements on the aggregate trade flow of China with the rest of the world2

(e.g., Brada, Kutan, & Zhou, 1993; Bahmani‐Oskooee & Fariditavana, 2015; Weixian, 1999; Zhang,

1998, 1999a, 1999b). Studies such as these implicitly assume a common trade dynamics that applies

to all trading partners. Clearly, this need not be the case. More importantly, aggregation of the data at

this level can potentially hide true relationships, including any evidence pertaining to the J‐curve that

is present in the trade dynamics at the bilateral trade. With this in mind, Bahmani‐Oskooee, Bose, and

Zhang (2018) has disaggregated China's trade flows with 21 of its major trading partners and evalu-

ated the relationships between the exchange rate movements and the trade balance for each trading

partner within the linear as well as the non‐linear ARDL framework.3 The results offer more insights,

and some meaningful significant long‐run asymmetric effects of real bilateral exchange rate changes

were found in the case of trade with Brazil, Malaysia, Mexico and Italy. In the case of the largest part-

ner, the United States, although the long‐run effects of the real yuan–dollar rate were not found to be

asymmetric, the yuan depreciation was found to have significantly favourable effects on China's trade

balance and yuan appreciation was found to have adverse effects.

Despite being a step forward in the right direction, the use of the bilateral level data is not free

from the aggregation bias since it assumes a common trade dynamics across all the sectors of the

economy. As a result, the analysis cannot inform about the differential output and employment effects

of devaluation across the sectors. It is therefore important that we disaggregate the data even further at

the industry level so that we gain insight into how the short and the long‐run trade dynamics of each

industry is affected by the exchange rate movements. What is true of one industry may not be true for

2 For a detailed review of studies related to China see Bahmani‐Oskooee and Zhang (2013).

3 Ahmad and Yang (2004), Narayan (2006) and Bahmani‐Oskooee and Wang (2006) are other studies that have estimated

China's bilateral trade balance models. However, none has engaged in asymmetry analysis.

2856

|

BAHMANI‐OSKOOEE Et Al.

another industry. Since different industries are subject to different trade contracts and different prices,

they face different price rigidities and this could reveal relatively more asymmetric effects of exchange

rate changes on the trade balance at industry level. In this paper, we offer such analysis within the

linear as well as the non‐linear framework for 97 commodities that are traded between China and the

United States.

The rest of the paper proceeds as follow. We describe the model and the methodology in Section

2. Our empirical results are reported in Section 3. Section 4 concludes with a summary. Data sources

and variable descriptions are reported in the Appendix A.

2

|

THE MODELS AND METHODS

The non‐linear analysis is the main focus of our paper. However, following the footstep of Bahmani‐

Oskooee and Fariditavana (2015) and Bahmani‐Oskooee et al. (2018), we begin with a linear specifi-

cation so that the results from the non‐linear model can be compared and contrasted with the linear

case4. Accordingly, we begin with the following long‐run specification:

Since commodity‐level data are reported by the US, we define the trade balance, TB, from the US

perspective and we define

TBi

as a ratio of the US imports of commodity i from China to the US

exports of commodity i to China. We follow the literature5 and include the US income (YUS), the

Chinese income (YCHN) and the real bilateral yuan–dollar rate (REX) as the three main determinants

of the trade balance for each industry. Since an increase in the US income is likely to generate more

demand for imports, we expect the estimate of b to be positive. By a similar logic, we expect the

estimate of c to be negative.6 As it is outlined in the Appendix, we define the real bilateral exchange

rate, REX, in such a manner so that a decline in REX represents a real depreciation of the dollar.

Therefore, we expect an estimate of d to be positive if dollar depreciation is to improve the US trade

balance of commodity i.

Next, we augment (1) to a standard error‐correction model (2) for assessing the short‐run effects

of exogenous variables:

Specification (2) is due to Pesaran et al. (2001) and has a distinctive advantage over a standard error‐

correction model as the specification allows for the estimation of the short‐run as well as the long‐run

4 This section closely follows Bahmani‐Oskooee and Fariditavana (2012) who used such methods to estimate bilateral trade

balance models between the US and six of her major trading partners.

(1)

LnTBi,t=a+bLnY US,t,+c LnYCHN,t+d Ln REX t+𝜀t.

5 Please refer to Rose and Yellen (2014) for a theoretical underpinning of the specification (1).

6 Note that these effects can go in the opposite direction if with an increase in income there is a tendency to substitute imports

(Bahmani‐Oskooee, 1986).

(2)

Δ

LnTBi,t=𝛼+

n

∑

j=1

𝛽t−jΔLnTBi,t−j+

n

∑

j=0

𝛿t−jΔLnYUS,t−j+

n

∑

j=0

𝛾t−jΔLnYCHN,t−j+

n

∑

j=0

𝜋t−jΔLnREXt−

j

+𝜆

1

LnTB

i,t

−

1

+𝜆

2

LnY

US,t

−

1

+𝜆

3

LnY

CHN,t

−

1

+𝜆

4

LnREX

t

−

1

+𝜇

t

.