The relation between the bilateral real exchange rate and the US-China trade balance

The relation between the bilateral real exchange

rate and the US-China trade balance

Berry Kramer

10553711

berry.afcajax@hotmail.com

MSc Economics

International Economics & Globalisation

Supervisor: John Lorié

Second Reader: Franc Klaassen

Universiteit van Amsterdam

10 July 2017

2 Statement of Originality

This document is written by Berry Kramer who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

3 Table of Contents

1. Introduction ... 4

2. Literature Review ... 7

2.1 Relation between the Real Exchange Rate and the US-China Trade Balance ... 8

2.2 The Marshall-Lerner Condition ... 8

2.3 The J-curve Effect... 9

3. Methodology ... 10

3.1 Data and Variables ... 10

3.2 The Trade Balance Model ... 13

3.3 The Autoregressive Distributed Lag – Error-Correction Model ... 15

3.3.1 ARDL First Stage: Bounds Cointegration Test ... 16

3.3.2 ARDL Second Stage: Short-Run Dynamics ... 19

4. Regression Results ... 20

4.1 Cointegration and Long-Run Coefficients ... 20

4.2 The Short-Run Effect ... 23

5. Conclusion ... 26

6. Reference list ... 29

4 1. Introduction

In a debate with Hillary Clinton at the end of 2016, Donald Trump said the following about China: “They‘re devaluing their currency, and there’s nobody in our government to fight them” (Authers, 2016). The current president of United States is pointing out that China is manipulating the Chinese renminbi at the cost of the United States. This opinion is shared by American politicians and economists who state that the renminbi is undervalued, which is seen as the reason for the large trade imbalance between the US and China (Yue & Zhang, 2013).

The undervalued renminbi could have complications for the US. Prices of Chinese export goods are cheaper when the renminbi is actually undervalued and higher demand causes the volume of US imports from China to increase. At the same time, prices of US export goods are more expensive and lower demand causes the volume of US exports to China to decrease. The higher price of US export goods and the lower price of US imports are positive for the US and are called the price effect. The increasing volume of US imports and decreasing volume of US exports are negative for the US and are called the volume effect. The volume effect is assumed to outweigh the price effect, which causes the bilateral trade balance to worsen for the US. In the case of a dollar depreciation, the volume effect is positive for the US and will more than offset the negative price effect which causes the trade balance to improve for the US. In this case, the Lerner condition is fulfilled. The Marshall-Lerner condition holds when the sum of the demand elasticities of import and export exceed one (Bahmani-Oskooee, 1991).

Koo and Zhuang (2007), Chiu, Lee and Sun (2010), Yue and Zhang (2013) and Shi and Li (2017) investigate the effect of the real exchange rate on the US-China trade balance. The real exchange rate is used as the main independent variable by previous literature and in this thesis. Since the nominal exchange rate was fixed in the period from the start of 1994 until July 21, 2005 (Morrison & Labonte, 2008), the real exchange rate has the advantage of experiencing more fluctuations (Bahmani-Oskooee & Wang, 2008). In these studies they raise the question whether a dollar depreciation is able to improve the US-China trade balance for the US and they find various outcomes. The results of Chiu et al. (2010) and Koo and Zhuang (2007) indicate that a depreciation of the US dollar against the renminbi improves the US-China trade balance for the US. However, Shi and Li (2017) suggest that the real exchange rate does not affect the bilateral trade balance. On the other hand, the results of Yue and Zhang (2013) suggest that a dollar depreciation results in a deterioration of the bilateral trade

5

balance for the US. Whether a dollar depreciation improves the bilateral trade balance for the US is further to be investigated. This thesis therefore aims to answer the following question: “Did real exchange rate movements between the US and China influence the bilateral trade balance?” This thesis will investigate the period from 1993 until 2016. Hereby contributing to the existing literature by investigating a long period in time and adding more recent data.

Figure 1 shows a negative slope of the trade balance/GDP in most of the years from 1993 until 2016. This implies that the US trade deficit with China has become relatively larger. Besides the trade deficit with China, the US has trade deficits with other trading partners. To illustrate, the US has trade deficits with the four largest trading partners behind China, namely Canada, Germany, Japan and Mexico (Chiu et al., 2010). However, the US government still blames China for the large US trade deficit and they look at the undervalued renminbi as a tool for the Chinese government to weaken the US economy. The accusation is also fueled by the increasing pressure on the economic and political relation between the US and China, since the US blames China for the loss of manufacturing jobs, the increasing debt and the growing economic insecurity (Liew, 2010).

Koo and Zhuang (2007) mention that the overall trade surplus of China is smaller than the trade surplus with the US. This implies that China has trade deficits with other countries. For example, China has a trade deficit with two of its most important trading partners, namely Japan and South Korea. According to Chiu et al. (2010), the economy of the United States, on the other hand, has been characterized by an increasingly negative overall trade balance since

-2,50% -2,00% -1,50% -1,00% -0,50% 0,00% 1992 1995 1998 2001 2004 2007 2010 2013 2016

Trade Balance Data: United States Census Bureau GDP Data: Federal Reserve Bank of St. Louis

6

1976. Since 2001, China has become the largest bilateral trade deficit for the US and the US has trade deficits with the other four largest trading partners, but also with the OPEC (Liew, 2010). Yue and Zhang (2013) name the position of the US dollar as reserve currency as reason for the US to be able to have large trade deficits. This suggests that the large trade imbalance between the US and China is not likely to be due to the bilateral exchange rate. The fact that China’s suspected undervalued currency does not cause China’s overall trade surplus to be larger than the trade surplus with the US also suggests that the exchange rate is not the reason for the large trade deficit. This argument is supported by Shi and Li (2017), who state that the large trade deficit between the US and China is not affected by the exchange rate since Chinese products rather complement than compete to US products.

Following Bahmani-Oskooee and Brooks (1999), the trade balance model is used to answer the research question. The ratio of US exports to China over US imports from China is used as a measure for the trade balance, which is the dependent variable. The use of a ratio is beneficial since the model will be in logarithmic form, enabling the coefficients of the variables to be interpreted as elasticities (Narayan, 2006). The coefficient of an independent variable then reflects the increase of the trade balance in percent when the independent variable changes by one percent. Besides that, this measure is advantageous since it is insensitive to the units of measurement and the imports and exports can be expressed in nominal terms (Bahmani-Oskooee, 1991). The main independent variable in the model is the real exchange rate, which is measured in US dollar per renminbi taking the consumer price index into account. According to Chiu et al. (2010), the sign of the coefficient of the real exchange rate is expected to be positive, since a dollar depreciation is expected to increase US exports and decrease US imports which i.e. will improve the trade balance for the US. The nominal GDP of both the US and China, measured in US dollars, are added as explanatory variables as well.

Some extra variables to control for third countries are added to the trade balance model. Koo and Zhuang (2007) add a weighted average real exchange rate between the US and Asian countries to control for the increase or decrease in trade between the US and China after a change in the Asian exchange rates. An appreciation against the US dollar decreases the prices of Asian goods relative to US goods, which decreases Chinese imports from the US. This leads to a deterioration of the bilateral trade balance for the US. Similarly to Koo and Zhuang (2007), a real effective exchange rate (REER) for both the US and China is added to the trade balance model. The REER is an index based on the exchange rates of a basket of major currencies that are linked to the country in question. The REER controls for changes in

7

trade between the US and China after changes in the exchange rates of major currencies. The paper of Andersen and Van Wincoop (2003) introduce a variable that consists of the GDP of a group of industrialized countries. This variable controls for the changes in trade after the changes in national incomes of these countries. Likewise, the GDP of all OECD member countries, measured as a volume index, is added to the balance trade model. This variable controls for the changes in trade after the changes in national incomes in the OECD countries.

The main narrative is to test whether the real exchange rate affects the trade balance. To determine whether there is a relationship between the variables, this thesis will test whether the variables are cointegrated in the long-run. Besides that, this thesis aims to determine whether there is J-curve effect by examining the long-run and short-run effect of a change in the real exchange rate on the trade balance. After a depreciation, the J-curve is characterized by a negative effect on the trade balance in the short-run followed by a positive effect on the trade balance in the long-run.

First, a test for unit roots in the variables will be performed to check whether the variables are integrated of order zero or order one and that none of the variables are integrated of the second order. Following Yue and Zhang (2013), the augmented Dickey-Fuller test will be used. Second, an error-correction model (ECM) following the autoregressive distributed lag (ARDL) approach, introduced by Pesaran, Shin and Smith (2001), is used to investigate whether the variables are cointegrated in the long-run. This is determined by a F-test which tests the joint significance of the variables, this is called the bounds cointegration test. If the F-statistic is significant, then there is a cointegration relationship. Third, an error-correction-term is added to the ARDL-ECM to estimate the ECM introduced by Engle and Granger (1987). The ECM will determine the short-run effect of the real exchange rate, which is used to investigate whether there is a J-curve effect.

In the second chapter, existing literature on this topic will be discussed. A presentation of the data and variables and the model used will be given in the third chapter and the

regression results of the empirical analyses will be given in the fourth chapter. The fifth chapter will consist of the conclusion of the results and in this chapter the research question will be answered.

2. Literature Review

This chapter focuses on the existing literature investigating the effect of the exchange rate on the trade balance. Some studies investigate the relation between the Chinese exchange rate and the US-China trade balance in particular. Some studies examine whether the

Marshall-8

Lerner condition holds between the US and other countries. Besides that, there are a few articles that investigate if there is a J-curve effect between China and the US, but also between the US and other countries.

2.1 Relation between the Real Exchange Rate and the US-China Trade Balance

Whether there is a long-run effect of the real exchange rate on the bilateral trade balance between the US and China has been investigated. For instance, the results of Koo and Zhuang (2007) and Chiu et al. (2010) suggest that a dollar depreciation would improve the bilateral trade balance for the US. Recent research on this topic by Shi and Li (2017) however suggest that the bilateral trade balance is not affected by the real exchange rate. Yue and Zhuang (2013) even indicate that a dollar depreciation results in a deterioration of the US-China trade balance for the US. These studies present varying results towards the long-run relation

between the real exchange rate and the bilateral trade balance and therefore show no clear outcome. This thesis aims to contribute to these studies, and in addition by adding the short-run effect. Besides that, more recent data is added since the most recent literature on this topic by Shi and Li (2017) investigates the period from 1993 until 2012, while this thesis includes the years 2013 until 2016 in addition.

2.2 The Marshall-Lerner Condition

The long-run effect of the exchange rate on the trade balance could also be examined by testing whether the Marshall-Lerner (M-L) condition holds. When the M-L condition is satisfied, a depreciation of a country’s currency improves the trade balance consisting of exports minus imports. More specific, a depreciation of a country’s currency results in a lower price of a country’s products, which causes the volume of exports to increase and the volume of imports to decrease. Since exports and imports are determined by the volume times the price, the trade balance will only improve when the change in volume outweighs the decrease in the price. The elasticities of exports and imports capture the response in volume after a price change. The M-L condition implies that when the sum of the elasticities of exports and imports are larger than one, then the volume effect outweighs the price effect resulting in an improvement of the trade balance in the long-run (Bahmani, Harvey & Hegerty, 2013).

The Marshall-Lerner condition has been tested between the US and its trading partners. When investigating bilateral trade between the US and three East Asian countries, Onafowora (2003) finds evidence that the Marshall-Lerner condition holds in the long-run. This indicates that a dollar depreciation will improve the bilateral trade balances to the benefit

9

of the US. Shirvani and Wilbratte (1997) find the same result when investigating bilateral trade between the US and the other G7 countries. However, the study of Dong (2017) reveals that the Marshall-Lerner condition does not hold between the US and all other G7 countries. His study suggests that the Marshall-Lerner condition holds between the US and Canada, Germany, Italy and Japan, but not between the US and France and the UK. The difference between the outcome of Shirvani and Wilbratte (1997) and Dong (2017) is most probably due to the time-spans in which the studies are conducted. Recent developments have increased the influence of capital flows on the trade balance. For example, China experienced largely increasing amounts of capital inflow in the 1990s which benefited the industrialization. The industrialization caused the export capacity in China to expand which i.e. increased the trade balance (Zhu, 2010). Besides that, they use different estimation approaches in their studies. To conclude, the outcome of these studies does not prove whether the Marshall-Lerner condition holds between the US and China specifically. However, the studies of Onafowora (2003) and Shivani and Wilbratte (1997) give an indication that a US dollar depreciation improves the bilateral trade balance for the US with a trading partner. The study of Dong (2017) gives the same indication, however in four out of six cases. This thesis could therefore contribute to these studies by investigating the US and China.

2.3 The J-curve Effect

Assuming that the M-L condition holds, the trade balance will improve in the long-run. However, the M-L condition does not include the short-run effect. The change in the exchange rate does not immediately result in the change of trade volumes, but it does affect the prices relatively fast. Since volumes are price inelastic in the short-run, the trade balance does not improve immediately before it will improve in the long-run. The J-curve effect suggests that a depreciation initially leads to a deterioration of the trade balance in the short-run followed by an improvement of the trade balance in the long-short-run (Lal & Lowinger, 2002).

When investigating the J-curve effect, both the short-run and long-run effect of the real exchange rate on the trade balance are examined. Narayan (2006), Bahmani-Oskooee and Wang (2006) and Bahmani-Oskooee and Wang (2008) investigate whether there is a J-curve effect between the US and China. All three studies find evidence that a dollar depreciation improves the bilateral trade balance for the US in the long-run. The difference however is revealed in the results on the short-run effect. Narayan (2006) and Bahmani-Oskooee and Wang (2006) find a positive short-run effect of a dollar depreciation on the trade balance,

10

while Bahmani-Oskooee and Wang (2008) find a negative short-run effect, and thus evidence for a J-curve effect, in 22 out of the 88 industries investigated.

Besides Bahmani-Oskooee and Wang (2008), Onafowora (2003) also finds a J-curve effect, he however does not investigate the US and China in particular. Onafowora (2003) finds a J-curve effect within two years between the US and three Asian countries. When investigating the J-curve effect between the US and six trading partners, Bahmani-Oskooee and Brooks (1999) does however find no statistical evidence for a short-run effect and thus no evidence for a J-curve effect. Concluding, studies regarding the J-curve effect of the US with trading partners other than China show varying outcomes.

The J-curve effect between the US and China has not been investigated since Bahmani-Oskooee and Wang (2008) have investigated these two countries over the period from 1978 until 2002, using commodity data. In 2006, Bahmani-Oskooee and Wang and Narayan have investigated the J-curve effect using bilateral trade data. They investigated the periods from 1983 until 2002 and 1979 until 2002, respectively. Since the J-curve effect between the US and China has not been investigated over the period from 2003 until 2016, this thesis aims to contribute to the literature by investigating this recent period in time.

3. Methodology

This section will discuss the empirical model and its implementation. First, the data will be presented. Second, the trade balance model and its variables are explained and the expected effects of the variables will be discussed. Third, a detailed description of the autoregressive distributed lag model (ARDL) is given and an explanation on how this model is capable of estimating the effect of the real exchange rate on the bilateral trade balance of the US with China.

3.1 Data and Variables

This section presents the data and variables which are used in the empirical analysis. The variables used are on a quarterly base. The period from 1993 until 2016 is investigated, therefore the dataset consists of a total of 96 observations. A presentation of the data will be given, including the name of the database where the data is obtained.

The dependent variable used in the empirical analysis is the bilateral trade balance between the US and China (TB). Following the trade balance model, the trade balance is defined as a ratio of US exports to China divided by US imports from China, in goods. The

11

data for both the exports and imports between the US and China are obtained from the United States Census Bureau. Figure 2 presents the trade balance ratio.

Besides the trade balance ratio, figure 2 also presents the real exchange rate between the US and China (RER). The RER is the main independent variable used in the empirical analysis. The following equation is used to calculate the RER:

𝑅𝐸𝑅𝑈𝑆,𝐶 ₌ 𝐸𝑅𝑈𝑆,𝐶 ∗ 𝐶𝑃𝐼𝐶 𝐶𝑃𝐼𝑈𝑆

where 𝐸𝑅𝑈𝑆,𝐶 _{is the nominal exchange rate expressed in US dollar per renminbi. 𝐶𝑃𝐼}𝐶 _{is the}

consumer price index of China and 𝐶𝑃𝐼𝑈𝑆 _{is the consumer price index of the US, where 2010}

is the reference year for both consumer price indices. The data for the nominal exchange rate, which is a period average of all daily exchange rates in a quarter, is obtained from the IMF IFS Database. The data for the consumer price indices of the US and China are obtained from the OECD Database.

When looking at figure 2, the trade balance ratio and the real exchange rate seem to move together. The trend of the trade balance ratio is however not as clear as the trend of the real exchange rate and the figure does not include the same scale on the vertical axes.

0,08 0,09 0,1 0,11 0,12 0,13 0,14 0,15 0,16 0,17 0,18 0,1 0,15 0,2 0,25 0,3 0,35 RER TB

Figure 2: Trade Balance Ratio and Real Exchange Rate,

Quarterly

12

Although, it does give an indication that the trade balance and the real exchange rate might be cointegrated.

The nominal gross domestic product of both the US and China are added to model as well. The data for the US GDP is measured in US dollars and is obtained from the OECD Database. The data for the GDP of China is measured in renminbi and is obtained from the National Bureau of Statistics of China. The GDP of China is converted into US dollars using the nominal exchange rate. Figure 3 presents the nominal GDP for both the US and China.

The GDP of both the US and China follow the well-known upward sloping trend. The GDP of China can be described as a jigsaw pattern. The GDP per quarter drops between the fourth quarter of the current year and the first quarter of the next year, while the quarterly GDP of China experiences a year on year growth larger than 6% for every quarter. However, no article or database gives an explanation about the jigsaw pattern. When observing figure 2, the trade balance ratio is mostly at its peak in the first quarter of the year. The jigsaw pattern of the GDP of China and the trade balance may therefore be related, which is to be

determined by the regression results.

Multiple additional control variables for third countries are added to the trade balance model. Similar to the weighted average exchange rate, which is included in the trade balance model of Koo and Zhuang (2007) a real effective exchange rate (REER) is added to the trade balance model. To balance the model, a REER for the US and a REER for China are added.

$0 $500 $1.000 $1.500 $2.000 $2.500 $3.000 $3.500 $4.000 $4.500 $5.000 Bil lio n s

Figure 3: Quarterly Nominal GDP

13

The REER is measured as an index, where 2010 is the reference year for both indices. The data of the real effective exchange rates are obtained from the IMF IFS Database.

Literature using gravity-type models, e.g. Andersen and Van Wincoop (2003), include a variable which consists of the GDP of a group of 20 countries. In a similar way, the trade balance model will control for the GDP of a group of third countries by adding the nominal GDP of all OECD countries. The GDP of all OECD countries is measured as a volume index where 2010 is the reference year, based on nominal volume estimates using current prices. The data for the GDP of all OECD countries is obtained from the OECD Database.

3.2 The Trade Balance Model

The trade balance model, used by Bahmani-Oskooee and Brooks (1999), is used as the baseline model in the empirical analysis. The trade balance in this model is the ratio of exports over imports. Therefore, the model can be estimated in logarithmic form (Bahmani-Oskooee and Brooks, 1999) and the coefficients can be interpreted as elasticities (Narayan, 2006). Besides that, a ratio is used since it is insensitive to the units of measurement and exports and imports can be measured in nominal terms (Bahmani-Oskooee, 1991). The original trade balance model is complemented with some additional variables to control for the effect of third countries. The trade balance model takes the following form:

𝑙𝑛𝑇𝐵_𝑡𝑈𝑆,𝐶 = 𝛼 + 𝛽 𝑙𝑛𝑅𝐸𝑅_𝑡𝑈𝑆,𝐶+ 𝛿 𝑙𝑛𝐺𝐷𝑃_𝑡𝑈𝑆 + 𝛾 𝑙𝑛𝐺𝐷𝑃_𝑡𝐶+ 𝜃 𝑙𝑛𝑅𝐸𝐸𝑅_𝑡𝑈𝑆+ 𝜑 𝑙𝑛𝑅𝐸𝐸𝑅_𝑡𝐶 (1) + 𝜔 𝑙𝑛𝐺𝐷𝑃_𝑡𝑂𝐸𝐶𝐷+ 𝜀_𝑡

where 𝑇𝐵𝑈𝑆,𝐶 is the trade balance, which consists of US exports to China divided by US imports from China. 𝐺𝐷𝑃𝑈𝑆 and 𝐺𝐷𝑃𝐶 are the nominal gross domestic products of respectively the US and China, measured in US dollars. 𝑅𝐸𝑅𝑈𝑆,𝐶 is the real exchange rate measured in US dollar per renminbi, which implies that a depreciation increases the RER. 𝑅𝐸𝐸𝑅𝑈𝑆 and 𝑅𝐸𝐸𝑅𝐶 are the real effective exchange rates of respectively the US and China, both measured as an index where 2010 is the reference year. 𝐺𝐷𝑃𝑂𝐸𝐶𝐷 is the gross domestic product of all OECD member countries which is measured as a volume index where 2010 is the reference year.

The trade balance model also controls for the effect of third countries on the US-China trade balance. Similar to the weighted average real exchange rate of Asian countries added by Koo and Zhuang (2007), the real effective exchange rates of both the US and China are added. The real effective exchange rate is a trade-weighted average index based on the

14

bilateral exchange rates of a basket of major currencies linked to the country in question, adjusted for consumer price indices. The REER is very useful to control for the effect of third countries since it expresses the changes in the exchange rates of considerable trading partners. To control for the growth in production of third countries, the GDP of all OECD member countries is added to the model. The same has been done by studies using gravity type models. For example, the study of Andersen and Van Wincoop (2003) uses the GDP of a group of 20 industrialized countries to control for the effect of all third countries.

The expected sign of 𝛽 is positive according to Bahmani-Oskooee and Wang (2008). A real depreciation implies that the RER variable will increase, which will encourage US exports to China and discourage US imports from China. The trade balance will therefore improve. According to Chiu et al. (2010), the expected sign of 𝛿 is negative since an increase in the US GDP raises US imports from China and i.e. will deteriorate the trade balance. However, the expected sign could also be positive if the increase of the US GDP is mostly due to the increase in the production of goods that can substitute imported goods. Therefore, relatively less products are imported from China which i.e. will improve the trade balance. The sign of 𝛾 is expected to be positive since an increase in China’s GDP will raise Chinese imports from the US and i.e. will improve the trade balance. On the other hand, the sign of 𝛾 is expected to be negative if the increase in the Chinese GDP is mostly due to the increase in the production of import substitutes. This will result in relatively less exports towards the US, which i.e. will deteriorate the trade balance. The effect of an increase in the GDP of both the US and China is therefore ambiguous.

The sign of the coefficient 𝜃 of the 𝑅𝐸𝐸𝑅𝑈𝑆 is expected to be positive. The REER is an index in which a decrease of the index number indicates a depreciation against other currencies. A dollar depreciation against a large group of major currencies will improve the competitiveness of the US compared to its trading partners and i.e. will increase the demand for US products. Assuming that the production of the US is rigid in the short-run, US exports towards China will be replaced by exports towards other trading partners. This will i.e. deteriorate the trade balance for the US since a smaller volume is exported towards China. The REER of the US consists for 21.7% of the bilateral exchange rate with China, according to the REER trade weights from the Bank for International Settlements (BIS). If a drop in the REER is caused by a dollar depreciation against the renminbi, then the expected sign of 𝜃 will be negative since the effect will be the same as the effect of the RER. The sign of the

coefficient 𝜑 of the 𝑅𝐸𝐸𝑅𝐶 is expected to be negative. A decrease of the REER of China indicates a depreciation against other currencies. This will increase the competitiveness of

15

China compared to its trading partners and will i.e. increase the demand for Chinese products. Assuming that the production of China is rigid in the short-run, Chinese exports towards the US are replaced by exports towards other countries. Therefore, the trade balance will improve for the US since a smaller volume is imported from China. However, according to BIS the REER of China consists for 17.8% of the bilateral exchange rate with the US. If a drop in the REER is caused by an appreciation of the dollar against the renminbi, then the trade balance will deteriorate for the US. The expected sign of 𝜑 will then be positive.

The expected sign of the coefficient 𝜔 of the 𝐺𝐷𝑃𝑂𝐸𝐶𝐷 _{is ambiguous. An increasing}

index number of the GDP of all OECD member countries indicates that the GDP of these countries increases. If this implies that the income of these countries increases, then this will encourage US and Chinese exports towards these countries. The increase in exports of both countries could result in the increase in GDP in the two countries. Besides that, the 𝐺𝐷𝑃𝑂𝐸𝐶𝐷

is added as indicator for ‘World GDP’, therefore an increase in this variable indicates that the GDP in all countries will increase. As mentioned before, the effect of an increase in the GDP of the US and China is expected to be either positive or negative. The GDP of the US is a large part of the GDP of the OECD. Therefore, it is likely that the GDP of the OECD moves together with the GDP of the US, this implies that 𝜔 will have the same sign as 𝛿.

The trade balance model is the main regression. This thesis aims to determine a long-run cointegration relationship among the variables and investigate whether there is a J-curve effect. Therefore, equation (1) is adjusted into an error-correction model (ECM) to include both long-run and short-run dynamics, using the autoregressive distributed lag (ARDL) approach. The following subsection will give a detailed description how the ECM following the ARDL approach will test for cointegration and determines the short-run dynamics.

3.3 The Autoregressive Distributed Lag – Error-Correction Model

The trade balance model is a limited model since it does not include lagged values of the variables. The trade balance model is not capable of determining whether the variables are cointegrated since it is not able to address stationarity accurately. Therefore, a cointegration model is used to examine whether the trade balance and the real exchange rate are

cointegrated. Following Pesaran et al. (2001), an error-correction model following the autoregressive distributed lag (ARDL) approach is used. The ARDL approach enables the error-correction model to determine whether there is a long-run cointegration relationship among the variables and i.e. determine the long-run effect. Besides the long-run effect, the

16

ARDL approach also determines the short-run effect to examine whether there is a J-curve effect.

The ARDL approach is divided into two stages. The first stage is the bounds cointegration test (Adom, Akoena & Bekoe, 2012). This test estimates whether there is a long-run cointegration relationship, which implies that the variables are cointegrated in the long-run. If the variables appear to be cointegrated, then the error-correction model including an error-correction term, introduced by Engle and Granger (1987), is constructed and

estimated in the second stage (Ma & Tian, 2010). The main focus of the second stage is to determine the short-run dynamics to investigate whether there is a J-curve effect.

3.3.1 ARDL First Stage: Bounds Cointegration Test

Whether the variables are cointegrated in the long-run is tested in the first stage of the ARDL approach. To do this, the autoregressive distributed lag - error-correction model (ARDL-ECM) of Pesaran et al. (2001) is used. The ARDL-ECM includes the same dependent variable and the same independent variables as in equation (1). Instead of using the level value of the variables, the ARDL- ECM uses the first difference of the variables. The first difference measures the change in the variable, which make it possible to examine whether a change in the exchange rate results in a change in the trade balance. The ARDL approach examines the effect of past values on current values, therefore the first differences of the independent variables are lagged. The lagged values of the first differences are used to determine whether a change in the variable of a previous period results in a change in the current period. The first differences are lagged for one or more periods or not lagged, which is to be determined. The ARDL-ECM also includes the lagged-level variables to replace the error-correction term that is originally included in the error-correction model (ECM) of Engle and Granger (1987). Pesaran et al. (2001) modified the original ECM to form the ARDL-ECM. Following Pesaran et al. (2001), the ARDL-ECM takes the following form:

∆𝑙𝑛𝑇𝐵_𝑡𝑈𝑆,𝐶 = 𝛼 + ∑ 𝜋_𝑖∆𝑙𝑛𝑇𝐵_𝑡−𝑖𝑈𝑆,𝐶 𝑛 𝑖=1 + ∑ 𝛽_𝑖∆𝑙𝑛𝑅𝐸𝑅_𝑡−𝑖𝑈𝑆,𝐶 𝑛 𝑖=0 + ∑ 𝛿_𝑖∆𝑙𝑛𝐺𝐷𝑃_𝑡−𝑖𝑈𝑆 (2) 𝑛 𝑖=0 + ∑ 𝛾_𝑖∆𝑙𝑛𝐺𝐷𝑃_𝑡−𝑖𝐶 𝑛 𝑖=0 + ∑ 𝜃_𝑖∆𝑙𝑛𝑅𝐸𝐸𝑅_𝑡−𝑖𝑈𝑆 𝑛 𝑖=0 + ∑ 𝜑_𝑖∆𝑙𝑛𝑅𝐸𝐸𝑅_𝑡−𝑖𝐶 𝑛 𝑖=0 + ∑ 𝜔_𝑖∆𝑙𝑛𝐺𝐷𝑃_𝑡−𝑖𝑂𝐸𝐶𝐷 𝑛 𝑖=0 + 𝜆₁𝑙𝑛𝑇𝐵_𝑡−1𝑈𝑆,𝐶+ 𝜆₂𝑙𝑛𝑅𝐸𝑅_𝑡−1𝑈𝑆,𝐶+ 𝜆₃𝑙𝑛𝐺𝐷𝑃_𝑡−1𝑈𝑆 + 𝜆₄𝑙𝑛𝐺𝐷𝑃_𝑡−1𝐶 + 𝜆₅𝑙𝑛𝑅𝐸𝐸𝑅_𝑡−1𝑈𝑆 + 𝜆₆𝑙𝑛𝑅𝐸𝐸𝑅_𝑡−1𝐶 + 𝜆₇𝑙𝑛𝐺𝐷𝑃_𝑡−1𝑂𝐸𝐶𝐷+ 𝜇_𝑡

17

The proper test to investigate for cointegration is the standard F-test. The F-test is used to test for joint significance of the lagged-level variables in equation (2) (Bahmani-Oskooee & Wang, 2008). Following Bahmani-Oskooee and Wang (2006), the following hypothesis is tested:

𝐻₀: 𝜆₁ = 𝜆₂ = 𝜆₃ = 𝜆₄ = 𝜆₅ = 𝜆₆ = 𝜆₇ = 0 𝐻₁: 𝜆₁ ≠ 𝜆₂ ≠ 𝜆₃ ≠ 𝜆₄ ≠ 𝜆₅ ≠ 𝜆₆ ≠ 𝜆₇ ≠ 0

where the null hypothesis implies that there is no existence of a long-run relationship. If the null hypothesis is rejected, then this would suggest that there is a cointegration relationship between the variables.

The asymptotic distribution of the F-statistic is non-standard, regardless of whether the variables are integrated of order zero, I(0), or order one, I(1) (Bahmani-Oskooee & Brooks, 1999). The variables are therefore expected to be either all I(0) or I(1). This is the main advantage of the model of Pesaran et al. (2001), since no unit root pre-testing is needed. Besides that, the ARDL approach is more robust and it is more accurate for a small sample size (Liam & Ma, 2010), which is the case in this study. Pesaran et al. (2001) provide two tables of critical values for the F-test. The first table assumes that all the variables are I(0), while the second table expects all variables to be I(1). The critical values assuming that all the variables are I(0) are called the lower bound values and the critical values assuming that the variables are I(1) are called the upper bound values (Ma & Tian, 2010). If the F-statistic turns out to be below the lower bound critical value, then the null hypothesis of no cointegration cannot be rejected. If the F-statistic lies between the lower and upper bound critical value, then the result is indecisive. If the F-statistic lies above the upper bound critical value, then the null hypothesis should be rejected indicating that the variables are cointegrated. The ECM used in the second stage includes an error-correction term, which will be used as an

alternative method to indicate cointegration among the variables. The error-correction term will determine whether the variables are cointegrated in the case of an indecisive result (Bahmani-Oskooee & Brooks, 1999).

According to Bahmani-Oskooee and Brooks (1999), the F-test is sensitive to the lag order. Therefore, following Bahmani-Oskooee and Wang (2008) Akaike’s information

18

variables in equation (2). Following Bahmani-Oskooee and Brooks (1999), the number of lags will be limited to a maximum of four lags since the data is quarterly.

As stated before, there is no need for unit root pre-testing since the F-statistic has a non-standard distribution regardless of whether the variables are I(0) or I(1) (Bahmani-Oskooee & Brooks, 1999). However, Ma & Tian (2010) and Baek and Chi (2013) use the ARDL-ECM and they choose to perform a unit root test to test for non-stationarity of the variables. Since the F-test is not valid when one of the variables is integrated of the second order, I(2), it is necessary to perform a unit root test (Baek & Chi, 2013). Besides that, a unit root test has to be performed eventually since all variables have to non-stationary, I(1), in the second stage ECM (Bahmani-Oskooee & Wang, 2008). Therefore, an augmented Dickey-Fuller (ADF) test is performed to test whether the variables are indeed I(0) or I(1). Following Yue and Zhang (2013), the ADF procedure uses the following equation:

∆𝑋_𝑡= 𝛼 + 𝛽𝑋_𝑡−1+ ∑ 𝛿_𝑖∆𝑋_𝑡−𝑖

𝑛

𝑖=1

+ 𝜀_𝑡 (3)

where 𝑋_𝑡 is the variable of interest and 𝛽 is the explanatory coefficient. The lagged first difference values are included in equation (3) to filter any serial correlation in ∆𝑋𝑡 (Yue &

Zhang, 2013). The null hypothesis in the ADF test assumes that the variable is non-stationary and thus I(1). The null hypothesis is rejected if the coefficient 𝛽 is significantly different from zero, which indicates that the variable is I(0). The variables are expected to be non-stationary at its level value, which implies that the null hypothesis is not rejected. The first difference of the variables is expected to be stationary, so in this case the null hypothesis is rejected (Yue and Zhang, 2013). Again, AIC is used to estimate the optimal number of lags for each variables’ first differences in equation (3).

If the variables are even I(0) or I(1) and none of these variables are thus I(2), then the F-test can be performed. Once the outcome of the F-test indicates that the variables are cointegrated, then the long-run effect of the independent variables can be estimated using the coefficients of the lagged-level variables in equation (2). The long-run effect of the RER on the trade balance is estimated by 𝜆₂/𝜆₁ (Bahmani-Oskooee & Wang, 2008). This estimate determines the change of the trade balance in percentage after a change in the RER by one percent. The long-run effect of the other variables on the trade balance can be estimated by dividing the coefficients 𝜆₃− 𝜆₇ individually by 𝜆₁.

19

Once a long-run cointegration relationship has been established in the first stage, then the error correction model of Engle and Granger (1987), including an error-correction term, is constructed and estimated in the second stage. The second stage mainly focuses on the short-run dynamics to determine whether there is a J-curve effect.

3.3.2 ARDL Second Stage: Short-Run Dynamics

The ECM of Engle and Granger (1987) is constructed and estimated in the second stage to determine the short-run dynamics. Pesaran et al. (2001) modified the original ECM of Engle and Granger (1987) by specifying the error-correction term. They solve (1) for 𝜀_𝑡−1 to form a lagged error-correction term, which include the lagged level of the variables (Bahmani-Oskooee & Wang, 2008). Following Pesaran et al. (2001), the error-correction term is as follows:

𝐸𝐶𝑇_𝑡−1 = 𝑙𝑛𝑇𝐵_𝑡−1𝑈𝑆,𝐶− 𝛼 − 𝛽 𝑙𝑛𝑅𝐸𝑅_𝑡−1𝑈𝑆,𝐶− 𝛿 𝑙𝑛𝐺𝐷𝑃_𝑡−1𝑈𝑆 − 𝛾 𝑙𝑛𝐺𝐷𝑃_𝑡−1𝐶 − 𝜃 𝑙𝑛𝑅𝐸𝐸𝑅_𝑡−1𝑈𝑆 (4) − 𝜑 𝑙𝑛𝑅𝐸𝐸𝑅_𝑡−1𝐶 − 𝜔 𝑙𝑛𝐺𝐷𝑃_𝑡−1𝑂𝐸𝐶𝐷

Pesaran et al. (2001) use the specification of the error-correction term to form the ARDL-ECM in (2). When substituting (4) into (2), then the ARDL-ECM that was originally introduced by Engle and Granger (1987) is constructed. Following Engle and Granger (1987), the ECM takes the following form:

∆𝑙𝑛𝑇𝐵_𝑡𝑈𝑆,𝐶 = 𝛼 + ∑ 𝜋_𝑖∆𝑙𝑛𝑇𝐵_𝑡−𝑖𝑈𝑆,𝐶 𝑛 𝑖=1 + ∑ 𝛽_𝑖∆𝑙𝑛𝑅𝐸𝑅_𝑡−𝑖𝑈𝑆,𝐶 𝑛 𝑖=0 + ∑ 𝛿_𝑖∆𝑙𝑛𝐺𝐷𝑃_𝑡−𝑖𝑈𝑆 (5) 𝑛 𝑖=0 + ∑ 𝛾_𝑖∆𝑙𝑛𝐺𝐷𝑃_𝑡−𝑖𝐶 𝑛 𝑖=0 + ∑ 𝜃_𝑖∆𝑙𝑛𝑅𝐸𝐸𝑅_𝑡−𝑖𝑈𝑆 𝑛 𝑖=0 + ∑ 𝜑_𝑖∆𝑙𝑛𝑅𝐸𝐸𝑅_𝑡−𝑖𝐶 𝑛 𝑖=0 + ∑ 𝜔_𝑖∆𝑙𝑛𝐺𝐷𝑃_𝑡−𝑖𝑂𝐸𝐶𝐷 𝑛 𝑖=0 + 𝜆𝐸𝐶𝑇_𝑡−1+ 𝜇_𝑡

The optimal number of lags is estimated by AIC. The error-correction term is calculated by equation (4) to form a new time series (Bahmani-Oskooee & Wang, 2006). Therefore, the coefficients of the lagged-level variables and the constant in (1) are first estimated by OLS (Chang, Chiang & Jiang, 2012). The coefficient of 𝐸𝐶𝑇𝑡−1, 𝜆, should be

20

Wang, 2008). To substantiate, 𝜆 determines the speed of adjustment to equilibrium after a shock (Narayan, 2006). Since the 𝐸𝐶𝑇_𝑡−1 in equation (4) increases when 𝑇𝐵_𝑡−1 increases, a significantly negative coefficient of 𝐸𝐶𝑇_𝑡−1 would imply that the trade balance will return to its equilibrium value. For example, a coefficient of -0.5 would indicate that after two quarters the trade balance is exactly back at its equilibrium value. The error-correction term is

therefore an alternative method to determine whether there is a long-run cointegration relationship (Bahmani-Oskooee & Brooks, 1999).

Once a long-run cointegration relationship is determined in the first stage, then the ECM estimates the short-run effect (Bahmani-Oskooee & Wang, 2008). A possible J-curve pattern will be revealed by the sign and significance of the coefficients of 𝛽_𝑖 in (5), which include the short-run effect (Bahmani-Oskooee & Wang, 2008). There is a J-curve effect when there are significant negative coefficients followed by significant positive coefficients for the lagged values of the first difference of the RER (Bahmani-Oskooee & Brooks, 1999). In this case, a negative short-run effect is followed by a positive long-run effect.

4. Regression Results

In this section, the steps that are described in the methodology will be executed. The results of the regressions will be tabulated and the coefficients will be interpreted on their size and significance. Hereby, the main aim is to investigate whether the bilateral real exchange rate influences the US-China trade balance. In addition, this thesis focuses on both the long-run and short-run effect to investigate whether there is a J-curve effect. This section is divided into two stages, according to the ARDL approach by Pesaran et al. (2001). The first stage will investigate whether the variables are cointegrated in the long-run using the ARDL-ECM by Pesaran et al. (2001). Besides that, the long-run coefficients are estimated. Once cointegration among the variables is determined, the ECM by Engle and Granger (1987) is used to

determine the short-run effect.

4.1 Cointegration and Long-Run Coefficients

In the first stage of the ARDL approach by Pesaran et al. (2001), the ARDL-ECM is estimated. An F-test will be conducted to test for the joint significance of the lagged-level variables, which will determine whether there is a long-run cointegration relationship between the variables. This is called the bounds cointegration test. Subsequently, the long-run effects are determined by the size and the significance of the coefficients of the lagged-level

21

Before the regression is performed, the order of integration of all variables has to be estimated first by the ADF test. The ADF test will ensure that none of the variables is I(2). The variables are I(1) when the level values of the variables are non-stationary and the first differences of the variables are stationary. In effect, the null is not rejected for the level values and rejected for the first differences. The optimal number of lags for the first differences is determined by AIC. The number of lags is (4,1,3,4,1,2,3). This implies that the lag length is 4 lags for the trade balance, 1 lag for the RER, 3 lags for the US GDP, 4 lags for the GDP of China, 1 lag for the US REER, 2 lags for the REER of China and 3 lags for the OECD GDP. The results of the augmented Dickey-Fuller test are presented in table 1.

Table 1. Augmented Dickey-Fuller Test

Variable Levels First Difference Lags Order of

Integration Constant Constant and

Trend Constant Constant and Trend 𝒍𝒏𝑻𝑩𝑼𝑺,𝑪 -1.652 -2.360 -4.879*** -5.158*** 4 I(1) 𝒍𝒏𝑹𝑬𝑹𝑼𝑺,𝑪 -1.151 -2.369 -7.383*** -7.388*** 1 I(1) 𝒍𝒏𝑮𝑫𝑷𝑼𝑺 -1.745 -1.640 -3.017** -3.345* 3 I(1) 𝒍𝒏𝑮𝑫𝑷𝑪 -1.422 -1.582 -4.045*** -3.956** 4 I(1) 𝒍𝒏𝑹𝑬𝑬𝑹𝑼𝑺 -0.988 -2.649 -7.232*** -7.263*** 1 I(1) 𝒍𝒏𝑹𝑬𝑬𝑹𝑪 -1.379 -1.273 -5.222*** -5.228*** 2 I(1) 𝒍𝒏𝑮𝑫𝑷𝑶𝑬𝑪𝑫 -1.688 -1.781 -3.809*** -4.031** 3 I(1)

*** denotes significance at the 1% level. ** denotes significance at the 5% level. * denotes significance at the 10% level.

The results in table 1 suggest that all variables are non-stationary at levels since the null hypothesis of non-stationarity cannot be rejected. Besides that, the first differences of all variables are significantly stationary. The first difference for the US GDP is stationary at 5% significance when no trend is included and stationary at 10% with a trend. The first difference of the GDP of China is significant at 5% when a trend is included, while it is significant at 1% when no trend is included. The remaining first differences are stationary at 1% significance. The outcome of the ADF test therefore indicate that all variables are integrated of the first order and are thus I(1).

Since the order of integration of all variables meets the condition for the bounds cointegration test, the next step is to use the ARDL-ECM to test for a long-run cointegration relationship between the variables. AIC has determined that the optimal number of lags is (3,3,1,3,4,0,0). The F-test is used to test the joint significance of the lagged-level variables. The F-statistic will then determine whether the variables are cointegrated, depending on the

22

critical values provided by Pesaran et al. (2001). If the F-statistic exceeds the upper bound critical value, then the variables are cointegrated. If the F-statistic falls below the lower bound critical value, no cointegration relationship is established. Cointegration among the variables is indecisive when the F-statistic lies between the lower and upper bound critical value. The t-value of the coefficients determines whether they are significant. The resulting F-statistic in the bounds cointegration test is presented in table 2.

Table 2. Bounds Cointegration Test – (3,3,1,3,4,0,0)

F-Statistic Significance level Critical values

I(0) I(1) 4.904*** 1% 5% 10% 3.15 2.45 2.12 4.43 3.61 3.23

*** denotes significance at the 1% level.

The F-statistic of joint significance of the lagged-level variables exceeds the upper bound critical value at the 1% significance level. The null hypothesis of no levels cointegration relationship is thus rejected. Therefore, the result of the F-test in table 2 suggests that there is long-run cointegration relationship between the variables.

The F-test revealed that the variables are cointegrated in the long-run, so now the long-run effects of the independent variables can be calculated. The long-run effect of the RER on the trade balance is determined by 𝜆₂/𝜆₁. The long-run effects of the other independent variables are calculated similarly by dividing the coefficients 𝜆₃− 𝜆₇

individually by 𝜆₁. The values of 𝜆 are an elasticity and therefore determine the effect on the trade balance in percent when the independent variable changes with 1%. If the long-run effect of the RER is significantly positive, then a dollar depreciation will improve the US-China trade balance for the US. This would also indicate that the M-L condition holds between the US and China in the long-run. The long-run effects of the independent variables are presented in table 3.

23 Table 3. The Long-Run Effects

Variable Coefficient Absolute

t-value 𝒍𝒏𝑹𝑬𝑹𝑼𝑺,𝑪 3.59** 2.35 𝒍𝒏𝑮𝑫𝑷𝑼𝑺 _{6.44*** 2.83} 𝒍𝒏𝑮𝑫𝑷𝑪 -1.92*** 5.92 𝒍𝒏𝑹𝑬𝑬𝑹𝑼𝑺 _{-0.97 0.74} 𝒍𝒏𝑹𝑬𝑬𝑹𝑪 0.38 0.37 𝒍𝒏𝑮𝑫𝑷𝑶𝑬𝑪𝑫 _{-2.24 0.67}

*** denotes significance at the 1% level. ** denotes significance at the 5% level.

The long-run effect of the RER is given by 3.59, which is significant at the 5% level. It has the expected positive sign and this estimate suggests that when the RER changes with 1%, the trade balance significantly changes with 3.59%. This implies that a dollar depreciation

improves the bilateral trade balance for the US. Besides that, it is evidence that the M-L condition holds between the US and China in the long-run.

The long-run estimates of the other independent variables are also presented in table 3. The coefficients of the GDP of both the US and China are significant at the 1% level. The coefficient of the GDP of the US is 6.44, which indicates that the trade balance changes with 6.44% when the GDP of the US changes with 1%. This estimate could imply that an increase in the production of the US is mainly in the production of import substitutes. More products will be produced in the US and i.e. relatively less products are imported from China which results in an improvement of the bilateral trade balance for the US. The coefficient of the GDP of China is -1.92, which indicates that the trade balance changes with -1.92% when the GDP of China changes with 1%. This result could imply that an increase in the production of China is mainly in the production of import substitutes. This results in relatively less imports from the US, which i.e. would diminish US exports to China. Therefore, the bilateral trade balance deteriorates for the US. The coefficients of the remaining independent variables are not significant.

4.2 The Short-Run Effect

In the second stage of the ARDL approach by Pesaran et al. (2001), the ECM of Engle and Granger (1987) is estimated. The ECM in equation (5) is constructed as in section 3.3.2. The ECM captures the short-run effect which is used to determine whether there is a J-curve

24

effect. There is a J-curve effect when the short-run effect of the RER on the trade balance is negative, while the long-run effect is positive.

The ECM includes an error-correction, which is estimated first and then added to the ECM as a new variable. The error-correction term is a time series consisting of the lagged residuals of equation (1). Accordingly, equation (1) has to be estimated first by OLS to determine the coefficients of the variables and the constant. The error-correction term (𝐸𝐶𝑇_𝑡−1) is then estimated as in equation (4). The constant and the coefficients of the variables in equation (1) are tabulated in table 4.

Table 4. OLS Estimate of Equation (1) – Trade Balance Model

Variable Coefficient Absolute

t-value Constant -40.65 1.22 𝒍𝒏𝑹𝑬𝑹_𝒕𝑼𝑺,𝑪 2.23** 2.16 𝒍𝒏𝑮𝑫𝑷𝒕𝑼𝑺 2.34 1.39 𝒍𝒏𝑮𝑫𝑷𝒕𝑪 0.22 1.30 𝒍𝒏𝑹𝑬𝑬𝑹𝒕𝑼𝑺 -1.95* 1.94 𝒍𝒏𝑹𝑬𝑬𝑹𝒕𝑪 1.16 1.31 𝒍𝒏𝑮𝑫𝑷𝒕𝑶𝑬𝑪𝑫 -5.79* -2.06

** denotes significance at the 5% level. * denotes significance at the 10% level.

The OLS estimate of equation (1) in table 4 reveals that a 1% increase in the RER changes the trade balance with 2.23% with 5% significance. This indicates that a dollar depreciation improves the bilateral trade balance for the US. Furthermore, a decrease of the US REER by 1% improves the trade balance with 1.95% at 10% significance. This indicates that when the dollar depreciates against a basket of major currencies, which consists for 17.8% of the renminbi, that the bilateral trade balance with China improves for the US. Apparently, the US REER has a similar effect on the trade balance as the RER. When the OECD GDP increases with 1%, then the trade balance drops with 5.79% at 10% significance. This indicates that when the GDP in the OECD member countries increases, then the bilateral trade balance with China deteriorates.

Following Bahmani-Oskooee and Brooks (1999), only the coefficients of the lagged values of the first difference of the RER and the error-correction term from the estimation of equation (5) are tabulated in table 5. The number of lags has been determined by estimating the AIC for several lag combinations. The starting combination of lags was the combination

25

used in the first stage ARDL-ECM. From this combination onwards, one lag is added to or subtracted from one specific variable to create a new combination. The combination that reaches the lowest AIC estimate is the optimal lag order. The optimal lag order is given by (3,2,1,3,4,0,0).

The number of lags for the first difference of the RER is two, therefore only the first difference and the lagged first difference are estimated. There is evidence for a J-curve pattern when there is a negative coefficient for the first difference followed by a positive coefficient for the lagged first difference. This would indicate that there is a negative short-run effect followed by a positive long-run effect. The error-correction term is an alternative method to determine a cointegration relationship between the variables. A significantly negative coefficient of the error-correction term indicates that the variables are cointegrated.

Table 5. Coefficient Estimates of Real Exchange Rate and Error-Correction Term – (3,2,1,3,4,0,0)

Variable Coefficient Absolute t-value

∆𝒍𝒏𝑹𝑬𝑹_𝒕𝑼𝑺,𝑪 0.77 0.69

∆𝒍𝒏𝑹𝑬𝑹_𝒕−𝟏𝑼𝑺,𝑪 -1.12* 1.67

𝑬𝑪𝑻𝒕−𝟏 -0,388*** 2.93

*** denotes significance at the 1% level. * denotes significance at the 10% level.

The effect of the first difference of the RER is positive and not significant, which is followed by a negative coefficient of the lagged first difference of the RER at 10%

significance. Therefore, the results from table 5 suggest that there is no evidence for a J-curve effect since there is no negative effect followed by a positive effect. While investigating the J-curve effect, several variables are excluded to create different variable combinations. When both the REERs and the GDP of the OECD countries are dropped, no J-curve effect is found. When dropping only the REERs or the OECD GDP, the same result is visible. However, in the case where you expect the RER to have influence on the trade balance after one quarter, the coefficient of the lagged first difference in table 5 can be seen as the short-run effect. The short-run effect is then significantly negative at the 10% significance level. Since the long-run effect was proved to be significantly positive in the first stage of the ARDL approach, a J-curve effect can be observed. Now, a significantly negative short-run effect is followed by a positive long-run effect indicating a J-curve effect. This is weak evidence however since it does not satisfy the strict interpretation provided by literature regarding the J-curve effect.

26

The coefficient of the error-correction term is -0.388, which is negative and significant at 1%. Therefore, there is a long-run cointegration relationship between the variables. The size of the coefficient indicates the speed of adjustment to the equilibrium value of the trade balance after a shock. So after one quarter, the trade balance returns for 38.8% back to its equilibrium value. Therefore, the trade balance returns to its equilibrium value within 2.58 quarters.

5. Conclusion

Donald Trump and American politicians and economists name the manipulation of the renminbi as the reason for the large trade deficit of the US with China. To check the validity of this claim, this thesis aimed to investigate whether the trade balance between the US and China is influenced by the bilateral real exchange rate in the period from 1993 until 2016. Quarterly bilateral trade data between the US and China is used to examine the effect of the RER on the trade balance.

Research regarding the relation between the bilateral trade balance and the RER have been conducted in the past. Several studies investigated whether there was a long-run

relationship between the US and China, while other studies investigated both the short-run and the run effect for these countries. This thesis will investigate whether there is a long-run relationship to examine whether the Marshall-Lerner condition holds. Besides that, the short-run dynamics are investigated to determine whether there is a J-curve effect. To contribute to the existing literature, more recent data is added.

Methodologically, this thesis follows the two stage ARDL approach by Pesaran et al. (2001). In the first stage, cointegration among the variables is examined and the long-run effects are estimated by the ARDL-ECM, which includes the lagged-level variables. In the second stage, the short-run effect is estimated by the ECM, which replaces the lagged-level variables by the error-correction term.

The results from the bounds cointegration test indicate that there is a long-run cointegration relationship between the variables. This implies that the RER affects the trade balance in the long-run. The long-run effect of the RER is 3.59, which indicates that a 1% dollar depreciation results in a 3.59% improvement of the bilateral trade balance for the US. To conclude, a dollar depreciation results in the improvement of the US-china trade balance for the US. Therefore, the Marshall-Lerner condition is fulfilled. Besides that, the result complements to the statements of Donald Trump and American politicians and economists

27

since the Chinese exchange rate is able to influence the bilateral trade balance in the advance of China.

Compared to previous literature, the result supports Koo and Zhuang (2007) and Chiu et al. (2010), who suggest that a dollar depreciation improves the bilateral trade balance for the US. The result however contradicts Yue and Zhang (2013), who suggest the opposite. The result also differs from research in a similar more recent time period, since Shi and Li (2017) find no statistical evidence for a long-run relationship when investigating the period from 1993 until 2012.

The long-run cointegration relationship between the variables is confirmed by the coefficient of the error-correction term, since it is significantly negative. The coefficient of the error-correction term is -0,388, which means that the trade balance returns to its equilibrium value in 2.58 quarters after a shock.

The results from the ECM in the second stage show that there is no pattern that

indicates that there is a J-curve effect. This result is in line with Narayan (2006) and Bahmani-Oskooee & Wang (2006), who find no statistical evidence for a J-curve effect between the US and China. On the other hand, weak evidence for the J-curve effect can be observed when the coefficient of the lagged first difference of the RER is considered as the short-run effect. Since the coefficient is negative and significant at the 10% level, a negative short-run effect is established. The negative short-run effect is then followed by a significantly positive long-run effect, which is established in the first stage, indicating a J-curve effect. This is weak evidence however since previous literature do not use this consideration to detect a J-curve effect.

The long-run coefficient of 3.59 of the RER is large compared to the long-run coefficients found by Bahmani-Oskooee and Wang (2006), which are smaller than one. To elaborate, the long-run coefficient could overestimate the effect of the RER on the trade balance. A possible reason for any overestimation is the selection of variables. The GDP variables of both countries and the OECD are for instance nominal data while real data is used in all previous studies regarding this topic, e.g. Bahmani-Oskooee and Brooks (1999) and Bahmani-Oskooee and Wang (2008). This is due to unavailability of real GDP estimates in public databases. The use of nominal GDP data could have caused the difference in the results. Besides that, the REER of both countries and the GDP of OECD member countries are added to the model. Both the REER of the US and China consist of the bilateral exchange rate between the US and China and the GDP of the OECD largely depends on the GDP of the US. Therefore, the RER and the REER of both countries could be correlated with each other as well as the US GDP and the GDP of the OECD, which indicates multicollinearity. This

28

could have caused a bias, which can explain the difference in the results. Further research on this topic using similar variables could be improved by using real instead of nominal data. Besides that, the exchange rate between the US and China should be filtered out of the REER for both countries. The same applies for the GDP of the OECD, which should be filtered for the US GDP.

To conclude, the main aim in this thesis was to investigate whether the real exchange rate between the US and China influences the bilateral trade balance in the period from 1993 until 2016. Following the ARDL-ECM approach by Pesaran et al. (2001), no J-curve pattern has been found. However, a positive long-run cointegration relationship between the RER and the bilateral trade balance has been found. To explain, a depreciation of the dollar results in the improvement of the bilateral trade balance for the US in the long-run. The US trade deficit with China, and i.e. the overall US trade deficit, can therefore to some extent be allocated to changes in the US-China real exchange rate.

29 6. Reference list

Adom, P. F., Akoena, S. K. K., & Bekoe, W. (2011). Modelling aggregate domestic

electricity demand in Ghana: An autoregressive distributed lag bounds cointegration approach. Energy Policy, 42, 530-537.

Anderson, J., & Van Wincoop, E. (2003). Gravity with Gravitas: A Solution to the Border Puzzle. American Economic Review, 93(1), 170-192.

Authers, J. (2016, October 14). Where Trump is right and wrong on the renminbi. The

Financial Times. Retrieved from

https://www.ft.com/content/5a89693a-91db-11e6-8df8-d3778b55a923

Bahmani, M., Harvey H., & Hegerty, S. W. (2013). Empirical tests of the Marshall-Lerner condition: A literature review. Journal of Economic Studies, 40(3), 411-443.

Bahmani-Oskooee, M. (1991). Is there a long-run relation between the trade balance and the real effective exchange rate of LDCs?. Economics letters, 36(4), 403-407.

Bahmani-Oskooee, M., & Brooks, T. J. (1999). Bilateral J-curve between US and her trading partners. Weltwirtschaftliches Archiv, 135(1), 156-165.

Bahmani‐Oskooee, M., & Wang, Y. (2006). THE J CURVE: CHINA VERSUS HER TRADING PARTNERS. Bulletin of Economic Research, 58(4), 323-343.

Bahmani-Oskooee, M., & Wang, Y. (2008). The J-curve: evidence from commodity trade between US and China. Applied Economics, 40(21), 2735-2747.

Baek, J., & Chi, J. (2013). Dynamic relationship between air transport demand and economic growth in the United States: A new look. Transport Policy, 29, 257-260.

Chang, M., Chiang, C., & Jiang, S. (2012). Price discovery in stock index: An ARDL-ECM approach in Taiwan case. Quality & Quantity, 46(4), 1227-1238.

Chiu, Y. B., Lee, C. C., & Sun, C. H. (2010). The US trade imbalance and real exchange rate: an application of the heterogeneous panel cointegration method. Economic

Modelling, 27(3), 705-716.

Dong, F. (2017). Testing the Marshall-Lerner condition between the U.S. and other G7 member countries. North American Journal of Economics and Finance, 40, 30-40.

Engle, R. F., & Granger, C. W. (1987). Co-integration and error correction: representation, estimation, and testing. Econometrica: journal of the Econometric Society, 55(2), 251-276.

Koo, W. W., & Zhuang, R. (2007). The role of exchange rate in Sino‐US bilateral trade. Contemporary Economic Policy, 25(3), 362-373.

30

Lal, A. K., & Lowinger, T. C. (2002). The J-curve: evidence from East Asia. Journal of

Economic Integration, 17(2), 397-415.

Liew, L. (2010). US Trade Deficits and Sino-US Relations. Journal of Contemporary

Asia, 40(4), 656-673.

Ma, S., & Tian, G. (2010). The relationship between stock returns and the foreign exchange rate: The ARDL approach. Journal of the Asia Pacific Economy, 15(4), 490-508.

Morrison, W. M., & Labonte, M. (2008). China’s currency: Economic issues and options for US trade policy. Currency Interventions, Fluctuations and Economic Issues, 1. Narayan, P. K. (2006). Examining the relationship between trade balance and exchange rate:

the case of China's trade with the USA. Applied economics letters, 13(8), 507-510.

Onafowora, O. (2003). Exchange rate and trade balance in East Asia: is there a J-curve. Economics Bulletin, 5(18), 1-13.

Shi, C., & Li, J. (2017). Does dollar-pegging matter? A closer look at US trade deficits with China and Germany. The Journal of International Trade & Economic

Development, 26(4), 451-472.

Shirvani, H., & Wilbratte, B. (1997). The Relationship Between The Real Exchange Rate And The Trade Balance: An Empirical Reassessment. International Economic

Journal, 11(1), 39-50.

Yue, K., & Zhang, K. H. (2013). How Much Does China's Exchange Rate Affect the US Trade Deficit?. Chinese Economy, 46(6), 80-93.

Zhu, Y. (2010). Trade, capital flows and external balance: Is China unique in two hundred years of globalisation?. Journal of Chinese Economic and Business Studies, 8(1), 1-22.

31 Appendix

Table 6. Estimation of the ARDL-ECM in Equation (2) – (3,3,1,3,4,0,0)

Obs 92

R-squared 0.82693919 Adj R-squared 0.77818967

D.Ln_TB Coef. Std. Err. t P>|t| [95% Conf. Interval]

Ln_TB L1. -.8213034 .1618984 -5.07 0.000 -1.14412 -0.4984871 Ln_RER L1. -2.948817 1.252324 -2.35 0.021 -5.445879 -.4517542 Ln_GDPUS L1. -5.289893 1.866995 -2.83 0.006 -9.012576 -1.567211 Ln_GDPChina L1. 1.577602 .2664326 5.92 0.000 1.04635 2.108853 Ln_REERUS L1. .7998901 1.087636 0.74 0.464 -1.368795 2.968575 Ln_REERChina L1. -.3124127 .8467639 -0.37 0.713 -2.000812 1.375987 Ln_OECDGDP L1. 1.839826 2.762644 0.67 0.508 -3.668729 7.348382 Ln_TB LD. 0.0543705 .1237298 0.44 0.662 -.1923398 .3010807 L2D. -.2057014 .0968366 -2.12 0.037 -.3987881 -.0126146 Ln_RER D1. -0.0742671 .9193132 -0.08 0.936 -1.907325 1.758791 LD. .0661752 .6288292 0.11 0.916 -1.187675 1.320025 L2D. 1.248063 .5796488 2.15 0.035 .0922761 2.40385 Ln_GDPUS D1. .577499 2.206032 0.26 0.794 -3.821203 4.976201 Ln_GDPChina D1. .1952529 .1463869 1.33 0.187 -.0966343 .4871402 LD. -.4592114 .2235328 -2.05 0.044 -0.9049232 -0.0134996 L2D. .4100091 .1479087 2.77 0.007 .1150876 .7049306 Ln_REERUS D1. -0.0137457 .9198558 -0.01 0.988 -1.847886 1.820395 LD. .7012465 .5285028 1.33 0.189 -.3525581 1.755051 L2D. -1.0882 .5464094 1.99 0.050 -2.177709 .0013098 L3D. 1.24995 .2666148 4.69 0.000 .7183357 1.781565 Ln_REERChina D1. -.2565856 .6981422 -0.37 0.714 -1.648642 1.13547 Ln_OECDGDP D1. 1.511056 2.309999 0.65 0.515 -3.094952 6.117063 _cons 74.90967 32.93479 2.27 0.026 9.239555 140.5798

32

Table 7. Estimation of the ECM in Equation (5) – (3,2,1,3,4,0,0)

Obs 92

R-squared 0.7705 Adj R-squared 0.7252

D.Ln_TB Coef. Std. Err. t P>|t| [95% Conf. Interval]

Ln_TB LD. -2.227065 .1209874 -1.76 0.083 -.4539738 .0282607 L2D. -.2959062 .0996502 -2.97 0.004 -.4943768 -.0974355 Ln_RER D1. .7661595 1.10899 0.69 0.492 -1.442585 2.974904 LD. -1.124384 .6737691 -1.67 0.099 -2.466311 .2175439 Ln_GDPUS D1. -1.892213 2.840877 -0.67 0.507 -7.55031 3.765884 Ln_GDPChina D1. -1.001434 .1433157 -0.70 0.487 -.3855814 .1852946 LD. .2998203 .1797445 1.33 0.186 -.118172 .5978126 L2D. .8488579 .1387183 6.12 0.000 .5725765 1.125139 Ln_REERUS D1. -.6065996 1.097979 -0.55 0.582 -2.793415 1.580216 LD. .6149758 .5571039 1.10 0.273 -.4945928 1.724544 L2D. -.711994 .2792804 -2.55 0.013 -1.268235 -.1557643 L3D. .8244945 .2635716 3.13 0.002 .2995461 1.349443 Ln_REERChina D1. -.7673803 1.083527 -0.71 0.481 -2.925411 1.39065 Ln_OECDGDP D1. .8578124 3.691723 0.23 0.817 -6.494891 8.210516 ECT L1. -.3878555 .1323875 -2.93 0.004 -.651528 -.124183 _cons -.0185933 .0247723 -0.75 0.455 -.0679315 .030745