The Effect of Credit Growth in the U.S. on the Asian Current Account Balance

(1)

The Effect of Credit Growth in the U.S. on

the Asian Current Account Balance

University of Groningen

Faculty of Economics and Business

Master Thesis International Economics and Business

Focus area: International Capital and Globalization

Name: I.M.E. Erich

Student number: s3519104

Student e-mail: i.m.e.erich@student.rug.nl

Supervisor: dr. A.C. Steiner

Co-assessor: dr. P.A. IJtsma

(2)

ABSTRACT

Based on the theory of Borio and Disyatat (2011), the research question of this paper is: What is the effect of credit growth on the Asian current account balance? Using domestic private credit growth and credit growth to households as measures for credit growth in the U.S., the relationship is investigated by a system GMM model as baseline model. The main finding is that when households in the U.S. receive one percent more credit, they will consume more Asian goods and services. As a result, on average, the current account balances of Asian countries increase with approximately 0.607 percentage points.

(3)

1. Introduction

During the last years, global trade imbalances are a much-discussed topic. Particular the current account balances of the United States (U.S.) and China, the two largest economies in the world, are striking. Figure 1.1 depicts the proportions of the current accounts of both countries. The U.S. has had a current account deficit for at least 35 years. Whereas China has had a large current account surplus for 20 years. Attention to global trade imbalances is needed due to its high persistence which worries policymakers. The large current account deficit of the U.S. may cause protectionist pressure in the country. At the time of writing, the trading relationship between the U.S. and China is under tension. One of the results is increasing import tariffs. These rising import tariffs cause a deterioration of the current account balance of both nations because importing becomes more expensive. The import and export of goods and services are represented in the current account balance. Initially, the U.S. produced many industrial goods which were exported to the rest of the world. Nowadays, the production of many goods is offshored to low-wage countries like China. Therefore, the U.S. exports less and China exports more to the U.S. As a result, global trade imbalances occur and protectionists actions are increasing in the U.S., like rising importing tariffs on Chinese goods. Higher import tariffs are not good for international trade and might affect welfare negatively in both countries in the long-run. Therefore, large global trade imbalances are risky for the economy (Bettendorf, 2017). In addition, when one country has a very large deficit, it depends very much on foreign investment. The country has high debt levels and might get into trouble when foreign investors do not want to invest in the country anymore. Figure 1.1 represents that during the global financial crisis the gap between the current account balances of the U.S. and China reduced. Still, households, enterprises, and government in the U.S. have to borrow on the international capital markets to cover the difference between the costs of imports and receipts from exports.

Figure 1.1 Current Account Balance as a percentage of GDP. Source: World Bank Indicators

-8% -6% -4% -2% 0% 2% 4% 6% 8% 10% 12% 198 2 198 3 198 4 198 5 198 6 198 7 198 8 198 9 199 0 199 1 199 2 199 3 199 4 199 5 199 6 199 7 199 8 199 9 200 0 200 1 200 2 200 3 200 4 200 5 200 6 200 7 200 8 200 9 201 0 201 1 201 2 201 3 201 4 201 5 201 6 201 7

Current Account Balance (% of GDP)

(5)

The United States depends on foreign credit to maintain its current account deficit. Furthermore, the net foreign asset position of the U.S. is declining, which is represented in figure 1.2 (Bernanke, 2005). Whereas, China has an extensive increase in its net foreign asset position.

Figure 1.2 Net Foreign Asset Position U.S. Source: World Bank Indicators

There are different explanations for the widening global current account imbalances. Bernanke (2005) argues that the saving glut in emerging economies causes capital inflows into the U.S. These capital flows maintain the current account deficit of the U.S. and give a downward pressure on the interest rate. Therefore, borrowing becomes cheap in the U.S. which results in an increase in consumption and investment. On the other hand, Borio and Disyatat (2011) argue that the large current account surplus of China and other emerging countries is not the cause of the growing imbalances and the financial crisis, but the outcome. Before the global financial crises, U.S. citizens received relatively easy a loan. Besides, the interest rate was too low, which made financing cheap. U.S. citizens used credit to consume Asian goods and services. Therefore, the current account deficit of the U.S increased and the current account surplus of Asia increased. Asia stored the surplus in international reserves. These reserves are invested in safe assets like U.S. Treasury bills. As in Bernanke’s view, these capital inflows cause downward pressure on the interest rate and results in more consumption and investment. The current literature has investigated the determinants of the current account balance. One of those determinants is credit growth. Researchers have found that an increase in credit leads to a deterioration of the domestic current account balance. Especially an increase in credit to households results in a decline in the saving rate and hence more consumption. This has a negative impact on the current account balance (Büyükkarabacak and Krause, 2009). Besides, Taylor (2009) confirms the theory of Borio and Disyatat (2011) and argues that the monetary policy was too loose in the U.S. during the period of a deterioration of the current account deficit. Furthermore, Atoyan et al. (2013), Wyplosz (2013), and Unger (2017) investigate trade imbalances in the Eurozone and confirm that the domestic private credit growth is negatively

-700 -600 -500 -400 -300 -200 -100 0 100 200 300 197 7 197 8 197 9 198 0 198 1 198 2 198 3 198 4 198 5 198 6 198 7 198 8 198 9 199 0 199 1 199 2 199 3 199 4 199 5 199 6 199 7 199 8 199 9 200 0 200 1 200 2 200 3 200 4 200 5 200 6 200 7 200 8 200 9 201 0 201 1 201 2 201 3 201 4 201 5 201 6

(6)

correlated with the current account balance. In addition to the current literature, this paper elaborates on the existing literature by examining the effect of credit growth in one country on the current account balance in another set of countries. Therefore, the research question of this paper is: What is the effect of credit growth in the U.S. on the current account balance in Asia? The theory of Borio and Disyatat (2011) is examined based on two hypotheses. The first hypothesis is: Credit growth in the U.S. has a positive effect on the current account balance of Asian countries. Based on Büyükkarabacak and Krause (2009), the second hypothesis is derived: Credit growth to households in the U.S. has a positive effect on the current account balance of Asian countries. A dynamic model (system GMM) is used in order to investigate multiple countries in the period from 1982 to 2016. In addition, the robustness of the estimates is checked by using static models. Two datasets are used. First, a dataset including 140 countries examines the relationship by using an interaction variable with an Asian dummy. Secondly, the relationship is examined by a dataset including only Asian countries. This paper finds highly significant results for the second hypothesis. This implies that when American households receive one percent more credit (relative to GDP), the current account balance of Asian countries increases with approximately 0.607 to 0.866 percentage points. The estimations for the first hypothesis were not robust enough.

The remainder of this paper is organized as follows. Section two contains a review of the literature. Based on this literature review, the two hypotheses are formulated. Section three discusses the data, data transformations, pre-estimation tests, and explains the empirical model. The results are discussed based on the baseline model of this paper in section four. Section five includes robustness checks based on static models. Lastly, section six discusses the conclusion of the research that has been done.

2. Literature review and hypotheses

(7)

This chapter explains the two theories of Bernanke (2005) and Borio and Disyatat (2011). Paragraph 2.1 discusses the theory of Bernanke (2005) that is followed by a review of empirical articles that examined his theory. Thereafter, paragraph 2.3 discusses the theory of Borio and Disyatat (2011). The empirical evidence for this theory is explained in paragraph 2.4. In addition, the determinants for credit are described in paragraph 2.5. Lastly, paragraph 2.6 discusses what this paper adds to the existing literature.

2.1 Theory: U.S. current account deficit made by the rest of the world

Bernanke (2005) argues that the large current account deficit of the U.S. is due to an excessive desire to save in the world. He shows that the large increase in the current account deficit of the U.S. between 1996 and 2004 is mainly due to developing countries. Bernanke (2005) refers to this phenomenon as the ‘global saving glut’ and ‘excess saving’ which causes also the low level of long-term real interest rates.

Between 1990 and 2000, most developing countries imported more capital than they exported, which means that they borrowed capital on the international markets. However, this capital was not always located effectively and efficiently. Capital did not move to the entrepreneurs and projects that could make high returns. Besides, those countries had large amounts of debt denominated in foreign currencies and overvalued domestic currencies made developing countries vulnerable which resulted in several financial crises. For Asian countries, the crises occurred in the period 1997-1998. As a result, capital in developing countries from foreign investors has pulled away and the exchange rate of domestic currencies depreciated. Therefore, emerging markets choose new strategies for managing international capital flows and became net exporters of financial capital. The high saving ratio made emerging countries net lenders on international capital markets instead of borrowers. In addition, they build large international reserves and became surplus countries, while in industrialized countries the desire to save declined. These large international reserves of developing countries are invested in the U.S. due to its attractive investment environment. In early 2000, productivity in the U.S. increased due to new technologies and developments. Besides, the U.S. is characterized as a country with strong property rights, a good regulatory environment, and low political risk. These factors make investors willing to invest in the U.S. In addition, central banks prefer to invest in government bonds because they are seen as safe assets.

(8)

2.2 Empirical evidence: U.S. current account deficit made by the rest of the world

A number of theories argue that global trade imbalances are due to a high saving ratio in emerging countries. Those theories explain also the determinants of the high saving ratio in Asia. This ratio has a positive effect on the current account balance. First, the ‘Bretton Woods II’ hypothesis is examined by Dooley et al. (2003, 2004, 2009). Asian economies peg their domestic currencies to the U.S. dollar to keep the relative prices of Asian goods low. Hence, Asian countries become very competitive in international trading which results in current account surpluses. In this theory, the U.S. current account deficit is not made by the U.S. itself but made by Asian countries. In addition, Asian financial markets are less integrated and less developed than the financial markets in developed countries. People in Asia save more because they might not invest easily in financial assets. A high saving rate leads to an increase in international reserves and an improvement of the current account balance. Furthermore, there are other reasons why the saving rate in Asia is high. For instance, people save based on precautionary reasons or there is an insufficient supply of safe assets (Bettendorf, 2017). These motives cause higher saving rates and higher current account surpluses. Higher current account surpluses might imply high international reserves which will be invested in safe assets, like U.S. Treasury bills. However, Gruber et al. (2007) examine the determinants of the current account deficit of the U.S. The authors are not able to explain the deficit with indicators that capture the special attractiveness of financial markets in the U.S. Nevertheless, the authors support the view of Bernanke (2005), but it is not clear how those large current account surpluses of Asian developing countries end up in the U.S. instead of being spread over the world.

(9)

favour of a global saving glut. He shows that the global saving rate is actually declining. Taylor (2009) mentions that in the period 2002-2004, the U.S. invested more than it saved. Therefore, the current account deficit originated.

2.3 Theory: ‘Homemade’ U.S. current account deficit

The other explanation for the large current account deficit of the U.S. is from Borio and Disyatat (2011). First of all, the authors emphasize that saving and finance are not equal. Saving is the part of income which is not consumed. Whereas, finance is a cash flow concept that facilitates the funding of investments and expenditures (Borio and Disyatat, 2011). The authors argue that there are not enough anchors in the international monetary and financial system that prevent the world for excessive credit growth. Before the financial crisis, credit expansion in the U.S. caused an increase in the consumption of Asian goods and services. Therefore, the current account surplus of Asia increased because export to the U.S. increased. On the other hand, the current account deficit of the U.S. deteriorated because import increased and exceeded exports. This caused the global current account imbalances to become larger. Asian countries build up international reserves due to the increasing surplus. They invested their reserves in safe assets like U.S. Treasury bills, which drove up asset prices in the U.S. As a result, the interest rate in the U.S. decreased which made credit cheaper. This loop caused a boom which will be followed by a bust. When the economy is in a boom phase, asset prices are high, and the economy enters a bubble. At the highest peak of the boom, one investor sells its stocks and receives high profits from doing so. Therefore, more investors will follow. This results in declining stock prices and eventually in a declining economy. Borio and Disyatat (2011) refer to this as the ‘excess elasticity’ view. In other words, we could say that the U.S. deteriorates its own current account deficit by having no strong constraints on excessive credit growth. The excess elasticity in the financial markets causes an increase in consumption and hence an increase in consumption of foreign goods. U.S. citizens would mainly consume more goods and services from Asian countries, according to Borio and Disyatat (2011). Therefore, the current account surplus in Asia becomes larger. Based on the theory, the following hypothesis is formulated.

Hypothesis I: Credit growth in the U.S. has a positive effect on the current account balance of Asian countries.

2.4 Empirical evidence: ‘Homemade’ U.S. current account deficit

(10)

depth. They find that the degree of financial excess has a higher impact on the current account balance of countries that have lower levels of financial depth. Despite the fact that the U.S. has a high level of financial depth, the article of Ekinici et al. (2015) proves that high levels of credit growth deteriorate the current account balance of a country. Furthermore, Taylor (2009) argues that the current account deficit of the U.S. is not due to the global saving glut, but due to too loose monetary policy of the central bank. The Taylor-rule indicates that the interest rate was far too low in the 1990s. In other words, the rule indicated that the central bank should have had increased the interest rate. An interest rate that is (too) low, stimulates the economy and results in a large increase in credit and hence an increase in consumption, which in turn causes a deterioration of the current account balance of the U.S. The findings of Chinn and Ito (2007, 2008) and Laibson and Mollerstrom (2010) are in line with Taylor (2009). They investigate trade imbalances by solving an equilibrium in a Cobb-Douglas economy during the period 1997 to 2006. They find that the current account deficit of the U.S. is caused by a decline in the domestic saving ratio in the U.S. The theory of Laibson and Mollerstrom (2010) is that asset price bubbles cause household savings to decline and consumption to increase.

The following articles examined the link between credit growth and the current account balance. First, Atoyan et al. (2013) investigate the adjustments in global current account balances after the crisis. Figure 1.1 in the introduction represents that before the global financial crisis, the current account imbalances increased rapidly. After the crisis, there are adjustments in both emerging and developed countries. Atoyan et al. (2013) use the growth of private credit to GDP as a measure for the degree of financial excess, like Ekinci et al. (2015). Furthermore, the authors explore the current account dynamics in the Euro area. They find that in emerging Europe (Bulgaria, Estonia, Latvia, and Lithuania), an increase in investment contributes mostly to current account imbalances. These investment booms cause an increase in asset prices. This is the same mechanism that is mentioned by Borio and Disyatat (2011). Besides, in the Eurozone periphery (Greece, Ireland, Portugal, and Spain), a decrease in private sector savings contributes mostly to current account imbalances. Secondly, a recent article by Unger (2017) provides evidence for the euro-area that one determinant of the current account balance is credit growth to the non-financial private sector. Credit growth is actually the most important factor that widens current account imbalances in deficit countries due to an increase in domestic demand. Third, Davis et al (2016) link credit growth to financial crises and the current account balance. They examine the marginal effects of credit growth. The authors provide evidence that when the economy is running a sizable current account deficit, the marginal effect of rising private sector debt levels is large. A one percentage point increase in credit growth makes the probability of a crisis larger by 0.5 percent point. These articles substantiate the first hypothesis. Credit growth in the U.S. would have a negative effect on its own current account balance, but according to Borio and Disyatat (2011), it has a positive effect on the current account balance of Asia.

(11)

for explaining foreign trade imbalances than the distribution of loans between households and enterprises. In addition, Adam et al. (2011) examine that when house prices decline, the current account improves due to subdued consumption. Linking this to the hypothesis of Borio and Disyatat (2011), an increase in housing prices may be the result of ‘excess elasticity’ of the monetary and financial system. Higher asset prices lead to more consumption and hence more import from Asian countries. Borio and Disyatat (2011) do not make the distinction between credit to households and firms. Based on the finding of Büyükkarabacak and Krause (2009), and Adam et al. (2011), the first hypothesis is expanded by the following hypothesis.

Hypothesis II: Credit growth to households in the U.S. has a positive effect on the current account balance of Asian countries.

2.5 Determinants of credit

(12)

2.6 Extension to the current literature

So far, many researchers investigated the determinants of the current account balance and found explanations for global trade imbalances. This article focuses on the financial determinants of the current account balance of the U.S. and Asia. One financial variable is credit growth. The literature examined the effect of credit growth on the current account balance, but it did not investigate the effect of credit growth in one country on the current account balance in another country or a group of countries (like Asia). Based on the theory of Borio and Disyatat (2011), this paper examines the effect of credit growth in the U.S. on the current account balance of a set of Asian countries. This specific relationship is important and relevant due to current political and economic tensions between the U.S. and China. Furthermore, the finding of Büyükkarabacak and Krause (2009) is very important for understanding global current account imbalances because it indicates an important determinant of import from Asian countries and hence their current account balances. Therefore, this paper examines credit growth further by looking at credit to households specifically.

3. Research design

This chapter explains the variables, data, and econometric model that will be used in order to investigate the relationship between credit growth and the current account. The first paragraph explains the dependent variable, explanatory variables, and the control variables. The second paragraph provides a description of the sample and tests that are needed in order to generate the right regressions. Lastly, the third paragraph provides an econometric model that explains the relationship.

3.1 Variables

3.1.1 Dependent variable

The dependent variable is the current account balance. The current account includes the trade balance which measures imports and exports in a country. Exporting results in an inflow of capital whereas importing results in an outflow of capital. Therefore, an increase in imports deteriorates the current account. In the regression, the current account balance is measured as the current account balance as a percentage of GDP. This measurement is used because countries have different economy sizes. As a result, it makes this variable the best measure to compare countries with each other. A negative value of the dependent variable indicates a current account deficit. The dependent variable is equal for the first and the second hypothesis. 3.1.2 Explanatory variables

(13)

Asian countries specifically. The variable that will be used in the regression is domestic credit to the private sector as a percentage of GDP which is also used by Ekinci et al (2015), and Davis (2016). In the second hypothesis, the explanatory variable is credit to households. In the regression, the variable is measured as credit to households and NPISHs from all sectors at market value as a percentage of GDP. According to Büyükkarabacak and Krause (2009), credit growth to households is expected to have a negative correlation with the current account balance of Asia which is also discussed in the literature review.

3.1.3 Control variables

This paragraph captures the other determinants of the current account balance. Their relevance is based on the existing academic literature. Ariyani et al. (2018) investigate factors that have an impact on export activity in the ASEAN region. The ASEAN region includes countries like Indonesia, Singapore, Thailand, Malaysia, the Philippines, and Vietnam. The authors show that GDP per capita, the interest rate, foreign direct investment (FDI), and the exchange rate have an impact on the current account. Furthermore, Chinn and Prasad (2003) provide empirical evidence of other determinants of the current account balance for industrial and developing countries. In their study, the authors focus on the medium run and they found, additionally to Ariyani et al. (2018), that the most important determinants are the government budget balance, the net foreign assets position, demographics, terms of trade, a dummy for oil exporting counties, and financial deepening.

GDP

Gross Domestic Product (GDP) is the sum of the value added by all resident producers in a country. In the short run, GDP has a negative effect on the current account. Ariyani et al. (2018) link this with the theory of Keynes. Income has a positive effect on consumption. This is also true on the national level. A higher GDP value indicates that citizens have a higher purchasing power and hence they will consume more domestic and foreign goods and services. Therefore, the value of import increases. In addition, Krugman and Obstfeld (1999) mention that an increase in GDP per capita leads to more consumption and hence import of foreign goods increases. An increase in import means that the current account balance decreases. Therefore, GDP per capita is used in the regression.

Interest rate

(14)

inflows in the country which results in a higher level of imports. Therefore, there will be a shift to a current account deficit.

FDI

Foreign Direct Investment (FDI) is the investment of foreign investors in a company in which at least 10 percent of the shares are bought. The variable that is used in the regression is net FDI that is derived from the balance of payments in U.S. dollars. According to empirical evidence, FDI has a negative effect on the current account balance. Research of Sahoo et al. (2014) confirms this. Besides, Hobza and Zeugner (2014) argue that more capital flows cause an appreciation of the domestic currency. This makes imports cheaper, which means that it will increase, exporting decrease, and the current account balance deteriorates. In addition, an increase in FDI leads to more innovation and accumulation of knowledge. This increases total factor productivity. More FDI leads to an increase in import of technology, capital goods, and raw materials, which again deteriorates the current account balance (Ariyani et al. 2018).

Exchange rate

Another determinant which is mentioned by Ariyani et al. (2018) is the exchange rate. The variable that will be used in the regression is the real effective exchange rate. Households, producers, and government determine their spending decisions on relative prices, like the real interest rate, and the real interest rate. These prices determine whether consumers or producers purchase goods and services from abroad or that foreign consumers and producers purchase from the ‘home’ country due to cheaper relative prices. Therefore, the real effective exchange rate determines import and export of a country and hence the current account position (World Bank, 2019). In the long run, the exchange rate has a positive effect on the current account balance of ASEAN countries. This is explained in the Marshall-Lerner’s elasticity theory. The theory states that a depreciation of the domestic currency means that the country becomes cheaper relative to other countries. This is favorable for the current account balance because exports will increase, and import will decrease.

Government budget balances

(15)

Net foreign assets

The sixth control variable is explained by using a definition of the current account balance. The current account balance is the sum of the trade balance and the balance of factor income. The balance of factor income is the sum of receipts from income, like interest, and payments of income. The balance of factor income is determined by the net foreign asset position. The net foreign asset position positively determines the current account balance. However, the sum of the current account should be zero for all countries in the world, with interest payments on the net foreign assets offsetting the trade balance. Therefore, in the long run, the current account and the stock of net foreign assets would globally have no relationship. However, for an individual country, there can be a relationship. According to Chinn and Prasad (2003), the sum the current account balances is called the steady state that is zero and is calculated by the growth rate of nominal GDP times the net foreign asset position. Various factors may influence this relationship during the transition to the long run. In addition, dynamics in real exchange rates determine the current value of the net foreign asset position. This would cause collinearity between the control variables. Therefore, a lagged variable of the net foreign asset position relative to GDP is used to deal with this collinearity.

Demographics

Two variables are used to indicate the demographics of a country. The first variable is a youth dependency ratio which is the ratio of the population between 0 and 14 years of age relative to the working population (ages 15-64). The second variable is the old dependency ratio which is the ratio of the population of 65 years and older relative to the working population. The relation between these variables and the current account balance is explained by the life-cycle theory of consumption and saving. This theory states that young people borrow in order to consume. Then, middle-age people save for their retirement and retired people dissave. This implies that when the population of a country is relatively young or relatively old, it is likely that the current account will be a deficit (Gruber et al. 2007). Therefore, there will be a negative relationship between the demographic variables and the current account balance. The World Bank created a variable that is the combination of the youth and old dependency ratio. This variable is called the age dependency ratio as a percentage of the working population. For this paper, the age dependency ratio has been transformed into the relative age dependency ratio. It is relative to the sample’s average in order to determine whether a country is relatively young or old. When the value of this variable is larger than one, the population consists of more people that spend more. In other words, the population consists of more people, relative to the world, that are younger than 15 or older than 65 years of age.

Terms of trade

(16)

Financial deepening

Chinn and Prasad (2003) measure financial deepening as M2 relative to GDP which is also used in this paper. The theory about the relation between the current account and financial deepening is not very clear. There could be a positive relationship since more broad money results in higher savings. On the other hand, more broad money might also mean that there is more borrowing. This implies less saving and a lower current account balance. Chinn and Prasad (2003) found evidence for the first explanation which implies that there would be a positive relationship.

Oil exporting countries

Lastly, a dummy for oil exporting countries is included in the regression. Chinn and Prasad (2003) found a statistically significant and positive relationship between the dummy and the current account. This implies that, on average, oil-exporting countries have more favorable positions of their current account balance than others.

Chinn and Prasad (2003) also examined potential determinants of the current account balance like trade openness, capital controls of the current account and capital account. The authors found insignificant results for those variables. Therefore, those variables are not included for the regressions in this paper. All the variables are presented in table A.1 in the appendix. Behind the descriptives of domestic private credit and credit to households is stated ‘lagged variable’. This is explained in paragraph 3.3.

3.2 Descriptive statistics

In the literature review, two hypotheses are defined. The first hypothesis is based on the variable domestic private credit in the U.S. and the second hypothesis is based on the variable credit to households in the U.S. These hypotheses are examined by using two samples: a sample based on all the countries in the world (full sample) and a sample that only includes Asian countries. Paragraph 3.2.1 describes characteristics of the full sample and paragraph 3.2.2 describes characteristics of the sample including Asian countries.

3.2.1 Full sample description

(17)

normality, the correlation between explanatory variables, heteroskedasticity, and serial correlation.

Trends

There might be some trends in the variables. First, the current account balance as a percentage of GDP is examined. Figure 3.1 represents the mean of the observations in the sample of the dependent variable. It is striking that the mean of the current account balance is negative for the whole period 1982-2016. Then, the mean values of the variables domestic private credit in the U.S. and credit to households in the U.S. are represented in figures 3.2 and 3.3. Both variables show an upward trend during the years 1982-2016. However, when the global financial crisis started, there is a decline in credit to households in the U.S. ever since. Especially the variable credit to households in the U.S. show a decline since the global financial crisis.

(18)

Figure 3.1 Trend in the sample of the mean of the current account balance as a percentage of GDP; Figure 3.2 Trend in the sample of the mean of domestic private credit in the U.S.

Figure 3.3 Trend in the sample of the mean of credit to households in the U.S.

Summary sample

A summary of the data provides an overview of the main descriptives of the data. The summary shows that the observations of the real interest rate and the real effective exchange rate include large outliers. First, the maximum value of the real interest rate is 1,158 percent. This observation is from Zimbabwe. In the period 2004-2008, the country experienced hyperinflation which explains these high values. Secondly, the real effective exchange rate which has a maximum value of 3,522, is an observation for Ghana in 1983. In 1982, Ghana also experienced a real exchange rate value of 2,135. Inflation was also a problem in Ghana in these years. Furthermore, the variable credit to households has a maximum growth rate of 800 percent. This is an observation in Turkey. In the year 1993, credit to households relative to GDP increased from 0.2 percent to 1.8 percent. These examples of extreme values can be defined as outliers. Outliers might be a problem because they influence the least squares of the estimates. Extreme values can distort estimated results and standard errors. Reasons for this are errors in the data, mistakes in the data collection, or mismeasurement. There are several solutions that deal with outliers. First, one can drop the outliers from the data (trimming). However, this is not preferable because other observations of that country in that specific year will be lost as well. Secondly, variables could be transformed into logarithms. However, many observations

(19)

could be winsorized, which is the best solution for this paper. This implies that extreme values are limited below the 0.5th_{percentile and above the 99.5}th_{percentile. Observations below the}

0.5th_{percentile are equalized to the 0.5th percentile and observations above the 99.5}th_percentile

are equalized to the 99.5th_{percentile. Usually, a 1}st_{or 5}th_{percentile is used. However, if those}

percentiles were used, variables would be adjusted while they are not really outliers in the data. An advantage of winsorizing is that none of the observations has to be removed from the dataset. The mean, minimum, and maximum values of a variable determine whether it should be winsorized or not. All explanatory variables have extreme minimum and maximum values. Therefore, all explanatory variables are winsorized, except for the relative age dependency ratio and the dummy variables. Table 3.1 presents a description of the dataset in which countries with a population below three million are removed and includes that variables that are not winsorized yet. In addition, the table includes variables that are transformed into growth rates in order to make them stationary.

Variable Number of observations Mean Standard deviation Minimum maximum Current account 3,550 -2.641 8.106 -86.088 45.454

Credit U.S. (growth rate)

4,464 2.264 4.527 -8.860 48.423

Household credit U.S. (growth rate)

4,464 1.491 3.374 -5.176 38.510 Credit 3,643 42.756 42.755 0.443 233.211 Household credit (growth rate) 1,050 4.782 27.054 -80.000 800.000 Dummy Asia 4,604 0.266 0.442 0 1 Dummy*credit U.S. 4,464 0.605 2.594 -8.860 48.423 Dummy*household credit U.S. 4,464 0.394 1.904 -5.176 38.510 GDP per capita (growth rate) 3,262 4.260 6.628 -63.831 127.245

Real interest rate 2,243 7.243 36.607 -97.615 1,158.026

FDI (in millions) 3,432 -451.63 19,449.55 -231,651.60 177,277.00

Real exchange rate 2,243 112.445 113.785 18.735 3,522.154

NFA/GDPt-1 3,516 0.107 0.254 -14.311 3.299 Fiscal balance 1,843 -1.752 6.284 -83.768 129.184 Relative age dependency ratio 4,564 1.016 0.305 0.246 1.778 Terms of trade (growth rate) 3,265 0.636 12.886 -62.287 349.439 Dummy oil 4,604 0,233 0,423 0 1 M2/GDP (growth rate) 1,834 24.281 183.347 -80.789 6,477.133

Table 3.1 Data descriptive full sample.

(20)

rate. The real interest rate is the lending interest rate adjusted for inflation as measured by the GDP deflator (World Bank). In addition, the real exchange rate is replaced by the nominal exchange rate, which is measured as the official exchange rate in local currency relative to the U.S. dollar. Real values are better measurements, but generating more observations is more important for the reliability of the results. Furthermore, an alternative for the variable fiscal balance is central government debt as a percentage of GDP. However, the available data of this variable contains 1,358 observations, which is even less than the observations for the initial variable fiscal balance. As a result, the variable fiscal balance is removed from the baseline model because the number of observations in the regression decreases with 909 observations by leaving the variable in the regression. Lastly, m2/GDP is replaced by broad money growth, which is explained as currency outside the banking sector. For example, loans on the interbank market do not count as broad money, whereas loans to consumers do count as broad money. The expected effect on the current account balance, as is explained in paragraph 3.1.3, does not change for all new variables. Table 3.2 represents the descriptives of the new variables. The minimum and maximum values are extreme values. Therefore, these variables will also be winsorized by the 0.5th_{and 99.5}th_{percentile. In addition, the growth rate of the nominal}

exchange rate is used to make the variable stationary. New variables Number of

observations Mean Standard deviation Minimum Maximum Lending interest rate 2,357 74.846 2513.100 0.500 121,906.000 Nominal exchange rate (growth rate) 3,770 124.572 4,365.059 -99.991 262,676.70 Broad money growth rate 3,481 45.932 361.719 -88.339 12,513.140

Table 3.2 Data descriptives 'new variables' full sample Normality

(21)

is also shown by the kurtosis value. Nevertheless, the data is assumed to be normal because of its bell-shaped form.

Normality is also examined for the second hypothesis. The system GMM model is the baseline model for hypothesis two. Figure 3.5 represents a histogram with the distribution of the residuals. The graph is not bell-shaped but is skewed to the right. The value of the skewness is 1.102. On the other hand, the kurtosis has a value of 3.778, which is acceptable. The Jarque-Bera test generates a p-value of 0.000. This implies that the distribution of the residuals is not perfectly normal. The data might become more normal by transforming some variables into a logarithm. However, by doing so, many variables will be lost and lead to less reliable estimates. The normality assumption is optional for regression models. Charter Hill et al. 2012 (p. 178) argue that despite errors are not normally distributed, the least squares estimators are approximately normally distributed in large samples. Since this paper uses a large sample, it is not worrying that the error is not perfectly normal.

Pre-estimation tests

Correlation

A correlation matrix provides information about the correlation between the explanatory variables. The correlations must not be too high because this would bias the results of the regressions. In that case, one independent variable is a function of another independent variable. The literature review discusses the determinants of the credit as well. These are the interest rate, GDP per capita, financial deepening, and the exchange rate. Table A.4, in the appendix, presents the correlations between the explanatory variables. None of the explanatory variables is too high, like higher than 0.800 for example. This implies that there is no multicollinearity in the model. In addition, multicollinearity may be detected by using the Variance Inflation Factor (VIF). The results are shown in table A.6 in the appendix. Note that the variables are not winsorized by performing this test. A rule of thumb is that VIF values above ten are marked as too high. The VIF values for both the first and second hypotheses are all very low. This implies that the explanatory variables are not near linear combinations of each other.

0 5 10 15 Pe rce n t -40 -20 0 20 40 Linear prediction

System GMM model residuals

0 5 10 15 20 25 Pe rce n t -10 0 10 20 Linear prediction

(22)

Heteroscedasticity

The White test is conducted to detect heteroskedasticity in each regression. For the regression with the variable domestic private credit, the null hypothesis can be rejected. This implies that heteroscedasticity is present in the data. Therefore, robust standard errors are used to deal with this problem. Secondly, for the regression with credit to households, the null hypothesis of the White test cannot be rejected. This implies that heteroscedasticity is not present, and it is not required to use robust standard errors in the regression. The scatterplots are presented in figures A.1 and A.2 in appendix D. Besides, the test statistics of the White test are presented in table 3.3.

White test T-statistic p-value Robust standard

errors

Cr_US 326.50 0.000 Yes

Housecr_US 211.56 0.000 Yes

Table 3.3 Test statistics White test

Serial correlation

Serial correlation implies that a variable is correlated over time. This makes the standard errors bias in linear panel data models. Therefore, the results are less efficient. In addition, serial correlation is a problem in the residuals. The Wooldridge test examines the residuals from a regression in first-differences (Drukker, 2003). The results of the Wooldridge test are represented in table 3.4. The null hypothesis can be rejected for both regressions based on ‘domestic private credit in the U.S.’ and ‘credit to households in the U.S.’ This implies that serial correlation is present in the error term. By clustering the standard errors in the regression, the estimates are consistent.

Wooldridge test F-statistic p-value Conclusion

Cr_US 61.760 0.000 First-order

autocorrelation

Housecr_US 44.210 0.000 First-order

(23)

3.2.2 Asian sample description

After mapping the characteristics of the full dataset, the same steps will be done for the dataset that only includes Asian countries. The number of individual countries is 39, after removing countries with a population below 3 million citizens. Furthermore, 35 years are investigated from the period between 1982 and 2016.

Trends

The trend in the dependent variable in the dataset of Asian countries is a bit different from the trend in the full sample. The trend in the current account balance relative to GDP is also increasing, but between 2000 and 2012, Asian countries have on average a positive value of the current account ratio. Whereas, the average value in the full dataset is negative for 35 years. The trends in the two variables of interest, domestic private credit in the U.S. and credit to households in the U.S., do not differ from figures 3.2 and 3.3.

Figure 3.6 Trend in the current account balance relative to GDP

The Fisher unit root test for panel data is performed in order to detect non-stationary. The results are presented in table A.3 in the appendix. The variables domestic private credit in the U.S., credit to households in the U.S., NFA/GDP, terms of trade, and the exchange rate are non-stationary. Therefore, these variables are transformed into growth rates in order to make them stationary.

Summary statistics

Table 3.5 presents a description of the dataset. The minimum and maximum values indicate that there are large outliers in the dataset. Therefore, all explanatory variables are winsorized for the 0.5th_{and 99.5}th_{percentiles, except for the relative age dependency ratio and the dummy}

(24)

Variable Number of observations Mean Standard deviation Minimum maximum Current account 854 -0.912 9.051 -30.688 45.454 Credit U.S. (growth rate) 1,183 2.282 4.645 -8.860 48.423 Household credit U.S. (growth rate)

1,183 1.487 3.472 -5.176 38.510 Credit 935 53.638 48.194 0.963 233.211 Household credit 241 43.050 20.676 6.200 92.600 GDP per capita 890 13,015.70 16,883.82 736.82 89,203.34 Lending interest rate 640 13.090 9.174 1.045 96.100

FDI (growth rate) 808 -2,444.14 23,647.67 -231,651.60 149,731.90

Nominal exchange rate (growth rate)

1,005 19.921 157.796 -29.352 3,070.245 NFA/GDPt-1 (growth rate) 856 -60.010 1,745.896 -49,330.090 4,029.156 Relative age dependency ratio 1,219 0.955 0.292 0.246 1.778 Terms of trade (growth rate) 794 0.404 9.749 -44.843 50.769 Dummy oil 1,222 0,257 0,437 0 1 Broad money growth 928 23.191 67.966 -43.738 1,116.547

Table 3.5 Data descriptive sample Asian countries Normality

(25)

Pre-estimation tests

Correlation

The correlation matrix differs from the correlation matrix for the full sample. Table A.4 in the appendix presents the correlation matrix for the Asian dataset. The lending rate has higher correlations with domestic private credit and credit to households. However, the correlations are not high enough to remove the variable from the dataset. Moreover, the VIF value of the lending rate is equal to 2.45, which indicates that the variable is not a linear combination of other variables. Other correlation values of the explanatory variables are not striking or too high.

Heteroscedasticity

The White tests for both hypotheses one and two determine that there is heteroscedasticity present in the data. Therefore, robust standard errors are needed to correct the initial standard errors. Figure 3.11 and 3.12 are scatterplots of the variables of interest and their residuals. Table 3.6 presents the test statistics for both hypotheses.

White test T-statistic p-value Robust standard

errors

Cr_US 188.90 0.000 Yes

Housecr_US 119.63 0.000 Yes

Table 3.6 Test statistics White test

Serial correlation

The tests for serial correlation are represented in table 3.7. The Wooldridge tests for both the first and second hypotheses are statistically significant. This implies that there is first-order serial correlation in the model. Therefore, clustered robust standard errors are used to deal with this problem. 0 5 10 15 20 25 Pe rce n t -40 -20 0 20 40 Linear prediction

0 5 10 15 20 25 Pe rce n t -10 0 10 20 Linear prediction

(26)

Wooldridge test T-statistic P-value Conclusion

Cr_US 30.669 0.000 First-order

autocorrelation

Housecr_US 23.965 0.001 First-order

autocorrelation Table 3.7 Test statistics Breusch and Pagan test

3.3 Econometric methodology

This paragraph explains the models that investigate the effect of credit growth in the U.S. on the Asian current account balance. Paragraph 3.3.1 explains and presents the models when the full sample is used. Secondly, paragraph 3.3.2 explains the models for the sample with Asian countries. Note, that they are very similar. Therefore, paragraph 3.3.1 is more elaborated than paragraph 3.3.2.

3.3.1 Full sample

This paper follows the empirical model of Büyükkarabacak and Krause (2009). The authors use a system Generalized Method of Moments (GMM) model. This model estimates a system of equations in the first-differences and levels. Lagged levels of the current account balance (relative to GDP) are used as instruments for the equations in the difference GMM model (first-differences). Besides, the lagged differences are used as instruments for the equations in levels. A reason to choose for a dynamic model is that there is a serial correlation for both the first and second hypotheses. Serial correlation might indicate that the relationship is dynamic instead of static. In a dynamic model, the dependent variable is also used as an explanatory variable. The lag of the dependent variable is a determinant for the value of the dependent variable of the current year. This could be logical in case of the current account balance. In addition, many economic relationships are dynamic by nature. The equilibrium value is not immediately reached in the dynamic adjustment process. For the baseline model, static models are not appropriate because those estimators would be biased and inefficient when the relationship is actually dynamic.

(27)

causality. In order to examine the first and second hypotheses, the following equation is derived from the theoretical framework and is used to investigate the dynamic relationship.

CAi,t = ∂CAi, t-1 + ß1*X t-1+ ß2*ADi,t + ß3(X t-1*AD i,t) + ß4*Control variables i,t + ei,t (1)

Where CA is the current account balance relative to GDP and has the subscriptions i and t. These subscriptions indicate that the current account differs per country (i) and per year (t). Then, the dependent variable is also stated on the right-hand side of the equation. The current account of the previous year determines the current account of this year. In addition, X is equal to domestic private credit in the U.S. (hypothesis I) or credit to households in the U.S. (hypothesis II). This variable does not contain the subscript i because the variable of interest includes only one country. Furthermore, a lag is used in order to deal with an endogeneity problem. The third parameter is AD that stands for Asian Dummy and equals one if the country in the dataset is an Asian country. The fourth parameter is an interaction variable that includes the Asian dummy and domestic private credit in the U.S. This interaction examines whether there is a specific relationship between the U.S. and Asia or not. The last parameter is based on a set of control variables. The Arellano-Bond test and the Hansen test must be performed in order to determine the validity of the model. The Arellano-Bond test determines serial correlation in the differences of the residuals. According to Roodman (2009), in order to check for first-order serial correlation in levels, second-order correlation in differences has to be examined. The null hypothesis of the Arellano-Bond test is equal to the case when there is no serial correlation in the first-differenced errors. Therefore, one does not want to reject the null hypothesis of this test. However, the null hypothesis for the first system GMM model cannot be rejected. Therefore, in the regression, the third until the ninth lags are used for the variable of interest, domestic private credit, GDP per capita and the exchange rate. These variables are endogenous because there could be reversed causality with the current account balance relative to GDP. In addition, the Hansen test whether overidentifying restrictions are valid. The Hansen test cannot be rejected for all regressions, which indicates that the instruments are valid. The expectation is that the coefficient of domestic private credit in the U.S. is positive. This implies that an increase in credit in the U.S. has a positive effect on the current account elsewhere in the world. In addition, the expectation is that the coefficient of the interaction variable is positive. This implies that an increase in credit in the U.S. has a positive effect on the current account of Asian countries. Lastly, the expectation is that credit to households has a higher impact on the current account balance than domestic private credit. The preferred outcome is that the interaction variable is significant. This would show the specific relationship between the U.S. and Asian countries.

3.3.2 Sample Asian countries

Another way of testing the relationship between credit growth in the U.S. and the current account balance of Asian countries is by using a sample that only includes Asian countries. In order to test the first hypothesis, the following equation is derived.

(28)

The equation is similar to equation one. However, the Asian dummy variable and the interaction are not needed in this dataset. The expectations for the Asian sample are similar to the expectations based on the full sample. Of course, the dummy for Asian countries and the interaction term are removed for this dataset. The expectation is that both domestic private credit in the U.S. and credit to households in the U.S. have a positive impact on the current account balance in Asian countries.

4. Results

Paragraph 4.1 discusses the results based on the full sample, where paragraph 4.1.1 examines the results based on the variable domestic private credit and paragraph 4.1.2 examines the results based on the variable credit to households. Furthermore, paragraph 4.2 discusses the results based on the Asian sample. This paragraph is also divided in results based on domestic private credit and credit to households. The tables from A.8 until A.11, in the appendix, represent the estimates. Note that all statistically significant effects are a ‘lagged’ values. This implies that the values of the interaction variables, domestic private credit in the U.S., and credit to households in the U.S. of previous year have an effect on the current account balance of this year. In addition, some variables are significant for a ten percent significance level. The effect is explained for some important variables, but for the conclusion of this paper, only estimates that have a significance level of five or one are used.

4.1 Results using full sample

4.1.1 Results based on domestic private credit

(29)

4.1.2 Results based on credit to households

The results based on credit to households are represented in table A.9 in the appendix. The model in which the control variables are included (column two) shows that the lag of the current account balance is negative and highly significant. This is striking because no other lag of the current account balance is negative. Furthermore, the model confirms the specific relationship between credit growth in the U.S. and the current account balance of Asian countries. The variable credit to households in the U.S. is not significant, which indicates that when households in the U.S. receive one percent more credit, the current account balance elsewhere in the world does not statistically change. However, the interaction variable is highly significant. Therefore, a one percent increase in the credit (growth) to households in the U.S. increases the current account of Asian countries with 0.866 percentage points. According to the literature, this positive effect was expected. In addition, the coefficient of the interaction term with domestic private credit is smaller than the interaction term with credit to households in the U.S. This implies that a one percent increase in credit growth to households has a larger and positive impact on Asian current account balances compared to domestic private credit in the U.S. This result was expected and confirmed hypothesis two. Note that the number of observations is quite low in this model. This might affect the reliability of these results. Therefore, chapter five includes a robustness check of this effect. Lastly, the dummy for Asian countries is not statistically significant.

4.2 Results using Asian sample

4.2.1 Results based on domestic private credit

The estimates of the models are represented in table A.10 in the appendix. In both models, the lag of the current account balance is statistically significant. This indicates that the level of this year depends on the previous year. The effect is higher (0.189 percentage points) than in the model when the full sample is used. By adding control variables in the second dynamic model, the lag of domestic private credit growth in the U.S. becomes significant for a ten percent significance level. This indicates that a one percent increase in domestic private credit growth in the U.S. increases the current account in Asia with 0.159 percentage points. Note that the number of observations dropped by approximately 300 observations in the second model. This is due to the missing values of the control variables that are added in the second model. The variable ‘terms of trade’ has a higher coefficient and has a higher level of significance than domestic private credit in the U.S. This implies that terms of trade have a higher impact on the Asian current account balance in this model.

4.2.2 Results based on credit to households

(30)

one-step is consistent but inefficient. Inefficiency implies that the estimated coefficients are less precise and that the stand errors are not as low as possible. However, efficiency is desirable but not that important as consistency in the model because, in the first place, a model has to reliable. The estimates based on credit to households are represented in table A.11 in the appendix. The lag of credit growth to households in the U.S. is significant for a ten percent significance level. The effect implies that a one percent increase in credit growth to households in the U.S. increases the Asian current account with 0.194 percentage points. Furthermore, the control variable credit to households caused a massive decline in the number of observations in the second model. Leaving this variable out of the regression might not be a good solution because it is a very important control variable. The number of observations would increase though (up to 472 observations) and would give the variable of interest a higher level of significance. Chapter five examines the robustness of this finding by using several static models.

One limitation of the results is that the estimated coefficients contain time fixed effects. The variable of interest contains data of the U.S. only. Whereas the dependent variable consists of multiple countries. However, in the way the system GMM models are done, is the best possible way in order to examine the relationship. The fitted regression lines are plotted in appendices I and J. The y-axis represents domestic private credit growth in the U.S. (or credit growth to households in the U.S. or the interaction term). The x-axis represents the current account balances relative to GDP.

5. Robustness

This chapter performs some robustness regressions in order to check the validity of the initial estimations. Paragraph 5.1 discusses the robustness checks for the full sample and paragraph 5.2. discusses the robustness checks for the Asian sample. First, several models are explained, where after the results are represented. The first robustness check is a fixed effects and/or random effects model. The second robustness check is an instrumental variable model (2SLS) that uses the lag of the variables of interest as an instrument. These models are static models. Whereas the baseline model of this paper is a dynamic model. Static models are used as robustness checks because they require less data. Especially many observations are lost by examining the second hypothesis. Besides, some researchers argue that static models are more reliable than dynamic models.

5.1 Robustness full sample

Fixed effects and random effects

(31)

a p-value of 0.000. This implies that at least one of the years is different from zero. A fixed effects model estimates the explanatory variable’s effect by looking at the variation with each country. The variable of interest varies over time, but this variation is for each country equal. This is due to the fact that the variable contains one country. Whereas the dependent variable contains 140 countries. Therefore, time fixed effects are included in order to control for the variable of interest.

CAi,t = 𝛼i + ß1*Cr_US + ß2*ADi,t + ß3(Cr_USt*ADi,t) + ß4*CV + d2*D2i + … + dn*Dn + ei,t (3)

In this model, the subscript i is added after the parameter for the intercept because individual heterogeneity is captured by the intercept. Therefore, the dummy variables that are constant over time are omitted from the regression because those effects are constant (fixed) over time. Secondly, Cr_US stands for domestic private credit in the U.S and AD is the dummy for Asian countries. Third, CV is a set of control variables. Lastly, dn*Dn are year dummies for 139 countries and 35 years instead of 140 countries in order to exclude a dummy variable trap.

Hausman test T-statistic p-value Conclusion

Cr_US 25.96 0.009 Fixed effects

Housecr_US 35.35 0.970 Random effects

Table 3.5 Test statistics Hausman test for baseline model - full sample

For the regression based on credit to households, the Hausman test indicates that the random effects model is most appropriate to use. In addition, the Breusch and Pagan test generates a p-value of 0.000. This confirms that random effects are required for this regression. The random effects model assumes that the differences between countries are random. Carter Hill et al. (2012) mention that “random individual differences can be included in our model by specifying the intercept parameters ß1 to consist of a fixed part that represents the population average, and random individual differences from the population average.” For the second hypothesis, equation four is derived.

CAit = 𝛼i + ß1*HC_USt + ß2*ADit + ß3(HC_US t *AD it) + ß4*Control variablesit + eit (4)

(32)

2SLS

In order to deal with reverse causality, a 2SLS model could be used. The instruments that are used are the lag values of the variables of interest. These lags have to be exogenous, relevant and do not affect directly the current account. Exogeneity can be tested with a Hansen J test, but only when the number of instruments exceeds the number of endogenous variables. This is not true for the regressions in this paper. The relevancy of the instrument is tested by performing the first stage of the 2SLS model. Then, the Kleinbergen-Paap rk Wald F-statistic determines whether the instrument is relevant. The F-statistic has to be larger than ten, which is true for both the lags of domestic private credit in the U.S. and credit to households in the U.S. Besides, the instrument variables and the endogenous variables have to be strongly correlated. The 2SLS model can be explained by showing the first and second stage of the model. The first stage includes a regression of the endogenous variable on the instrument and the exogenous explanatory variable. As a result, the following equation is obtained where X is either equal to domestic private credit in the U.S. or credit to households in the U.S.

Xi,t = γ1+ γ2*X t-1 + γ3*Control variables i,t + υ (5)

In the second stage, the predicted value (x-hat) of the endogenous variable x is used instead of the initial value in the main regression equation. As a result, the following equation presents the second stage of the 2SLS model.

CAit = 𝛼+ ß1*X t + ß2*AD + ß3(X t-1*AD i,t) + ß4*Control variables i, t-1 + ei,t (6) Results based on domestic private credit

(33)

Results based on credit to households

The first model is a random effects model that generates insignificant results for credit to households in the U.S. and the interaction variable. These results are equal to the baseline model, except for the interaction term that is not significant anymore. In the fixed effects model with control variables, credit to households in the U.S. becomes significant for a five percent significance level. However, the coefficient is quite large, namely 1.049. This might be due to high correlations with the year dummies that control for year fixed effects. However, the estimates would be still the best linear unbiased estimators. Another consequence might be that the standard errors are larger than usual. Furthermore, the interaction variable is insignificant. On the other hand, the 2SLS model with control variables shows a weakly significant coefficient for the interaction term based on the lag of credit to households in the U.S. However, the dummy for Asian countries is high significant and has a coefficient of 3.880. This implies that the current account balance increases with 3.880 percentage points for Asian countries. One limitation is that the number of observations declines dramatically by adding control variables to the model. This affects the reliability of the results. The results are represented in table A.13 in the appendix. In sum, the robustness regressions based on the full sample do not generate strong enough results in order to conclude that the second hypothesis is true.

5.2 Robustness Asian sample

Fixed effects and random effects

For both hypotheses, the Hausman test is insignificant. In addition, the Breusch and Pagan tests are significant for both hypotheses. This indicates that the random effects model is the most appropriate model to use. The model that is used, is derived in equation seven. Where X is equal to domestic private credit in the U.S. for hypothesis I and equal to credit to households in the U.S. for hypothesis II.

CA i,t = 𝛼i + ß1*X t + ß2*Control variables i,t + e i,t (7)

Hausman test T-statistic p-value Conclusion

Cr_US 6.74 0.456 Random effects

Housecr_US 8.85 0.355 Random effects

Table 3.10 Test statistics Hausman test for baseline model – Asian sample

The Effect of Credit Growth in the U.S. on the Asian Current Account Balance