Essays on Applied International Economics and Finance by Pramesti Resiandini A dissertation submitted to the Graduate Faculty of Auburn University in partial fulfillment of the requirements for the Degree of Doctor of Philosophy Auburn, Alabama May 9, 2011 Keywords: International trade, international finance, exchange rate Copyright 2011 by Pramesti Resiandini Approved by Henry Thompson, Co-chair, Professor of Agricultural Economics and Rural Sociology James R. Barth, Co-chair, Professor of Finance Hyeongwoo Kim, Assistant Professor of Economics ii Abstract Studies in international economics and finance have become increasingly important in understanding the ways economies are integrated. This dissertation comprises three essays on this topic. The first essay investigates the relationship between financial development and trade based on panel data of bilateral trade between the world?s three largest economies (United States, Japan, and Germany) and 47 partner countries. The second essay estimates common factors from a monthly panel of 51 commodity prices and further analyzes the most important common factor, which appears to be correlated with the U.S. dollar nominal exchange rates. Lastly, the third essay examines data on Japanese and Korean automobile exports to the United States to determine consistency with the Alchian-Allen theorem. iii Acknowledgments I am grateful to my dissertation committee, Dr. Henry Thompson (co-chair), Dr. James R. Barth (co-chair), and Dr. Hyeongwoo Kim, for their support and advice. I am also thankful to Dr. Daniel Gropper (College of Business) as the outside reader of this dissertation. I gratefully acknowledge Graduate Research Assistantship from the College of Business. I thank Dr. John D. Jackson (Department of Economics) and Dr. Shu-Ling Chen (National Tsing Hua University), who co-author ?What Drives Commodity Prices?? with Dr. Hyeongwoo Kim and me. I thank Dr. Robert Lawson (Department of Finance) for valuable information on some of the data I use in this dissertation. Different parts of this dissertation also benefited from helpful comments and suggestions from Dr. Giap Nguyen, Dr. Jennings Byrd, Malick Diarrassouba, and Thomas Doran. All remaining errors are mine. I appreciate my friendship with my colleagues: Ermanno Affuso, Nathan Forczyk, Dengjun Zhang, and others who I have not mentioned. Special thanks to Kellie Goodwin, a fellow classmate in the MBA program, for her friendship over the years. Finally, I am grateful to my family, Parangtopo Sutokusumo family, for their continuous support and encouragements. iv Table of Contents Abstract ......................................................................................................................................... ii Acknowledgments........................................................................................................................ iii List of Tables ............................................................................................................................... vi List of Figures ............................................................................................................................. vii Chapter 1: Financial Development and Trade ............................................................................. 1 1.1 Introduction ................................................................................................................. 2 1.2 Data ............................................................................................................................. 3 1.3 Literature review ....................................................................................................... 5 1.4 Gravity model ........................................................................................................... 9 1.5 Empirical results ....................................................................................................... 14 1.6 Conclusion .............................................................................................................. 19 Chapter 2: Commodity Prices and the Exchange Rate .............................................................. 21 2.1 Introduction ............................................................................................................. 22 2.2 Data ......................................................................................................................... 24 2.3 Literature review ..................................................................................................... 25 2.4 Econometric techniques .......................................................................................... 27 2.5 Empirical results ....................................................................................................... 32 2.6 Conclusion .............................................................................................................. 35 v Chapter 3: Japanese and Korean Automobile Exports to the United States .............................. 37 3.1 Introduction ............................................................................................................. 38 3.2 Literature review ..................................................................................................... 40 3.3 Data ......................................................................................................................... 42 3.4 The shipping model ................................................................................................. 44 3.5 Empirical method and results .................................................................................. 49 3.6 Conclusion .............................................................................................................. 51 References ................................................................................................................................. 53 Appendix 1: Tables and Figures for Chapter 1 .......................................................................... 59 Appendix 2: Tables and Figures for Chapter 2 .......................................................................... 68 Appendix 3: Tables and Figures for Chapter 3 .......................................................................... 84 vi List of Tables Table 1.1 List of countries .......................................................................................................... 61 Table 1.2 Variable descriptions and summary statistics ............................................................. 62 Table 1.3 Pair-wise correlations between variables .................................................................... 63 Table 1.4 Regression results ....................................................................................................... 64 Table 1.5 Fixed effects regressions without pair-country interactions ....................................... 67 Table 2.1 Commodity price data description .............................................................................. 69 Table 2.2 ADF test and CD test results: Nominal commodity prices ......................................... 71 Table 2.3 ADF test and CD test results: Relative commodity prices ......................................... 72 Table 2.4 PANIC test results: Nominal commodity prices......................................................... 73 Table 2.5 PANIC test results: Relative commodity prices ......................................................... 74 Table 2.6 Out-of-sample forecast performance: Nominal commodity price factors .................. 75 Table 2.7 Out-of-sample forecast performance: Relative commodity price factors ................... 75 Table 3.1 Import quantity and price data .................................................................................... 85 Table 3.2 Augmented Dickey-Fuller (ADF) test results ............................................................. 90 Table 3.3 Regressions and cointegration test results (1995m1-2005m12) ................................. 91 Table 3.4 Regressions and cointegration test results for higher quality cars, extending the period of observations (1995m1-2008m12) ......................................... 92 Table 3.5 Stock-Watson dynamic OLS regression results (1995m1-2005m12) ........................ 93 Table 3.6 Stock-Watson dynamic OLS regression results for higher quality cars, extending the period of observations (1995m1-2005m12) ......................................... 94 vii List of Figures Figure 1.1 Sources of funds for a corporation ............................................................................ 60 Figure 2.1 Number of factors estimation: Nominal commodity prices ...................................... 76 Figure 2.2 Number of factors estimation: Relative commodity prices ....................................... 77 Figure 2.3 Relative importance of common factors: Nominal commodity prices ...................... 78 Figure 2.4 Relative importance of common factors: Relative commodity prices ...................... 79 Figure 2.5 Factor loadings estimates: Nominal commodity prices............................................. 80 Figure 2.6 Factor loadings estimates: Relative commodity prices ............................................. 81 Figure 2.7 U.S. nominal exchange rates vs. common factor estimates ...................................... 82 Figure 2.8 Crude oil prices.......................................................................................................... 83 Figure 3.1 Import quantity data series ........................................................................................ 86 Figure 3.2 Car price data series................................................................................................... 87 Figure 3.3 Actual shipping cost data series ................................................................................ 88 Figure 3.4 Other data series ........................................................................................................ 89 1 CHAPTER 1 Financial Development and Trade: Evidence from the World?s Three Largest Economies 2 1.1 Introduction Numerous studies examine the relationship between international trade and economic growth, as well as between financial development and economic growth.1 The importance of trade and financial development in the growth literature provides motivation to study the relationship between the two. Many of the existing studies on this issue examine individually specific measures of financial development, e.g., private credit and foreign investment.2 The present study aims to contribute to the literature by investigating several measures of financial development with a gravity model of bilateral trade between the world?s three largest economies (United States, Japan, and Germany) and their 47 major trading partners from 2003 to 2007. This paper examines access to external funds and international financial indicators as the measures of financial development. The main hypothesis tested is that financial development in a country relates to the degree of bilateral trade with its trading partners. A gravity model is constructed and estimated with a fixed effects method. The model includes three variables commonly used in a gravity equation: distance between pair countries, land common border, and stage of development. Ease of access to loans and ease of access to the local equity market represent access to external funds. Three international financial indicators are included: country credit ratings, international capital market controls, and real exchange rates. A country?s credit rating and capital controls capture its access to foreign capital, while the real exchange rate plays a role in determining profitability of tradables. 1 Frankel and Romer (1999), among others, find a positive relationship between international trade and economic growth. Several studies suggesting the importance of financial development for economic growth are Levine and Zervos (1998), Levine, Loayza, and Beck (2000), and Demirg??-Kunt and Levine (2001). 2 The World Economic Forum (2009, p. 3) defines financial development as ?the factors, policies, and institutions that lead to effective financial intermediation and markets, as well as deep and broad access to capital and financial services.? 3 Throughout this paper, the term ?main countries? will refer to the three largest economies, and ?partner countries? will refer to the 47 trading partners. The study finds differences in the degree of bilateral trade when grouping partner countries into developed and less developed countries. Access to loans for businesses shows a strong positive relationship with bilateral trade. Access to the local equity market is negatively related to trade with developed countries, but positively related to trade with less developed countries. The study also finds that country credit ratings, international capital market controls, and real exchange rates are significant determinants of trade. The study suggests the roles policy can play in promoting both trade and development. The remainder of the paper proceeds as follows. Section 2 describes the data. Section 3 provides literature review. Section 4 and Section 5 discuss the empirical methods and results. Section 6 concludes the paper with a summary of findings. 1.2 Data The dataset is a balanced panel of bilateral trade between the three main countries (United States, Japan, and Germany) and 47 partner countries over the years 2003 to 2007 for a total of 690 observations. Sample selection is based on the average levels of GDP of IMF reporting countries over the five year period and availability of other data. The top three countries based on the average GDP are selected as the main countries and the rest as partner countries, shown in Table 1.1. The total trade between the main countries and partner countries during the period represents about 70% of total trade of the three main countries. [Table 1.1 here] 4 The major data source in this study is the International Monetary Fund (IMF) databases. The bilateral trade data are obtained from the IMF?s Direction of Trade (DOT) database, the GDP data are from the IMF?s World Economic Outlook (WEO) database, and the real effective exchange rate data are from the IMF?s International Financial Statistics (IFS) database. The IMF trade data follows United Nation?s guidelines, which sufficiently covers all merchandise entering or leaving a country, except goods in transit. The IFS database provides an indicator of real effective exchange rates based on relative consumer prices, allowing for comparison with a broad range of partner countries. The weighting method is based on disaggregated trade flows for manufactured goods and primary products over the period 1999 to 2001 (IMF, 2009). Another data source is the World Economic Forum?s (WEF) Global Competitiveness Reports (2003-2007). The WEF reports the indices of access to loans (Ease of Access to Loan) and access to the local equity markets (Ease of Access to Equity). These indices are scaled 1 to 7 with higher scale indicating easier access to obtain loans or to raise capital through the local equity market. They are based on surveys by the WEF on the opinions of business executives worldwide. The same report contains the Stages of Development index, dividing countries into different stages of development (Stages 1 to 3) based on the real GDP per capita. The last index is used in this study as a basis for grouping countries into developed countries (Stage 3: GDP per capita > USD 17,000) and less developed countries (Stage 1 and Stage 2: GDP per capita ? USD 17,000). The Economic Freedom of the World Report (Gwartney and Lawson, 2009) provides data on international capital market controls. The related index is scaled 1 to 10 with higher scale indicating more open capital flow policy. It is constructed based on two sub indices: (1) foreign ownership/investment restrictions, and (2) capital controls. The sources of the two sub 5 indices are the WEF?s Global Competitiveness Report and the IMF?s Annual Report on Exchange Arrangements and Exchange Restrictions. The data on country credit ratings are from various editions of the Institutional Investor (2003-2007). The creditworthiness of each country is evaluated on a scale of 1 to 100 with higher scale indicating higher rating. The rating is constructed based on a compilation of economic, financial, and political indicators assessed by senior economists and risk analysts worldwide. The data on distances are from Centre d'Etudes Prospectives et d'Informations Internationales (CEPII) Geodesic distances database (2010). CEPII calculates geodesic distances with latitudes and longitudes of the most important cities/agglomerations in terms of population. 1.3 Literature review (a) Finance and trade literature The literature indicates that there is a significant relationship between financial development and trade. Beck (2002) provides a theoretical model examining the relationship between financial development and trade, focusing on the role played by financial intermediaries in facilitating high-return manufacturing projects. Using panel data on private credit for 65 countries over a 30-year period, the study finds that financial development strongly affects export volume and trade balance of manufacturing industries (2002, p. 107). In a subsequent paper, Beck (2003) finds that countries in which financial systems are relatively highly developed tend to have higher export shares and trade balances when industries rely more on external finance. 6 A broad range of literature link international capital flows to trade. Some of the studies use macroeconomic models, commonly based on the adjustments to changes in trade regulations or capital market controls. McKinnon (1993) suggests the importance of capital market controls in determining trade flows. On the other hand, Mundell (1957) examines the effect of trade openness on capital flows. Based on a 2x2 Heckscher-Ohlin factor endowment model, Mundell suggests that barriers to trade encourage international capital flows that if unhindered increase the output of the host country?s import-competing sector. Several recent papers on this topic relate trade openness to the stability of capital flows. Cavallo and Frankel (2008) provide empirical evidence that trade openness makes countries less susceptible to sudden stops in capital inflows and thus less susceptible to crises. Other studies on trade and capital flows utilize microeconomic models, focusing on firm- level problems such as costs and sales. These studies examine specifically the relationship between trade and foreign direct investment (FDI). A model of trade-FDI relationship is known as the ?Proximity-Concentration Tradeoff? suggesting that multinational companies choose exporting over FDI when they face higher fixed costs in the host countries than trade costs, but choose FDI over exporting when the trade costs are higher than fixed costs (Smith [1987] and Neary [2002]). It assumes that trade and FDI are substitutes. Neary (2009) argues, however, that the proximity-concentration tradeoff applies to horizontal FDI only. A vertical FDI model suggests that trade and FDI can be complements if countries differ either in technology or endowments of specific factors (Markunsen [1983] and Neary [1995]). Trade liberalization can encourage FDI if the induced capital flows lead to an increase in production of the host country?s exporting sector. Helpman (1984) utilizes a Heckscher-Ohlin 7 model to make a point that when stages of production vary in factor intensities, differences in factor endowments between countries may encourage vertical disintegration by firms. Based on data of U.S. capital outflows, Ruffin and Rassekh (1986) argue that foreign direct investment and portfolio investment are perfect substitutes. They note that the way multinational corporations finance their operations may be unrelated to the net flow of capital between countries (p. 1126). A recent article on foreign direct investment and portfolio investment is Goldstein and Razin (2005). Using a model of information-based tradeoffs between direct investment and portfolio investment, they compare the expected yields on the two types of foreign investment. Goldstein and Razin point out that developed countries attract more portfolio investment relative to direct investment than less developed countries due to the higher fixed costs associated with direct investment in developed countries. They also note that the high levels of transparency make portfolio investment in developed countries more efficient. Real exchange rate and its role in determining the profitability of tradables have attracted more attention in the recent economic growth literature. Freund and Pierola (2008) find that a currency undervaluation and a decrease in exchange rate volatility increase exports in developing countries. They also find that exchange rate volatility has less impact on exports in developed countries. It is often suggested that exchange rate volatility reduces trade; however, some empirical studies find that the negative relationship is not robust to some specifications (Clark, Tamirisa, and Wei [2004]; Baxter and Kouparitsas [2006]). Furthermore, Gala (2008, p. 279) points out that ?currency misalignment measures are far from consensual? and contributes to the literature by theoretically and empirically connecting real exchange rate levels and economic development. Gala finds that real exchange rate levels could affect economic growth through capital accumulation and technological development. 8 (b) Gravity literature Gravity models have been widely used in examining international trade flows. Proposed by Tinbergen (1962), the gravity equation is later developed by several authors who provide the microfoundations (e.g., Anderson [1979], Hepman [1987], Anderson and van Wincoop [2003], and Feenstra [2004]). The gravity model predicts that trade flows between two countries are inversely related to the physical distance between them and directly related to the multiplicative interaction of each country?s size, which is commonly measured by GDP. Anderson (1979) develops a microfoundation for the gravity model using constant elasticity of substitution (CES) expenditure system. It proves algebraically that level of trade is related to the size (income) of countries as well as trade costs. Trade costs include: (1) transport costs as an increasing function of distance, and (2) tariffs. Another application of the gravity equation widely cited in subsequent papers is an empirical work by McCallum (1995) that finds that national borders matter for trade flows. Feenstra (2004) provides a theoretical model explaining border effects, such as transport costs or tariffs, using the CES utility function in a consumer utility maximization problem. Anderson and van Wincoop (2003) point out the role of prices in the applications of gravity equation for international trade. They find that ignoring prices in the cross-section gravity equation generates omitted variables bias. Using the general equilibrium structure of the model, they calculate the comparative statics of trade barriers and propose a framework that counts for ?multilateral price resistance.? They use a nonlinear least squares program to estimate the multilateral price resistance terms. Feenstra (2004) suggests an alternative method to estimate the multilateral price resistance terms in cross section by including country-specific fixed effects, controlling for the 9 effects of exporters by time as well as importers by time. After implementing the country- specific fixed effects, Baier and Bergstrand (2007) find that there is still a large amount of unobserved heterogeneity among country pairs, and thus they add country-pair fixed effects. The country-pair fixed effects capture the effects of free trade agreements (FTA) on trade. Baltagi, Egger, and Pfaffermayr (2003) note that a full interaction effect may be used in examining bilateral trade flows. The full interaction effects consist of country-specific fixed effects, country-pair fixed effects, as well as fixed exporter, importer, and time effects. The robustness of various variables in a gravity model is examined by Baxter and Koupartisas (2006). They utilize three methods for determining robust relationships: the extreme bounds analysis of Leamer (1983, 1985), the extreme bounds analysis of Sala-i-Martin (1997), and the general to specific approach of Hendry (1995). They find that common gravity variables are robust determinant of trade. These variables are distance, common border, cultural distance, and colonial ties. Other variables tested produce different results. Bilateral factor endowment and stage of development are robust determinants of trade when tested using all three methods. Industrial similarity, restrictions to flows of goods and capital, currency union membership, and exchange rate volatility show robustness under only one of the testing methods or when tested with a restricted sample. 1.4 Gravity model The bilateral trade model is specified with trade flows between country i and country j (Tij) directly related to the multiplicative interaction of each country?s size and inversely related to the physical distance between them, 10 Tij = ?o Yi?1Yj?2 Dij?3?ij , (1) where Yi(j) denotes GDP of country i(j) and Dij denotes the distance between country i and j; ?s are unknown parameters, and ?ij is a stochastic error term. Taking natural logarithms of both sides, the gravity equation can be estimated. ln Tij = ?o+ ?1 lnYi + ?2 lnYj + ?3 lnDij + ?ij (2) Adding Zij and Wij as continuous variables and dummy variables of interest in the study, the gravity equation is rewritten as follows. ln Tij = ?o+ ?1 lnYi + ?2 lnYj + ?3 lnDij + ? lnZij + ?Wij + ?ij (3) Following Anderson and van Wincoop (2003), a framework that accounts for multilateral (price) resistance terms is built by including Pi1-? and Pj 1-?, main country and partner country price indices. The model proposes that the degree of trade is determined not only by the national borders between country i and country j but also by the multilateral resistance from their trading partners in the rest of the world. It also moves the GDP terms from the right to the left side of the gravity equation. ln [Tij/(YiYj)] = ?o+ ?3 lnDij + ? lnZij + ? Wij ? ln Pi 1-?? ln Pj 1-? + ?ij (4) where, g1842g3036g2869g2879g3097 g3404 ? g1842g3036g3097g2879g2869g4672g3026g3284g3026 g3298 g4673g3015g3036g2880g2869 g1857g3081g2871 g3039g3041 g3005g3036g3037 g2878 g3082g3039g3041 g3027g3036g3037 g2878 g3083 g3024g3036g3037 (5) with i = 1?N equilibrium conditions; Yw denotes world GDP (constant across countries); and ? is the elasticity of substitution between countries. This paper follows the measure of bilateral trade by Helpman (1987) and Feenstra (2004). In a panel setting, they define bilateral trade between country i and country j in period t, Tijt, as: 11 Tijt= ln(Xijt + Xjit), where Xijt denotes exports from country i to country j in period t and Xjit denotes the reverse. The conceptual model is as follows. Trade = f(Gravity, External Funds, International Finance, Other) (6) where, Trade is the level of trade between pair countries relative to their level of GDP. Gravity is a group of variables commonly used in gravity equations. External Funds is a group of variables indicating access to external funds. International Finance is a group of international financial indicators. Other is a group of other variables controlling for fixed effects. [Figure 1.1 here] The variables of interests of this study are the financial development variables, represented by Zij in (4). These variables comprise the External Funds variables (access to loans and access to the local equity market) and the International Finance variables (country credit ratings, international capital market controls, and the real exchange rates). Dummy variables commonly used in gravity equations are represented by Wij in (4). These variables are land common border and stage of development. The present study uses a fixed effects method to estimate the gravity equation, controlling for country-specific (by time) fixed effects and country-pair fixed effects. The country-specific fixed effects count for the multilateral resistance terms as suggested by Anderson and van 12 Wincoop (2003) and Feenstra (2004), whereas the country-pair fixed effects capture the effects of free trade agreements (FTA) as suggested by Baier and Bergstrand (2007). The gravity equation for trade and financial development is specified as follows. TRAijt = ?o + ?1DISij + ?2LCBij + ?3DEVijt + ?4LNAijt + ?5EQAijt + ?6CCRijt + ?7CAPijt + ?8RERijt + ?it + ?jt + ?ij + ?ijt (7) where i denotes main countries, j denotes partner countries, t denotes time; and the variables are described as follows. TRAijt = log of trade between i and j relative to the product of their levels of GDP. DISij = log of the distance between i and j. LCBij = a dummy variable which is unity if i and j share a common land border. DEVijt = a dummy variable which is unity if i and j are in the same stage of development. LNAijt = log of the product of the scores of access to loans in i and j. EQAijt = log of the product of the scores of access to local equity market in i and j. CCRijt = the difference between the log of the maximum of the country credit ratings of i and j and the log of the minimum of the country credit ratings of i and j. CAPijt = log of the product of the scores of capital controls in i and j. RERijt = the difference between the log of the maximum of the real effective exchange rates of i and j and the log of the minimum of the real effective exchange rates of 13 i and j.3 ?it = interactions between country i and year t dummies ?jt = interactions between country j and year t dummies ?ij = interactions between country i and country j dummies ?ijt = other influences on bilateral trade The financial development variables are measured with bilateral interactions similar to those of the common gravity variables. Two forms of bilateral interaction are constructed for the financial development variables following the methods used by Baxter and Kouparitsas (2006) in measuring factor intensity. The first measure is the log of the product of the levels of ?factor? in the two countries, Bilateral interaction 1: ln(Fit * Fjt) (8) where Fit(jt) stands for the factor in country i(j) in period t. This measure of ?scale? can be interpreted as: ?the higher the measure, the higher the levels of factor in the two countries or in either of the two countries.? It also suggests that ?the higher the measure, the more equal the levels of factor in the two countries.? This interaction term is applied to the measures of access to loans (LNA), access to equity market (EQA), and international capital market controls (CAP). The second measure is the difference between the log of the maximum and the log of the minimum levels of factor in the two countries, or 3 Real exchange rate is defined as: g1844g1831g1844 g3404g1857g3400g1842 g3007g3004 g1842g3005g3004? , where g1857 is nominal exchange rate (g1830g1829 g1832g1829? ) and g1842g3007g3004 g1842g3005g3004? is the ratio of price levels in the two countries. g1830g1829 denotes domestic currency, and g1832g1829denotes foreign currency. However, the data used are real effective exchange rates, the averages of bilateral real exchange rates between the country and each of its trading partners, weighted by the respective trade shares of each partner. The IMF?s real effective exchange rates index (based on relative consumer prices) is used to allow for comparison with a broad range of partner countries. 14 Bilateral interaction 2: ln[max(Fit, Fjt)/min(Fit, Fjt)] (9) This measure can be interpreted as follows: ?the larger the measure, the larger the difference in the levels of factors in the two countries.? This interaction term is applied to the measures of country credit ratings (CCR) and real exchange rates (RER). Equation (8) is based on a similar method of measuring the scale of GDP for pair countries, whereas equation (9) is often used in gravity literature to measure the differences in GDP per capita between pair countries. Baxter and Kouparitsas (2006) also use both equations as the measures of human capital. [Table 1.2 here] 1.5 Empirical results The gravity model described in Section 1.4, based on Anderson and van Wincoop (2003) and Baier and Bergstrand (2007), provides estimation results as presented in Table 1.4(a). As comparisons, Table 1.4(b) and Table 1.4(c) list the regression results obtained from using only country-specific by time fixed effects (Anderson and van Wincoop, 2003) and Ordinary Least Squares (OLS), respectively. Ignoring country-pair fixed effects, the estimates in Table 1.4(b) generate lower values of adjusted R-squared compared to the estimations of the selected model (Table 1.4[a]). The estimates in Table 1.4(b) are biased due to the omission of the effects of FTA when in fact the countries are affected by at least two major FTAs (European Union and NAFTA). The OLS estimates (Table 1.4[c]) show the lowest values of adjusted R-squared among the three models. The regression for Dataset 1 shows that stage of development variable has a 15 significant negative relationship with bilateral trade at 1% level of significance. This result suggests that the dataset should be partitioned based on partner country?s stage of development: developed countries (Dataset 2) and less developed countries (Dataset 3). In addition, a Chow test is conducted, and the result rejects the null hypothesis of no differences between the two groups at 1% level of significance.4 Table 1.5 lists the estimation results obtained from using financial development variables without bilateral interactions (e.g., using LNAit and LNAjt instead of LNAijt). It implies that the coefficient estimate for a finance variable for country i(j) is generated by holding constant the same variable for country j(i) and other variables in the model. This model seems to work for data on less developed countries, but not on developed countries. [Tables 1.3, 1.4, and 1.5 here] The following paragraphs discuss separately and in details the regression results for developed and less developed countries. (a) Developed (partner) countries A developed (partner) country is a trading partner that shares the same stage of development as the three main countries, represented by a dummy variable that is in unity. Panel data of developed countries, named Dataset 2, comprise 318 observations (Table 1.4[a]). The adjusted R-squared shows that 95.8% of the variation in bilateral trade is explained by the model. An F-test is conducted for each group of variables and it shows that the Gravity variables (DIS and LCB) jointly contribute to bilateral trade at 1% level of significance. The External Funds 4Chow test, F-value = 3.27. 16 variables (LNA and EQA jointly) and the International Finance variables (CCR, CAP, and RER jointly) also significantly contribute to bilateral trade at 1% level. 5 Distance between pair countries has a negative relationship with bilateral trade at 1% level of significance. Main countries trade more with developed countries located closer to them. Land common border indicates a positive relationship with trade between developed countries, also significant at 1% level. Access to loans has a positive relationship with bilateral trade at 5% level of significance, suggesting that trade between countries increases when it is easier for businesses to obtain loans. Access to equity shows a negative relationship with bilateral trade at 1% level of significance, indicating that trade between countries decreases when it is easier for businesses to raise capital through the local equity markets. Since the pair countries in this data set are both developed countries, this result seems to be intuitively correct. The rationale for this result may relate to the degree of efficiency of the equity markets. In developed countries, the equity markets have high levels of transparency and thus more efficient than those in less developed countries (Goldstein and Razin, 2005). Easy access to highly efficient equity markets may encourage multinational companies to substitute trade with investment in local suppliers. Thus, easier access to equity market lowers trade. The country credit ratings variable indicates a strong negative relationship with bilateral trade at 1% level of significance. The larger the difference in credit ratings of the two countries, the less trade occurs between them. The capital controls variable also shows a strong negative relationship with bilateral trade at 1% level of significance. It suggests that trade between two 5 Gravity, F-value=83.58; External Funds, F-value=7.64; International Finance, F-value=8.50. 17 countries decreases when the countries relax their international capital market controls. This result indicates a substitution relationship between international trade and foreign investment in local companies, consistent with the horizontal FDI model proposed by Smith (1987) and Neary (2002). The real exchange rate variable is positively related to bilateral trade at 1% level of significance, indicating the larger the difference in real exchange rates of pair countries, the more trade occurs between them. (b) Less developed (partner) countries A less developed (partner) country is a trading partner that is in a lower stage of development than the three main countries. The panel data of less developed partner countries, named Dataset 3, comprise 372 observations (Table 1.4). The adjusted R-squared shows that 93.6% of the variation in bilateral trade is explained by the model. The F-test for each group of variables shows that Gravity variables (DIS and LCB) jointly contribute to bilateral trade at 1% level of significance. The External Funds variables (LNA and EQA jointly) and International Finance variables (CCR, CAP, and RER jointly) are also significant at 1% level.6 As expected, distance between pair countries shows a strong negative relationship with trade at 1% level of significance. Land common border, however, does not significantly contribute to bilateral trade when estimated using this data set. Access to loans has a positive relationship with bilateral trade at 1% level of significance. The degree of bilateral trade contributed by access to loans is higher when estimated using this data set compared to using Dataset 2. Access to equity also shows a positive relationship with 6 Gravity, F-value=86.76; External Funds, F-value=16.71; International Finance, F-value=48.87. 18 bilateral trade, significant at 5% level, indicating that trade increases when it is easier for businesses in the two countries to raise capital through the local equity market. The country credit ratings variable is negatively related to bilateral trade at 1% level of significance, whereas the capital controls variable is positively related to bilateral trade at 1% level of significance. In contrast to the negative result from Dataset 2, the positive coefficient suggests that trade between main countries and less developed partner countries increases when these countries relax their international capital market controls. This result is consistent with the vertical FDI model developed by Markunsen (1983) and Neary (1995). The real exchange rates variable shows a negative relationship with bilateral trade at 5% level of significance, indicating that the larger the difference in real exchange rates of pair countries, the less trade occurs between them. The regression without pair-country interaction terms for financial development variables reveals some interesting results (Table 1.5). An increase in real effective exchange rates in less developed countries seems to increase trade with main countries at 1% level of significance, holding constant real effective exchange rates in main countries. This result is consistent with the literature (Freund and Pierola, 2008). A real depreciation of currency in the main countries, however, decreases trade with less developed countries at 10% level of significance. Other coefficient estimates appear to be consistent with the estimation results from the model with interaction terms (Table 1.4a), except for access to loans variables (LNAit and LNAjt) that show insignificant results. 19 1.6 Conclusion This study investigates the relationship between financial development and trade using a gravity model with fixed effects based on Anderson and van Wincoop (2003) and Baier and Bergstrand (2007). The panel data of bilateral trade between the three main countries (United States, Japan, and Germany) and 47 partner countries covers the period 2003 to 2007. The study finds differences when grouping the partner countries into developed and less developed countries affecting the degree of bilateral trade. Consistent with the literature, physical distance is negatively related to trade between countries. In all cases, access to loans has a significant positive relationship with bilateral trade. Easy access to loans in less developed countries appears to contribute more to the bilateral trade than in developed countries. This result is expected since financial systems in less developed countries are heavily bank based. Access to equity shows a significant positive relationship with bilateral trade for less developed countries. Equity market development, as well as trade, is often promoted in less developed countries since financial systems in these countries are usually bank based rather than market based. In contrast, trade between main countries and other developed countries is lower with higher access to equity. A possible explanation is that the equity markets in developed countries have high levels of transparency and are more efficient than the equity markets in less developed countries. Easy access to highly efficient equity markets in developed countries may encourage multinational companies to substitute trade with investment in local suppliers, lowering trade. All three international financial indicators (country credit ratings, capital controls, real exchange rates) have significant relationships with bilateral trade. The three variables appear to 20 affect trade between main countries and less developed countries to a lesser degree. In all cases, main countries tend to trade more with partner countries with higher credit ratings. Trade between main countries and other developed countries is lower when the countries relax their international capital market controls. This result suggests substitution between trade and foreign investment in local companies that serve the local markets, consistent with horizontal FDI model. On the other hand, trade between main countries and less developed countries is higher when the countries relax their capital controls, consistent with vertical FDI model. The real exchange rate is positively related to bilateral trade when main countries trade with other developed countries. The positive relationship indicates that a larger difference in real exchange rates between two developed countries implies more trade. The rationale may relate to a shift in production location of multinational firms within developed countries to take advantage of misaligned currency. The opposite is the case for trade with less developed countries. This result, however, needs to be interpreted with caution and merits further study. In conclusion, this paper finds that there is indeed a significant relationship between financial development and trade. The results suggest implications for policies regarding access to external funds for businesses as well as capital controls and exchange rates in promoting trade and development, with some differences depending on whether the country is a developed or less developed country. 21 CHAPTER 2 Commodity Prices and the Exchange Rate 22 2.1 Introduction The global pricing of primary commodities is determined in the world markets connecting the supply of and demand for the goods. Consequently, time series data on these prices are expected to exhibit stationary or mean reverting process, reflecting the dynamic stability of market equilibria. However, empirical unit root tests of international commodity prices generally exhibit highly persistent or even nonstationary process for both individual and aggregate commodities. This phenomenon provides opportunity to explore what seems to generate the inconsistency between theory and empirical evidence. The present study investigates what factors determine commodity prices and provides a rationale for how these factors generate the inconsistency between theory and empirical evidence. Previous studies have observed this inconsistency between economic theory and empirical evidence suggested by the unit root test results on commodity prices. Wang and Tomek (2007) note that commodity prices should be level stationary according to price theory. Kellard and Wohar (2006) point out that commodity prices should be trend stationary to be consistent with the Prebisch-Singer hypothesis. They further report some evidence of nonlinear stationarity of commodity prices. The present paper uses a different approach in that it supports the evidence of nonstationarity of commodity prices and attempts to filter its factor. The proposed inference is: ?factoring out the nonstationary effect will generate the filtered commodity prices that are consistent with economic theory.? An important factor that is likely to explain a nonstationary effect of international commodity prices is the U.S. nominal exchange rate. The U.S. dollar is used in the global pricing of most internationally traded commodities. It is widely accepted that the U.S. nominal exchange rate, e.g., relative to the Euro, Japanese Yen, or some trade-weighted index of 23 currencies, is nonstationary. Its relationship with goods priced in U.S. dollars can be explained as follows. A depreciation of the U.S. dollar should drive an increase in the price of the goods priced in U.S. dollars to maintain the same world price. Another possible explanation is that a depreciation of the U.S. dollar results in lower commodity prices in terms of the foreign currency; and as the foreign demands for the commodities increase, their prices rise. Based on the above rationales, commodity prices should exhibit a dynamic behavior that closely mirrors the dynamics of the U.S. exchange rate and thus reflect its nonstationarity, which is common to all commodity prices. Furthermore, this inference holds for not only nominal commodity prices but also relative commodity prices, which are prices deflated by the U.S. Consumer Price Index (CPI). This similar dynamics of relative prices and nominal prices is expected since aggregate price indices such as the CPI exhibit less fluctuation than international commodity prices. This study begins with a factor analysis on a panel of 51 international commodity prices for the period January 1980 to December 2009 by performing the PANIC (Panel Analysis of Nonstationarity in Idiosyncratic and Common Components) method, which is a second- generation panel unit root test recently developed by Bai and Ng (2004).7 An out-of-sample forecasting analysis is then conducted to further investigate the link between the U.S. nominal exchange rate and commodity prices. The present paper examines the predictive power of international commodity prices for movements in the U.S. exchange rate. 7 PANIC method (Bai and Ng 2004) is chosen over other second-generation panel unit root tests such as Phillips and Sul (2004), Moon and Peron (2004), and Pesaran (2007) because these methods assume stationary common factors and thus do not apply to commodity prices based on the inference regarding the role of the U.S. nominal exchange rate. 24 The PANIC test results identify two common factors from both the nominal and relative commodity prices. The test results indicate that the first (most important) common factor is nonstationary, whereas the second common factor and the defactored (filtered) idiosyncratic components are both stationary. Graphically, the first common factor reflects the inverse of the U.S. nominal exchange rate. Out-of-sample forecasts using a simple model with the two common factors outperforms a benchmark random walk model. This forecasting result further supports the inference that the first common factor reflects the effect of the U.S. nominal exchange rate on commodity prices. The stationarity of the second common factor and the idiosyncratic components supports the existence of dynamic stability of market equilibria. Overall, these results provide a rationale for the previously identified inconsistency between economic theory and empirical evidence. This paper proceeds as follows. Section 2 describes the data and their sources. Section 3 provides literature review. Section 4 and section 5 discuss the econometric techniques and the empirical results. Section 6 concludes the paper. 2.2 Data The data set is constructed based on monthly data of U.S. nominal exchange rates and commodity prices for the period January 1980 to December 2009. The U.S. nominal exchange rate data are the trade-weighted exchange rate index for the U.S. dollar against a basket of major currencies including the Euro, Canada, Japan, the United Kingdom, Switzerland, Australia, and Sweden. The data for 51 commodity prices are from the IMF Primary Commodity Prices database available on the International Monetary Fund Website, except for the natural gas price from the U.S. Energy Information Administration Website. 25 [Table 2.1 here] 2.3 Literature review (a) Commodity prices and the exchange rate There have been attempts to explain the relationship between commodity prices and exchange rates, including those seeking to produce forecasts of commodity price movements. Existing economic models are rarely successful in predicting the exchange rate itself. Meese and Rogoff (1983) point out that nominal exchange rate models routinely fail to outperform the random walk model in predicting out-of-sample in the floating regime. The notion that commodity prices are perfectly arbitraged is also in question. One related literature is empirical work on the validity of the law of one price (LOP) in commodity markets. Since seminal work of Isard (1977), some find evidence against the LOP (e.g., Ardeni [1989], Engel and Rogers [2001], Parsley and Wei [2001], and Goldberg and Verboven [2005]), while others find evidence in favor of the LOP (e.g., Goodwin [1992], Michael et al.[1994], Obsfeld and Taylor [1997], Lo and Zivot [2001], and Sarno et al.[2004]). Several studies observe the inconsistency between economic theory and empirical evidence suggested by the unit root test results on commodity prices. Wang and Tomek (2007) note that price theory suggests that commodity prices should be level stationary. Kellard and Wohar (2006) point out that Prebisch-Singer hypothesis implies that commodity prices should be trend stationary. They recognize the weakness of conventional unit root tests, due to low power of the tests, and provide some evidence of nonlinear stationarity of commodity prices. 26 A growing body of literature investigates the information contents in commodity prices and other macroeconomic variables. Chen, Rogoff, and Rossi (2008) find substantial evidence of the predictive power of commodity-currency exchange rates for commodity prices, both in- sample and out-of-sample, but no evidence going in reverse direction. Using the exchange rate approach of the preceding researchers as well as a broader approach that includes exchange rates and other macrovariables, Groen and Presenti (2010) find that neither of both approaches has strong predictive power of exchange rates for commodity prices. They note, however, that both approaches occasionally outperform simple benchmark models. Gospodinov and Ng (2010) examine the link between commodity prices and inflation. They provide strong evidence that information contents in commodity prices have predictive power for inflation. (b) Panel unit root tests The application of panel unit root tests provides substantial advantages over the use of univariate unit root tests that in practice lack power by the short period of macroeconomic time series. Panel unit root tests increase the power of unit root tests by pooling information across units. Levin, Lin, and Chu (2002) provide evidence of the increase in power generated by panel unit root tests. Madala and Wu (1999) and Im, Pesaran, and Shin (2003) also propose the use of panel unit root tests instead of univariate unit root tests. The first generation unit root tests, however, assume independent units. The drawback of imposing this assumption is that the test results suffer from serious size distortions if the panel data are in fact cross section dependence. Pesaran (2004) proposes a test for cross section dependence based on averages of pair- wise correlation coefficients of the OLS residuals from the individual regressions in the panel. The test is based on an assumption of no a priori ordering of the cross section units. In a subsequent paper, Pesaran (2007) implements a cross-sectionally augmented Dickey-Fuller 27 (CADF) test. The CADF augments the standard ADF tests with the cross-section averages of lagged levels and first-differences of the individual series. Pesaran shows that panel unit root tests that do not account for cross section dependence are seriously biased if the degree of dependence is sufficiently large. The second generation unit root tests allow for cross section dependence among units and thus have better size property than the first generation tests. They suggest the presence of common effects, which are assumed to be stationary by Philips and Sul (2003), Moon and Perron (2004) and Pesaran (2007). The PANIC method of Bai and Ng (2004) use no a priori assumption on stationary or integrated processes of the common effects. It utilizes the factor structure of large dimensional panels to examine the nature of nonstationarity in the data. PANIC tests the unobserved components of the data instead of the observed series. Bai and Ng show that, if a factor structure exists in a panel data set, testing the presence of unit roots in the common factors and the idiosyncratic errors separately should be more effective than testing for individual unit roots in the observed series. 2.4 Econometric techniques This section illustrates the econometric techniques used in this paper. First, augmented Dickey-Fuller (ADF) tests for univariate unit roots are applied to each unit in the data set. The ADF test results provide an assessment of the time series properties and the number of integrated time series in the cross section. The cross section dependence test by Pesaran (2004) is applied to the residuals of the ADF regressions to determine whether there is cross section dependence among commodity prices. Since the present study finds evidence of cross section dependence, it justifies the use of the second-generation panel unit root test of Bai and Ng (2004). 28 Test for cross section dependence (CD) by Pesaran (2004) is based on the averages of pair-wise correlation coefficients of the OLS residuals from the individual regressions in the panel, g1829g1830 g3404 g4672 g2870g3021g3015g4666g3015g2879g2869g4667g4673g2869/g2870g3435? ? g2025g3548g3036,g3037g3015g3037g2880g3036g2878g2869g3015g2879g2869g3036g2880g2869 g3439 g3031g4657g4654 g1840g46660,1g4667 (1) where g2025g3548g3036,g3037 is the pair-wise correlation coefficients from the residuals of the ADF regressions. The above statistic has exactly mean zero for fixed values of T and N, under a wide class of panel data models. A rejection of the null of no cross-section dependence implies that the utilization of second-generation panel tests is appropriate and that the first-generation panel tests should not be used. Pesaran (2004) demonstrates that the test is valid under fairly general conditions even when T is small and N is large. The number of common factors is determined using g1842g1829g4666g1870g4667 and g1835g1829g4666g1870g4667 criteria of Bai and Ng (2002). The PANIC method of Bai and Ng (2004) is implemented to analyze the nonstationarity characteristics of commodity prices. A conventional method for out-of-sample forecast accuracy proposed by Diebold and Mariano (1995) and West (1996) is used to further examine the link between commodity prices and the U.S. nominal exchange rate. (a) The PANIC method The PANIC method of Bai and Ng (2004) is described as follows. Let g1868g3036,g3047 be the natural logarithm price of a good g1861 at time g1872 that obeys the following stochastic process8 g1868g3036,g3047 g3404g1855g3036 g3397 g2019g3036?g1806g3047 g3397 g1857g3036,g3047 (2) 8 All regularity conditions in Bai and Ng (2004, pp. 1130-1131) are assumed to be satisfied. 29 g46661g3398g1838g4667 g1806g3047 g3404g1775 g4666g1838g4667 g1821g3047 g46661g3398 g2025g3036g1838g4667g1857g3036,g3047 g3404g1828g3036g4666g1838g4667g2013g3036,g3047 where g1855g3036 is a fixed effect intercept, g1806g3047 g3404 g4670g1858g2869? g1858g3045g4671g1314 is a g1870 g34001 vector of (latent) common factors, ?g3047 g3404 g3427g2019g3036,g2869? g2019g3036,g3045g3431g1314 denotes a g1870 g34001 vector of factor loadings for good g1861, and g1857g3036,g3047 is the idiosyncratic error term. g1775g4666g1838g4667 and g1828g3036g4666g1838g4667 are lagg4666g1838g4667 polynomials. g1821g3047 , g2013g3036,g3047, and ?g3036 are mutually independent. An important point is that a factor model with N variables has N idiosyncratic components but a small number of common factors (r << N). A factor structure determines that the series have nonstationary process if one or more of the common factors are nonstationary, or the idiosyncratic error is nonstationary, or both (Bai and Ng [2004, p. 1128]). Estimations are performed by the method of principal components. When g1857g3036,g3047 is stationary, the principal component estimators for g1806g3047 and ?g3036 are consistent regardless of the order of g1806g3047. When g1857g3036,g3047 is integrated, however, the estimator is inconsistent because a regression of g1868g3036,g3047 on g1806g3047 is spurious. Implementing the method of principal components to the first-differenced data, PANIC tackles this problem as follows. Rewrite (1) as a model with differenced variables, ?g1868g3036,g3047 g3404g1855g3036 g3397 g2019g3036g4593?g1806g3047 g3397 g1986g1857g3036,g3047 (3) for g1872g34042,?,g1846. Let ?g1816g2919 g3404 g3427?g1868g2919,g2869? ?g1868g3021g3431g4593and ?g1816g3404 g4670?g1816g2869? ?g1816g3015g4671. After proper normalization9, the method of principal components for ?g1816?g1816g1314 yields estimated factors ?g1806g4632g3047, the 9 This is because the principal components method is not scale invariant. 30 associated factor loadings ?g3552g3036 , and the residuals ?g1857?g3036,g3047 g3404 ?g1868g3036,g2869 g3398 g2019g3036g4593?g1806g3047. Re-integrating these, we obtain g1857?g3036,g3047 g3404 ? ?g1857?g3036,g3046g3047g3046g2880g2870 (4) for g1861 g34041,?,g1840 and g1806g4632g3047 g3404 ? ?g1806g4632g3046g3047g3046g2880g2870 (5) Theorem 1 of Bai and Ng (2004, p.1134) shows that g1857?g3036,g3047 and g1806g4632g3047 can be tested as if they are observable. Specifically, the ADF test with no deterministic terms can be applied to each g1857?g3036,g3047 and the ADF test with an intercept can be used for g1806g4632g3047. When there are more than two nonstationary factors, cointegration-type tests can be used to determine the rank of g1775g46661g4667 in (2). Bai and Ng (2004) proposed a panel unit root test for idiosyncratic terms as follows. g1842g3032? g3404 g2879g2870 ? g2922g2924g3043g3280g3362g3550g2879 g2870 g3015 g3263g3284g3128g3117 g2870 g3015g3117/g3118 g3031g4657g4654 g1840g46660,1g4667 (6) where g1868g3032g3362g3549 is the p-value from the ADF test for g1857?g3036,g3047. Pooling these p-values across units tolerates the presence of as much heterogeneity across units as possible. The estimates for common factors g4666g1806g3553g3047g4667, factor loadings (?g3552g3036g4667, and idiosyncratic components (g1857?g3036,g3047) are obtained by applying the method of principal components. The importance of common factors for dynamics of the commodity prices relative to idiosyncratic components is evaluated by g1870g1874g3036g3038 g3404 g3097 g3435g3090g3553g3284g3033g4632g3295 g3286g3439 g3097 g3435g3032?g3284,g3295g3439 ,g1863 g34041,?,g1870 (7) where ?(?) denotes the standard deviation. 31 (b) Diebold-Mariano-West Test Statistics Out-of-sample forecasts of the U.S. nominal exchange rate are performed using a model based on the estimated common factors with the random walk model as a benchmark. A conventional method developed by Diebold and Mariano (1995) and West (1996) is used to evaluate the out-of-sample forecast accuracy of these models. Let g1871g3047 denote the natural logarithm U.S. nominal exchange rate. The random walk model of g1871g3047 implies, g1871g3047g2878g3038|g3047g3019 g3404 g1871g3047, (8) where g1871g3047g2878g3038|g3047g3019 is the k-step ahead forecast by the random walk model given information set at time t. The competing factor model is based on the following least squares regression. g2722g1871g3047g2878g3038 g3404 g1855g3397 g2010g4593g2722g1806g3047 g3397 g1873g3047 (9) Given the least squares coefficient estimate, the k-step ahead forecast by the factor model g1871g3047g2878g3038|g3047g3007 is g1871g3047g2878g3038|g3047g3007 g3404 ? g2722g1871g3047g2878g3046g3555g3038g3046g2880g2869 g3397g1871g3047, (10) where g2722g1871g3047g2878g3046g3555 is the fitted value from (7) and g1871g3047 is the actual data at time t. The forecast errors from the two models are ?g3047g2878g3038|g3047g3019 g3404 g1871g3047g2878g3038 g3398 g1871g3047g2878g3038|g3047g3019 , ?g3047g2878g3038|g3047g3007 g3404 g1871g3047g2878g3038 g3398 g1871g3047g2878g3038|g3047g3007 The loss differentials are defined as g1856g3047 g3404g1838g3435g2013g3047g2878g3038|g3047g3019 g3439g3398g1838g4666g2013g3047g2878g3038|g3047g3007 g4667 32 where g1838g4666g2013g3047g2878g3038|g3047g3037 ), g1862g3404g1844,g1832 is a loss function.10 To test the null of equal predictive accuracy, g1834g2868: g1831g1856g3047 g34040, the Diebold-Mariano-West statistic (DMW) is defined as g1830g1839g1849 g3404 g3031g3364g3493g3002g3049g3028g3045g3555 g4666g3031g3364g4667 (11) where g1856g1191 is the sample mean loss differential, g1856g1191 g3404 1g1846g3398 g1846 g2868 g3533 g1856g3047 g3021 g3047g2880g3021g3290g2878g2869 g1827g1874g1853g1870g3555g3435g1856g1191g3439 is the asymptotic variance of g1856g1191, g1827g1874g1853g1870g3555g3435g1856g1191g3439g3404 1g1846g3398 g1846 g2868 g3533 g1863g4666g1862,g1869g4667uni0413g3552g3037 g3044 g3037g2880g2879g3044 , and g1863(?) denotes a kernel function where g1863(?) = 0, g1862 g3408 g1869, and uni0413g3552g3037 is g1862g3047g3035 autocovariance function estimate.11 It is known that the DMW statistic is severely undersized with asymptotic critical values when competing models are nested, which is the case here. Therefore, the critical values of McCracken (2007) are used to avoid this size distortion problem. 2.5 Empirical results As a preliminary analysis, the ADF test for individual commodity prices are performed (see Table 2.2). The test rejects the null of nonstationarity for only 13 and 14 out of 51 10 We use the conventional squared error loss function, g4666g2013 g3047 g3037 g3397g1863|g1872g4667g2870, g1862 g3404 g1844,g1832 uni002E 11 Following Andrews and Monahan (1992), the quadratic spectral kernel with automatic bandwidth selection is used in the analysis. 33 commodity prices at the 5% significance level. Since the ADF test is known to suffer from low power in small samples, this test result cannot be taken as an evidence for overall nonstationarity of commodity prices. The cross-section dependence (CD) test of Pesaran (2004) rejects the null of no cross-section dependence at any significance level (see Table 2.2), which justifies the use of second-generation unit root tests and suggests that using the first-generation panel tests is not appropriate. [Table 2.2 here] The next step is to implement PANIC for the commodity prices. Using g1842g1829g4666g1870g4667 and g1835g1829g4666g1870g4667 criteria of Bai and Ng (2002), the number of common factors is determined. Figure 2.1 indicates that all criteria except g1842g1829g2871g4666g1870g4667 choose two factors (g1870g34042). [Figures 2.1and 2.2 here] Figures 2.3 and 2.4 show that dynamics of individual commodity prices are substantially influenced by the first common factor. For many commodity prices, g1870g1874g3036g3038 is greater than one, suggesting that the first common factor is more important than idiosyncratic components for those prices. The second common factor also plays an important role for some commodities such as crude oil prices. Similar evidence can be found in factor loading estimates (Figures 2.5 and 2.6). [Figures 2.3 through 2.6 here] The PANIC unit root test results are reported in Table 2.3. The ADF test fails to reject null of nonstationarity for the first factor (g1858g3047g2869g4667, but rejects the null for the second factor g4666g1858g3047g2870g4667 at 5% level of significance. The cointegration tests are not implemented in this case since there is only one nonstationary factor among two common factors, g1870g1853g1866g1863 g3435g1775g46661g4667g3439g3404 1. For the 34 defactored (filtered) idiosyncratic components, the ADF tests reject the null for 30 and 29 out of 51 nominal and relative commodity prices, respectively. The panel unit root test by (6) rejects the null hypothesis at any significance level. These results provide strong evidence that there is a nonstationary common factor that drives persistent movement of commodity prices. [Table 2.3 here] Although there is no obvious way of identifying the source of this nonstationarity (since the factors are latent variables), it can be observed that the estimated first common factor mirrors the image of the U.S. nominal exchange rate. Figure 2.3 shows that the exchange rate exhibits two large swings peaked in 1985 and in 2002, whereas the first common factor estimate shows similar patterns but in opposite directions. This phenomenon may make sense since it is previously acknowledged that most commodities are priced in dollar terms. A depreciation of U.S. dollar relative to other major currencies may raise commodity prices given the world price, and the reverse holds. The second common factor shows stable fluctuations which may indicate the stationary characteristics. The two common factors considerably affect oil prices (Figures 2.3 through 2.6). The three crude oil prices (Brent, Dubai, and WTI) in Figure 2.8 show some interesting dynamics illustrating the role of the common factors in these prices. Original oil prices are plotted in panel (a), and the defactored oil prices are in panel (b). Panel (a) clearly shows extremely persistent movements of oil prices. Defactored oil prices, however, exhibit much less persistent dynamics. The nonstationarity of nominal exchange rates seems to be widely accepted in the economic profession. If this is the case, and if commodity prices are largely governed by a single nonstationary common factor, it might be the case that such nonstationarity is due to the U.S. nominal exchange rate. The remaining factors and/or idiosyncratic components may reflect 35 changes in world demand and supply conditions, and their movements around the long-run equilibrium may be consistent with price theories. Lastly, the out-of-sample forecast results reported in Tables 6 and 7 provide additional evidence to support the possibility that the first common factor reflects the effect of exchange rate on commodity prices. Forecasting is conducted recursively by sequentially adding one additional observation from 180 initial observations toward 360 total observations for g1863 = 1, 2, 3, 4. First, the ratios of the root mean square prediction error (RMSPE) of the random walk model to the factor model were greater than one for all g1863, suggesting that the factor model outperforms the benchmark random walk model. Second, the DMW statistics with McCracken?s (2007) critical values reject the null of equal predictability for g1863 = 1, 4 at the 5% significance level and for g1863 = 3 at the 10% level. 2.6 Conclusion This paper examines common factors affecting the prices of 51 highly tradable world commodities. It recognizes an inconsistency between the implications of economic theory concerning the dynamic behavior of commodity prices and the implications of corresponding empirical tests as reported by previous studies (e.g., Wang and Tomek [2007]; Kellard and Wohar [2006]). Dynamic stability of market equilibria implies that time series data on commodity prices should exhibit a stationary or mean reverting process, but unit root tests on these prices generally show evidence of nonstationarity. This paper investigates this inconsistency between economic theory and empirical evidence by first identifying common factors driving the dynamics of highly tradable commodity prices. 36 Utilizing the PANIC method of Bai and Ng (2004), the study finds that the factor structure of commodity prices is well described by two common factors. The first common factor explains the largest proportion of the variation in the panel of commodity prices. It is nonstationary and closely related to the U.S. nominal exchange rate, as suggested by theoretical, graphical, and out-of-sample forecasting evidence. The second common factor and the defactored (filtered) idiosyncratic components are stationary. A simple model constructed with the two common factors significantly outperforms a random walk in forecasting the exchange rate. This is an important result because the random walk model consistently outperforms economic models for forecasting the exchange rate since it was reported by Meese and Rogoff (1983). Furthermore, the forecasting result indicates that the first common factor and exchange rates share information content. It suggests that factors that have a predictive power for the exchange rate will have a correspondingly predictive power for commodity prices. The stationarity of the second common factor and the idiosyncratic components of each series are consistent with equilibrium price dynamics, showing a mean reverting process. When the effects the first common factor, which has been identified to be closely related to the U.S. exchange rate, are filtered out of the panel of commodity prices, the remaining factors affecting commodity prices exhibit the type of dynamic behavior consistent with price theory. An obvious illustration is demonstrated by the three crude oil prices (Brent, Dubai, and WTI), which show extremely persistent (possibly nonstationary) movements of their original prices, but much less persistent dynamics (possibly stationary) movements of their defactored (filtered) prices. Overall, the results of this study support the proposed rationale for the inconsistency between economic theory and empirical evidence on international commodity prices. 37 CHAPTER 3 Japanese and Korean Automobile Exports to the United States: Shipping the Good Cars Out? 38 3.1 Introduction This paper analyzes Japanese and Korean automobile exports to the United States based on monthly data for the period 1995 to 2008 to examine consistency with the Alchian-Allen theorem (1964). Also called ?shipping the good apples out,? the theorem suggests that a per-unit charge applied to two similar goods will lower the relative price and increase the relative consumption of the higher quality good (e.g., Hummels and Skiba [2002]; Bauman [2004]). Japan has historically been the largest automobile exporter to the United States with a share of about 50% to 70% of U.S. automobile imports during the period of observations, whereas Korea has become one of the major automobile exporters to the United States with a share of about 14% of U.S. imports in 2008.12 One of the implications of the Alchian-Allen theorem may be related to the choices of production location as major Japanese and (more recently) Korean automobile manufacturers own production plants in the United States. The present study utilizes the dynamic OLS (DOLS) method of Stock and Watson (1993) to estimate car imports from Japan and Korea with car price, shipping costs, and the U.S. dollar exchange rate as explanatory variables. Trade costs include tariffs and shipping costs; and shipping costs include freight, insurance, and other charges. Since tariff is invariant in this case, the trade costs include only shipping costs. Moreover, the model does not include distance variable, which is commonly used in measuring shipping costs, because Japan and Korea are located similar distance from the United States. The main hypothesis tested in this paper is that as shipping costs increase, U.S. imports of higher-quality cars from Japan and Korea increase. 12 Import shares are calculated with data from the United States International Trade Commission (USITC, 2009). 39 The model proposed in this paper utilizes freight charge, the main part of shipping costs, as an increasing function of the price of oil based on an observation that the price of oil positively affects shipping rates with a one-month lag (Kilian, 2009). Shipping costs include other charges such as insurance and handling. The inference is that freight charge measures per- unit shipping cost, whereas insurance and other charges may vary with car prices. Thus, higher insurance and other charges reduce or eliminate the Alchian-Allen effect on the demand for higher quality cars. The nominal exchange rate is utilized in the present model. Given currency supply and demand conditions, an appreciation of the dollar is expected to lower the price of foreign goods and increase the level of exports. An indicator variable for the Asian financial crisis of 1997- 1998 is used to control for the large depreciation of the Korean won against the dollar. This paper argues that the Alchian-Allen theorem holds when shipping costs account for per-unit freight charge only without considering insurance and other charges that may be imposed based on the car price. Another condition for the Alchian-Allen theorem to hold is that the effect of per-unit charge is larger than the effect of the ad-valorem charges. Present estimates based on Baltic Dry shipping index (and the price of crude oil) as a measure of freight rate support the Alchian-Allen theorem. Estimations based on actual shipping cost data are also conducted, and the result indicates that the effect of insurance and other charges outweighs the effect of per-unit freight charge.13 It suggests that actual shipping cost data do not support the Alchian-Allen theorem. The relative price of higher quality cars is not necessarily lower when shipping costs rise, perhaps due to higher insurance and other charges. Furthermore, this paper 13 Actual shipping cost is calculated by deducting the Custom value from the CIF (cost, insurance, and freight) value, and the result is divided by the total quantity of car imports. The Custom value is defined by the United States International Trade Commission (USITC, 2009) as ?the price actually paid or payable for merchandise, excluding U.S. import duties, freight, insurance, and other charges.? The CIF value excludes U.S. import duties. 40 suggests that foreign car production is to be done domestically when trade costs are high, regardless differences in car quality. The remainder of the paper proceeds as follows. Section 2 provides literature review. Section 3 discusses the data. Section 4 describes the shipping model, and Section 5 discusses the empirical method and results. Section 6 concludes the paper. 3.2 Literature review Alchian and Allen (1964) propose that a per-unit charge applied to two similar goods will lower the relative price and increase the relative consumption of the higher quality good (e.g., Hummels and Skiba [2002] and Bauman [2004]). This theorem, often called ?shipping the good apples out,? has been proven algebraically by Borcherding and Silberberg (1978) and Saito (2007). The theorem originally applies to a two-good world only, but Borcherding and Silberberg (1978) subsequently provide a proof of its application with two goods in a many- good world given the two goods are close substitutes. In a more recent paper, Bauman (2004) relaxes the assumption of close substitutability between two goods in a many-good world. A recent application of ?shipping the good apples out? in international trade is reported by Hummels and Skiba (2002, 2004), who provide theoretical and empirical evidence based on bilateral trade data for six countries and their trading partners. With data on traded good prices, quantities, and shipping costs, they point out that shipping costs behave like a quantitative restriction similar to quotas (2002, p. 4). Another important contribution of Hummels and Skiba is their analysis of variation in ad-valorem trade costs such as tariffs, suggesting that they lower the relative demand for high quality goods and reduce the Alchian-Allen effect. 41 Issues on automobile trade between Japan and the United States have been on the political agenda of the two countries for a few decades. In 1981, Japan voluntarily limited automobile exports to the United States with a three-year export quota of 1.68 million automobiles (Tharp, 1981). The effectiveness of the voluntary export restraint (VER) system in initiating recovery of the U.S. automobile industry was in question as Japan shifted its concentration to exporting larger and higher-priced cars, increasing its export revenue while complying with the quantity restriction. The result was increased competition in the U.S. higher- priced automobile market dominated by U.S. automakers. A trade dispute occurred from 1993 to 1995. An agreement ended the two-year dispute after the Japanese agreed to deregulate its domestic automobile market. Another important deal reached in 1995 was that Japanese automobile manufacturers in the United States would increase production by 25% to 2.65 million automobiles in 1998 (Katzner and Nikomarov, 2008). Kilian (2009) notes that freight rates may increase during upward oil price shocks due to higher demand for commodities and the use of bunker fuel oil as an input in shipping services. Kilian finds that the changes in freight rates do not occur contemporaneously with the changes in the price of crude oil and suggests a one-month lag on the response of freight rates. Backus and Crucini (2000) report that there is a relationship between the prices of oil and the terms of trade similar to the relationship between nominal or real exchange rates and the terms of trade. Their finding suggests that the price of oil and exchange rate are important determinants of trade. Banik and Biswas (2007) point out that price competition among firms determines the degree of exchange rate pass-through in the U.S. automobile market with the degree of price competition negatively correlated with the degree of exchange rate pass-through. They find low degree of exchange rate pass-through for the Japanese and Korean automobile exporters, 42 indicating that exporters from these two countries try to protect their market share in the U.S. market by offsetting the effect of exchange rate changes on car prices. Korean won experienced a large depreciation against the dollar during the Asian financial crisis of 1997-1998. Korea was one of the Asian countries hardest hit by the crisis, along with Indonesia and Thailand. A rapid increase of short-term foreign borrowings before the crisis, combined with incomplete financial sector reforms, are identified as the main problems leading to corporate bankruptcies and financial sector collapses in these countries when the exchange rate movements turned against them (Radelet and Sachs, 1998). The Asian financial crisis did not significantly affect the Japanese economy, and the yen mildly depreciated against the dollar. 3.3 Data A time-series model is applied to monthly data of U.S. car imports from Japan and Korea to test the main hypothesis on consistency with the Alchian-Allen theorem. Four data sets are constructed for the period 1995 to 2005 with 132 observations in each. The four data sets represent higher and lower quality cars from the two exporting countries. Two types of automobiles are examined: passenger motor vehicles with engine sizes exceeding 1,000 cc but not exceeding 1,500 cc (small size cars), and passenger motor vehicles with engine sizes exceeding 1,500 cc but not exceeding 3,000 cc (medium size cars). In addition, two data sets of higher quality cars from 1995 and 2008 with 180 observations are examined to include more recent data. 14 14 The lower quality car imports (small size cars) from Korea discontinued after 2005. 43 The data on car imports (in quantities and values) are obtained from the Interactive Tariff and Trade Database of the United States International Trade Commission (USITC, 2009). The dollar exchange rates are based on noon buying rates for JPY/USD and KRW/USD reported by the Federal Reserve Bank of New York (2009). The shipping index used is the Baltic Dry Index, which is taken from the FactSet Research Systems database. The data on crude oil price is the WTI spot price FOB taken from the Energy Information Administration of the U.S. Department of Energy (2009). All prices are deflated using Consumer Price Index reported by the Bureau of Labor Statistics (2009). U.S. imports of small size cars from Japan exhibits an increasing trend during the last five years (Figure 3.1). In the same period, small size cars from Korea show a declining trend. In medium size category, car imports from Japan decreased during the period 1995-2003 and then increased during the last five years of observations, whereas car imports from Korea increased steadily over the period 1995-2008. The unit price of different car categories are presented in Figure 3.2, showing a persistent increase in prices for both small cars from Japan and Korea during the period 2000 to 2005. Further, Figure 3.3 shows actual shipping costs (CIF charges) for both small and medium size cars. Other data series are presented in Figure 3.4. The price of crude oil exhibits an increasing trend prior to 2006. After a short decline in 2006, it shows a rapid increase in 2007- 2008, with a peak in the second quarter of 2008. The Baltic Dry Index was relatively stable until 2003, and then considerably higher and more volatile through 2008. Issued daily by the London- based Baltic Exchange, the index measures worldwide international shipping rates of many dry bulk cargoes. The plots of crude oil price and Baltic Dry Index indicate a delay on the response of shipping rates to changes in oil price. The exchange rate data indicate that the Korean won 44 experienced a large depreciation against the dollar during the Asian financial crisis of 1997- 1998. The Japanese yen value was relatively stable over the period. [Figures 3.1 through 3.4 here] 3.4 The shipping model The model estimates the quantity of U.S. car imports from Japan and Korea with three main explanatory variables: car price, shipping costs, and the exchange rate. All variables in the model are in natural log form. g1855g3051g3036 g3404 g1832g4666g1868g3051g3036,g1878g3051g3036,g1857g3036g4667 (1) where subscript g1876 denotes small car g4666g1871g4667 or medium car g4666g1865g4667, and subscript g1861 denotes exporting country, Japan g4666g1862g4667 or Korea g4666g1863g4667. The variables are defined as follows: g1855 = quantity of car imports g1868 = car price per unit based on the U.S. Custom value (excluding shipping costs) g1878 = shipping costs, measured in two ways: (a) using freight rate g4666g1858g4667, constructed with Baltic Dry Index as an increasing function of oil price (o), where g1858 g3404 g1832g4666g1867g3047g2879g2869g4667. (b) using actual shipping costs(g1872g3051g3036g4667, calculated from the U.S. Custom value and CIF value of imported car. g1857 = dollar exchange rate (USD/JPY or USD/KRW) with an indicator variable controlling for the Asian financial crisis in the case of KRW. [Table 3.1 here] 45 The relative quality of each import category is determined by comparing its average price relative to those of other import categories as shown in Table 3.1. Average car prices are calculated based on the U.S. Custom value, which is the price actually paid or payable for merchandise, excluding U.S. import duties, freight, insurance, and other charges (USITC, 2010). The average prices of the four import categories can be compared as follows. The cars g1855g3046g3037 have a higher price than g1855g3046g3038, but the cars g1855g3040g3037 has a lower price than g1855g3040g3038. The present study assumes similar product variations in g1855g3046g3037 and g1855g3046g3038, and applies the same assumption to g1855g3040g3037 and g1855g3040g3038. Based on the comparison of car prices in Table 3.1, the relative quality of each import category can be determined: higher quality goods are small size cars from Japan and medium size cars from Korea, and lower quality goods are medium size cars from Japan and small size cars from Korea. To be clear, higher quality cars are marked as g1855g3046g3037g3035 and g1855g3040g3038g3035 , and lower quality cars are marked as g1855g3046g3038g3039 and g1855g3040g3037g3039 . A detailed illustration of the Alchian- Allen theorem is to be explained in the following paragraphs. A world with n goods comprises g1855g3046g3037g3035 , g1855g3046g3038g3039 , g1855g3040g3037g3039 , g1855g3040g3038g3035 , and g1877, where g1877 is a composite of ?other? goods. The first and second goods are small cars with different qualities, higher and lower. The third and fourth goods are medium cars with higher quality and lower quality. Higher quality cars are represented by g1855g3035 and lower quality cars are represented by g1855g3039; then by assumption, the car prices follow g1868g3035 g3408g1868g3039 g3408 0. A per-unit shipping cost (freight charge), g1858, is added to the prices of g1855g3035 and g1855g3039. The prices become g1868g3035 g3397g1858, g1868g3039 g3397g1858, and g1868g3052, respectively. Following Borcherding and Silberberg (1978), g2034g4666g1855g3035/g1855g3039g4667/g2034g1858g34080 (2) 46 where g1855g3035g4666g1868g3035,g1868g3039,g1868g3052,g1847g4667 and g1855g3039g4666g1868g3035,g1868g3039,g1868g3052,g1847g4667 are Hicksian demand functions. That is, as the shipping cost rises, the consumption of higher quality cars increases relative to the lower quality cars, holding real income and the prices of all other goods constant. Given g2013g3036g3037 g3404g4666g1868g3037/g1855g3036g4667 g4666g2034g1855g3036/g2034g1868g3037g4667 be the compensated elasticities, applying the chain rule and quotient rule on (2) will result in g2034g4666g1855g3035/g1855g3039g4667/g2034g1858 g3404 g3030g3283g3030 g3287 g3427g4666g2013g3035g3035/g1868g3035g4667 g3397 g4666g2013g3035g3039/g1868g3039g4667 g514 g4666g2013g3039g3035/g1868g3035g4667 g514 g4666g2013g3039g3039/g1868g3039g4667g3431g3408 0 (3) which confirms the Alchian-Allen theorem. In a two-good world, Borcherding and Silberberg (1978) note that Hick?s third law (1946) applies (? g2013g3036g3037g3036 g3404 0g4667uni002E This implies that the two cars in a two-car world must be substitutes g4666g2013g3035g3039 g34080g4667 and own-price elasticities are negative g4666g2013g3039g3039 g34070g4667, which can be obtained by substituting for g2013g3035g3035 and g2013g3039g3035 (given g2013g3035g3039 g3404g3398g2013g3039g3035g4667. In an n-good world (introducing a third, composite of ?other? goods), Borcherding and Silberberg show that Alchian-Allen holds if the two goods are assumed to be close substitutes. Bauman (2004) relaxes the assumption of close substitutability between the two goods in an n-good world. So long as the two goods are not close complements g4666g2013g3035g3039 g3408g2013g3039g3039g4667 and are not close in price g4666g1868g3035 g3408g3408g1868g3039g4667 a per-unit charge will increase the relative consumption of the higher quality good (2004, p.535). The implication is that the two goods bearing per-unit charge do not have to be close substitutes. In the car import case, higher quality cars from Japan g4666g1855g3046g3037g3035 ) can be substitute to lower quality cars from Japan g4666g1855g3040g3037g3039 g4667 or lower quality cars from Korea g4666g1855g3046g3038g3039 g4667uni002E When per-unit shipping costs increase, U.S. demand for higher quality cars from Japan increases relative to lower quality cars from both Japan and Korea. 47 The consumer utility is based on Hummels and Skiba (2002) and applied specifically to the utility obtained by the U.S. consumer from buying imported cars. The U.S. consumer obtains greater utility from higher quality cars, g1847g3051 g3404g4674? g2019g3051g3036g3039 g3435g1855g3051g3036g3039 g3439g3087g3051,g3036 g3397 g2019g3051g3036g3035 g3435g1855g3051g3036g3035g3439g3087g4675 g2869/g3087 , (4) denoting g1855 as car quantity and ? as car quality, where g2019g3051g3036g3035 g3408g2019g3051g3036g3039 and g2016 g34041g3398g2869g3097, and ? is the elasticity of substitution between higher and lower quality cars based on the CES utility function. Subscript g1876 denotes small car g4666g1871g4667 or medium car g4666g1865g4667, and subscript g1861 denotes exporting country, Japan g4666g1862g4667 or Korea g4666g1863g4667. The price of cars depends on the U.S. customer price g4666g1868g3051g3036g4667 and trade costs. Trade costs include tariffs and shipping costs; and shipping costs include freight, insurance, and other charges. Since tariff is invariant in this case, the trade costs include only shipping costs g4666g1878g4667uni002E Let g1858 denotes freight rate and g1859 denotes ad-valorem insurance and other shipping charges, then the price of car at destination follows g1868g3051g3036g3022g3020 g3404g1858g3397g1859g3051g3036 g1868g3051g3036 (5) where g1859 g3408 1. Freight rate g1858 represents a per-unit cost that follows the Alchian-Allen theorem, assumed the same for Japan and Korea given similar distance to the United States; and g1859 is ad-valorem charges that increase with car price, g1859g3404 g1832g4666g1868g4667uni002E As described in g46661g4667, g1858 increases with oil price g4666g1867g3047g2879g2869g4667. The model in g46661g4667 does not count for ad-valorem charges g4666g1859g4667 when freight rate g1858 is used, based on Baltic Dry Index and the price of oil. The ad-valorem charges g1859 are taken into account when the model utilizes the actual CIF (cost, insurance, and freight) shipping costs instead of 48 freight rate. The actual data are calculated based on the U.S. Custom value and CIF value of car imports as well as import quantity reported by the USITC (2010). The actual data do not allow for decomposing the CIF charges into cost, insurance, and freight separately. Hummels and Skiba (2002) note that ad-valorem charges reduce the Alchian-Allen effect on shipping cost. Consequently, Alchian-Allen theorem is expected to hold when shipping costs do not count for ad-valorem charges, or when per-unit charge is larger than the ad-valorem charges. The present model assumes that the effect of insurance and other charges is similar to that of ad-valorem tariffs in reducing the Alchian-Allen effect. An expenditure minimization problem for U.S. consumer buying imported cars can be constructed with the car price in (5) and consumer utility in (4). Assuming competitive firms, the consumer?s first order conditions implies g3030g3299g3284 g3283 g3030g3299g3284g3287 g3404 g3436 g3043g3299g3284g3287 g3034g3299g3284g2878 g3033 g3043g3299g3284g3283 g3034g3299g3284g2878 g3033g3440 g3097 g3436g3090g3299g3284 g3283 g3090g3299g3284g3287 g3440 g3097 . (6) Rearranging (6) results in g3030g3299g3284 g3283 g3030g3299g3284g3287 g3404 g3436 g3043g3299g3284g3287 g2878 g3033/g3034g3299g3284 g3043g3299g3284g3283 g2878 g3033/g3034g3299g3284g3440 g3097 g3436g3090g3299g3284 g3283 g3090g3299g3284g3287 g3440 g3097 . (7) Further derivations of (7) with respect to g1858 and g1859 will give results that suggest the relative consumption of higher quality cars increases with shipping costs when the effect of per- unit freight charge is larger than the effect of ad-valorem insurance and other charges. 49 3.5 Empirical method and results The augmented Dickey-Fuller (ADF) test for the stationarity of individual variables is summarized in Table 3.2. The actual shipping cost variables are all stationary in their log forms. Other variables are mostly difference-stationary in their log forms. The spurious regressions are presented in Table 3.3 and Table 3.4. Two regressions are run for each data set: the first includes freight rate g4666g1858g4667, whereas the second includes actual shipping costs g4666g1872g4667 instead. Estimations using freight rate are cointegrated by the Engle-Granger (EG) for all four data sets, g1855g3046g3037g3035 , g1855g3046g3038g3039 , g1855g3040g3037g3039 , and g1855g3040g3038g3035 (Table 3.3[a]). Estimations using actual shipping costs are cointegrated for three data sets (g1855g3046g3037g3035 , g1855g3046g3038g3039 , and g1855g3040g3038g3035 ), but not cointegrated for g1855g3040g3037g3039 (Table 3.3[b]). The ARCH test results indicate that heteroskedasticity exists in the data, and the Breusch-Godfrey Lagrange Multiplier test results indicate the presence of autocorrelations. [Tables 3.2 through 3.4 here] The model is estimated using the dynamic OLS (DOLS) regression proposed by Stock and Watson (1993). It estimates long-run equilibria in cointegrated systems that may have variables integrated with different orders. The DOLS estimates parameter g2010 of the following regression: g1877g2869g3047 g3404 g2010g4593g1877g2870g3047 g3397 g1856g4666g1838g4667g1986g1877g2870g3047 g3397 g1874g3047 (8) where g1856g4666g1838g4667 is the leads and lags of g1986g1877g2870g3047, the first differences of any g1835(1) variables; and g1874g3047 is a stochastic error term. This process eliminates asymptotically possible bias caused by endogeneity or serial correlation. DOLS is applied to the shipping model as follows, g1855g3051g3036g3047 g3404 g2010g2869g4593g1868g3051g3036g3047 g3397 g1856g4666g1838g4667g1986g1868g3051g3036g3047g3397 g2010g2870g4593g1878g3051g3036g3047 g3397 g1856g4666g1838g4667g1986g1878g3051g3036g3047g3397 g2010g2871g4593g1857g3036g3047 g3397g1856g4666g1838g4667g1986g1857g3036g3047 g3397 g1874g3047 (9) 50 substituting g1878g3051g3036g3047 = g1858g3047 (9a) or g1878g3051g3036g3047 = g1872g3051g3036g3047 (9b) where freight rate g4666g1858g3047g4667 and actual shipping costs g4666g1872g3051g3036g3047g4667 are the variables of interest. The estimation results are discussed separately based on (9a) and (9b). [Tables 3.5 and 3.6 here] (a) Regressions with freight rate Estimations based on monthly data from 1995 to 2005, using freight rate (Baltic Dry Index and the price of oil) as a measure of shipping cost, show expected results on the response of import quantity to the changes in freight rate (Table 3.5[a]). An increase in freight rate is related to an increase in U.S. imports of higher quality cars from Japan and Korea (g1855g3046g3037g3035 and g1855g3040g3038g3035 ) and a decrease in U.S. imports of lower quality cars from Japan (g1855g3040g3037g3039 g4667, holding car price and exchange rate constant. These results are significant at the 1% level of significance. The coefficient estimate for lower quality cars from Korea (g1855g3046g3038g3039 g4667 is not significant. Exchange rate variable shows the expected significant negative coefficients for three data sets g4666g1855g3046g3037g3035 , g1855g3046g3038g3039 ,and g1855g3040g3038g3035 ) and negative but not significant coefficient for g1855g3040g3037g3039 . A depreciation of the U.S. dollar increases the foreign car price and thus lowers the demand for those cars, reflected in the lower quantity of car imports. Extending the period of observations to 2008 (Table 3.6[a]), the regression results for higher quality cars show similar results on the coefficient estimates of the freight rate variable, but weaker in magnitudes. Still, it seems that the model works even in the period of oil price shock in 2006 to 2008. 51 (b) Regressions with actual shipping costs Estimations based on monthly data from 1995 to 2005 using actual CIF shipping charges as a measure of shipping costs are presented in Table 3.5[b]. The significant negative coefficient of shipping cost variable for the U.S. import of higher quality cars from Korea (g1855g3040g3038g3035 g4667 implies that an increase in shipping costs is related to a decrease in import quantity of higher quality cars, holding car price and exchange rate constant. This result seems to counter the Alchian-Allen theorem. Similarly, the significant positive coefficient of shipping cost variable for the lower quality cars from Korea (g1855g3046g3038g3039 ) shows inconsistency with Alchian-Allen theorem. The exchange rate variable shows significant negative coefficients consistent with the results from (a) for two data sets (g1855g3046g3038g3039 and g1855g3040g3037g3039 ). The exchange rate variable for higher quality cars from Japan has a significant positive coefficient, suggesting that a lower U.S. dollar value, relative to the Japanese yen, increases the import quantity of higher quality cars from Japan. This result counters the result from (a). Extending the period of observations to 2008 (Table 3.6[b]), the regression results for higher quality cars from Japan and Korea (g1855g3046g3037g3035 and g1855g3040g3038g3035 ) show coefficient estimates similar to those from the data sets of shorter period of observations (1995 to 2005). 3.6 Conclusion This paper examines data on Japanese and Korean automobile exports to the United States to determine consistency with the Alchian-Allen theorem using the dynamic OLS estimators. The result suggests that the Alchian-Allen theorem holds when shipping costs account for per-unit freight charge only without considering insurance and other shipping 52 charges that may be imposed based on the car price. It also suggests that another condition for Alchian-Allen theorem to hold is when the effect of per-unit charge is larger than the effect of the ad-valorem charges. Hummels and Skiba (2002) report that ad-valorem trade costs such as tarrifs reduce the Alchian-Allen effect. Similarly, the present study finds that insurance and other shipping charges seem to reduce or eliminate the Alchian-Allen effect. Estimations based on Baltic Dry shipping index and the price of oil as a measure of freight rate support the Alchian-Allen theorem. As shipping rates increase, U.S. imports of higher quality cars from Japan and Korea increase relative to the lower quality cars. In contrast, estimations based on the actual CIF (cost, insurance, and freight) charges indicate that the relative price of higher-quality car imports is not necessarily lower when shipping costs rise. A possible rationale is that insurance and other shipping charges are imposed based on the car price. These charges reduce or eliminate the Alchian-Allen effect of the per-unit freight charge. Therefore, the present shipping cost data do not support the Alchian-Allen theorem. The model seems to work better for car imports from Korea than for car imports from Japan, as indicated by the cointegration test and DOLS regression results. A possible explanation is there are more product variations in the cars imported from Japan compared to those imported from Korea. Another possible explanation is there are more automobile trade agreements between Japan and the United States that may influence the types of cars imported from Japan. The result of this study suggests that foreign car production is to be done domestically in the United States when trade costs are high, regardless differences in car quality. Policy that promotes domestic production by foreign car manufacturers is recommended. 53 References Alchian, Arment A. and William R. Allen. 1964. University Economics. Belmont, Wadsworth Publishing Co. Anderson, James E. 1979. A theoretical foundation for the gravity equation. American Economic Review, 69(1), pp. 106-116. Anderson, James and Eric van Wincoop. 2003. Gravity with gravitas: a solution to the border puzzle. American Economic Review,93(1), pp. 170-92. Andrews, Donald W. K. and J. Christopher Monahan. 1992. An improved heteroskedasticity and autocorrelation consistent covariance matrix estimator. Econometrica, 60(4), pp. 953-66. Ardeni, Pier Giorgio. 1989. Does the law of one price really hold for commodity prices? American Journal of Agricultural Economics Association,71(3), pp. 661-669. Backus, David K. and Mario J. Crucini. 2000. Oil prices and the terms of trade. Journal of International Economics, 50, pp.185-213. Bai, Jushan and Serena Ng. 2002. Determining the number of factors in approximate factor models. Econometrica,70 (1), pp. 191-221. Bai, Jushan and Serena Ng. 2004. A PANIC attack on unit roots and cointegration. Econometrica,72 (4), pp. 1127-1177. Baier, Scott L. and Jeffrey H. Bergstrand. 2007. Do free trade agreements actually increase members? international trade? Journal of International Economics, 71, pp. 72-95. Baltagi, Badi H., Peter Egger and Michael Pfaffermayr. 2003. A generalized design for bilateral trade flow models. Economics Letters, 80 (3), pp. 291-97. Banik, Nilanjan and Basudeb Biswas. 2007. Exchange rate pass-through in the U.S. automobile market: a cointegration approach. International Review of Economics and Finance, 16, pp.223-236. Bauman, Yoram. 2004. Shipping the good apples out: a new perspective. Economic Inquiry,42, pp.534-536. Baxter, Marianne and Michael A. Kouparitsas. 2006. What determines bilateral trade flows? NBER Working Paper, 12188. Beck, Thorsten. 2002. Financial development and international trade: is there a link? Journal of International Economics, 57, pp. 107-131. 54 Beck, Thorsten. 2003. Financial dependence and international trade. Review of International Economics,11, pp. 296-316. Borcherding, Thomas E. and Eugene Silberberg. 1978. Shipping the good apples out: the Alchian and Allen theorem reconsidered. The Journal of Political Economy, 95(1), pp. 131-138. Bureau of Economic Analysis. 2009. The U.S. National Economic Accounts. http://www.bea.gov/National/index.htm. Bureau of Labor Statistics. 2009. Consumer Price Index. http://www.bls.gov/CPI/. Calvallo, Eduardo A. and Jeffrey A. Frankel. 2008. Does openness to trade make countries more vulnerable to sudden stops, or less? Using gravity to establish causality. Journal of International Money and Finance, 27, pp. 1430-1452. Centre d'Etudes Prospectives et d'Informations Internationales (CEPII). 2010. Geodesic distances database. http://www.cepii.fr/anglaisgraph/bdd/distances.htm Chen, Yu-chin, Kenneth Rogoff, and Barbara Rossi. 2008. Can exchange rates forecast commodity prices? NBER Working Paper,13901. Clark, Peter, Natalia Tamirisa, and Shang-Jin Wei. 2004. Exchange rate volatility and trade flows ? some new evidence, IMF Occasional Paper,235. Washington, DC: IMF. Demirg??-Kunt, Asli and Ross Levine. 2001. Financial structure and economic growth: perspectives and lessons. In Financial Structure and Economic Growth, ed. Asli Demirg??-Kunt and Ross Levine. Cambridge, MA: The MIT Press. Diebold, Francis X. and Roberto S. Mariano. 1995. Comparing predictive accuracy. Journal of Business and Economic Statistics, 13(3), pp. 253-63. Enders, Walter. (2004). Applied Econometric Time Series. John Wiley & Sons, New York. Engel, Charles and John H. Rogers. 2001. Violating the law of one price: should we make a federal case out of it. Journal of Money, Credit, and Banking, 33(1), pp.1-15. FactSet Research Systems. 2010. Baltic Dry Index (BDI-BAX). Federal Reserve Bank of New York. 2009. Foreign Exchange, Historical Data. http://www.newyorkfed.org/markets/foreignex.html. Federal Reserve Bank of St. Louis. 2010. Economic Research Database, Exchange Rates, http://research.stlouisfed.org/fred2/ Feenstra, Robert C. 2004. Advanced International Trade: Theory and Evidence. Princeton, NJ: Princeton University Press. 55 Frankel, Jeffrey A., and David Romer. 1999. Does trade cause growth? American Economic Review, 89, pp. 379-99. Freund, Caroline and Martha Denisse Pierola. 2008. Export surges: the power of a competitive currency. Policy Research Working Paper, 450. Washington, DC: World Bank. Gala, Paulo. 2008. Real exchange rate levels and economic development: theoretical analysis and econometric evidence. Cambridge Journal of Economics, 32, pp. 273-288. Goldstein, Itay and Assaf Razin. 2005. Foreign direct investment vs. foreign portfolio investment. NBER Working Paper,11047. Goodwin, Barry K. 1992. Multivariate cointegration tests and the law of one price in international wheat markets. Review of Agricultural Economics,14(1), pp. 117-24. Gospodinov, Nikolay and Serena Ng. 2010. Commodity prices, convenience yields and inflation. Working paper at http://www.columbia.edu/~sn2294/papers/cy.pdf Groen, Jan J. J. and Paolo A. Pesenti. 2010. Commodity prices, commodity currencies, and global economic development. NBER Working Papers,15743. Gwartney James and Robert Lawson. 2009. Economic Freedom Dataset, published in Economic Freedom of the World: 2009 Annual Report, Economic Freedom Network, http://www.freetheworld.com/2009/2009Dataset.xls. Helpman, Elhanan. 1984. A simple theory of international trade with multinational corporations. Journal of Political Economy,92, pp. 451-471. Helpman, Elhanan. 1987. Imperfect competition and international trade: evidence from fourteen industrial countries. Journal of the Japanese and International Economies, 1, pp. 62-81. Hendry, David. 1995. Dynamic Econometrics. Oxford, UK: Oxford University Press. Hummels, David and Alexandra Skiba. 2002. Shipping the good apples out? An empirical confirmation for the Alchian-Allen conjecture. NBER Working Paper, 9023. Hummels, David and Alexandra Skiba. 2004. Shipping the good apples out? An empirical confirmation for the Alchian-Allen conjecture. Journal of Political Economy, 112(6), pp.1384-1402. Im, Kyung So, M. Hashem Pesaran, and Yongcheol Shin. 2003. Testing for unit roots in heterogeneous panels. Journal of Econometrics, 115(1), pp. 53-74. Institutional Investor. 2003-2007. Country Credit Ratings, September 2003-2007. International Monetary Fund. 2008. Direction of Trade Statistics Online. http://www.imfstatistics.org/DOT/ 56 International Monetary Fund. 2008. World Economic Outlook Database, April 2008. http://www.imf.org/external/pubs/ft/weo/2008/01/weodata/index.aspx International Monetary Fund. 2009. International Financial Statistics CD-ROM, February 2009. International Monetary Fund. 2010. Data and Statistics, Primary Commodity Prices, http://www.imf.org/external/np/res/commod/index.asp Isard, Peter. 1977. How far can we push the ?law of one price?? American Economic Review, 67(5), pp. 942-48. Katzner, Donald W. and Mikail J. Nikomarov. 2008. Exercises in futility. The Japanese Economy, 35, pp.29-58. Kellard, Neil and Mark E. Wohar. 2006. On the prevalence of trends in primary commodity prices. Journal of Development Economics, 79(1), pp. 146-67. Kilian, Lutz. 2009. Not all price shocks are alike: distangling demand and supply shocks in the crude oil market. American Economic Review, 99(3), pp.1053-1069. Leamer, Edward. 1983. Let?s take the con out of econometrics. American Economic Review, 73(1), pp. 31-43. Leamer, Edward. 1985. Sensitivity analysis would help. American Economic Review, 75(3), pp. 308-313. Levin, Andrew, Chien-Fu Lin, and Chia-Shang James Chu. 2002. Unit root tests in panel data: asymptotic and finite sample properties. Journal of Econometrics,98 (1), pp. 1-24. Levine, Ross, and Sara Zervos. 1998. Stock markets, banks, and economic growth. American Economic Review,88(3), pp. 537-558. Levine, Ross, Norman Loayza, and Thorsten Beck. 2000. Financial intermediation and growth: causality and causes. Journal of Monetary Economics, 46(1), pp. 31-88. Lo, Ming Chien and Eric Zivot. 2001. Threshold cointegration and nonlinear adjustment to the law of one price. Macroeconomic Dynamics, 5(4), pp. 533-76. Maddala, G. S. and Shaowen Wu. 1999. A comparative study of unit root tests with panel data and a new simple test. Oxford Bulletin of Economics and Statistics, Special Issue v.61, pp. 631-52. Markusen, James R. 1983. Factor movements and commodity trade as complements. Journal of International Economics,14, pp. 341-356. 57 McCallum, John. 1995. National borders matter: Canada-U.S. Regional trade patterns. American Economic Review, 85(3), pp. 615-23. McCracken, Michael W. 2007. Asymptotics for out of sample tests of Granger causality. Journal of Econometrics, 140( 2), pp. 719-52. McKinnon, Ronald I. 1993. The Order of Economic Liberalization: Financial Control in the Transition to a Market Economy, 2d ed. Baltimore, MD: The John Hopkins University Press. Meese, Richard and Kenneth Rogoff. 1982. The out-of-sample failure of empirical exchange rate models: sampling error or misspecification? International Monetary Fund?s International Finance Discussion Papers,204. Michael, Panos, A. Robert Nobay, and David Peel. 1994. Purchasing power parity yet again: evidence from spatially separated commodity markets. Journal of International Money and Finance, 13(6), pp. 637-57. Moon, Hyungsik Roger and Benoit Perron. 2004. Testing for a unit root in panels with dynamic factors. Journal of Econometrics, 122(1), pp. 81-126. Neary, Peter J. 1995. Factor mobility and international trade. Canadian Journal of Economics, 28, pp. S4-S23. Neary, Peter J. 2002. Foreign direct investment and the single market. The Manchester School,70(3), pp. 291-314. Neary, Peter J. 2009. Trade costs and foreign direct investment. International Review of Economics and Finance, 18, pp. 207-218. Obstfeld, Maurice and Alan M. Taylor. 1997. Nonlinear aspects of good-market arbitrage and adjustment: Heckscher?s commodity points revisited. NBER Working Papers, 6053. Parsley, David C. and Shang-Jin Wei. 2001. Explaining the border effect: the role of exchange rate variability, shipping costs, and geography. Journal of International Economics, 55(1), pp. 87-105. Pesaran, M. Hashem. 2004. General diagnostic tests for cross section dependence in panels. CESIFO Working Paper, 1229. Pesaran, M. Hashem. 2007. A simple panel unit root test in the presence of cross-section dependence. Journal of Applied Econometrics,22 (2), pp. 265-312. Phillips, Peter C. B. and Donggyu Sul. 2003. Dynamic panel estimation and homogeneity testing under cross section dependence. Econometrics Journal, 6(1), pp. 217-59. 58 Radelet, Steven and Jeffrey D. Sachs. 1998. The East Asian financial crisis: diagnosis, remedies, prospects. Brookings Papers on Economic Activity, 1. Ruffin, Roy J. and Farhad Rassekh. 1986. The role of foreign direct investment in U.S. capital outflows. American Economic Review, 76(5), pp. 1126-1130. Sala-i-Martin, Xavier. 1997. I have just run two million regressions. American Economic Review, 87(2), pp. 178-183. Saito, Tetsuya. 2007. Shipping the good apples out: another proof with a graphical representation. MPRA Paper, 1297.http://mpra.ub.uni-muenchen.de/1297/. Sarno, Lucio, Mark P. Taylor, and Ibrahim Chowdhury. 2004. Nonlinear dynamics in deviations from the law of one price: a broad-based empirical study. Journal of International Money and Finance,23(1), pp. 1-25. Silberberg, Eugene and Wing Suen. 2001. The Structure of Economics: A Mathematical Analysis. (3rd ed.). McGraw-Hill, New York. Smith, Alasdair. 1987. Strategic investment, multinational corporations and trade policy. European Economic Review, 31, pp. 89-96. Stock, James H. and Mark W. Watson. 1993. A simple estimator of cointegrating vectors in higher order integrated systems. Econometrica, 61(4), pp.783-820. Tharp, Mike. 1981. U.S., Japan in car pact. New York Times, May 1, 1981, A1. Tinbergen, Jan. 1962. Shaping the World Economy. New York: The Twentieth Century Fund. United States Department of Energy Energy Information Administration. 2009, 2010. World Crude Oil Prices and Natural Gas Prices. http://www.eia.doe.gov/ United States International Trade Commission. 2009, 2010. Interactive Tariff and Trade Database. http://dataweb.usitc.gov/ Wang, Dabin and William G. Tomek. 2007. Commodity prices and unit root tests. American Journal of Agricultural Economics Association,89 (4), pp. 873-889. West, Kenneth D. 1996. Asymptotic inference about predictive ability. Econometrica, 64, pp. 1067-1084. World Economic Forum. 2003-2007. The Global Competitiveness Report, 2003-2007. World Economic Forum. 2009. The Financial Development Report, 2009. 59 APPENDIX 1 Tables and Figures for Chapter 1 60 Figure 1.1. Sources of funds for a corporation Internal Within group External fundssssourcess Retained earnings Intra-company debt Funds from parent company or corporate affiliates Loans Equity Debt securities 61 Table 1.1. List of countries Main countries Partner countries Germany Japan United States Algeria Australia Austria Belgium Bulgaria Canada Chile China Colombia Costa Rica Croatia Czech Republic Denmark Dominican Republic Ecuador Finland France Germany Greece Hungary Ireland Israel Italy Japan Luxembourg Malaysia Morocco Netherlands New Zealand Nigeria Norway Pakistan Philippines Poland Portugal Romania Russian Federation Singapore Slovak Republic South Africa Spain Sweden Switzerland Tunisia Ukraine United Kingdom United States Venezuela 62 Table 1.2. Variable descriptions and summary statistics (a) Variable descriptions Variable Variable label Description TRAijt ln[(xij+xji)/(gdpi*gdpj)] xij+xji is trade between country i and country j in million US$ gdpi(j) is GDP of country i(j) in billion US$ DISij ln(disij) disij is distance between country i and country j in kilometers LCBij land common border 1 = share common border, 0 = no common border DEVijt stage of development 1 = same stage, 0 = different stage LNAijt ln(lnai*lnaj) lnai(j) is score of access to loan of country i(j) EQAijt ln(eqai*eqaj) eqai(j) is score of access to equity of country i(j) CCRijt ln(max ccri,ccrj/min ccri,ccrj) ccri(j) is country credit rating of country i(j) CAPijt ln(capi*capj) capi(j) is score of capital controls in country i(j) RERijt ln(max reri,rerj/min reri,rerj) reri(j) is real effective exchange rate in country i(j) (b) Summary statistics Variable Observation Mean Standard Deviation Minimum Maximum TRAijt 690 -5.036 1.336 -7.621 -1.754 DISij 690 8.540 1.020 5.156 9.843 LCBij 690 0.0725 0.259 0 1 DEVijt 690 0.4609 0.499 0 1 LNAijt 690 2.681 0.339 1.658 3.352 EQAijt 690 3.395 0.229 2.591 3.728 CCRijt 690 0.346 0.350 0 1.525 CAPijt 690 3.642 0.375 2.480 4.345 RERijt 690 0.210 0.170 0.000 0.953 63 Table 1.3. Pair-wise correlations between variables TRAijt DISij LCBij DEVijt LNAijt EQAijt CCRijt CAPijt RERijt TRAijt 1 DISij -0.702* 1 (0.000) LCBij 0.479* -0.679* 1 (0.000) (0.000) DEVijt 0.034 -0.156* 0.190* 1 (0.376) (0.000) (0.000) LNAijt 0.089* -0.139* 0.122* 0.551* 1 (0.019) (0.000) (0.001) (0.000) EQAijt 0.066 0.039 0.087* 0.567* 0.489* 1 (0.082) (0.301) (0.022) (0.000) (0.000) CCRijt -0.064 0.171* -0.206* -0.725* -0.578* -0.568* 1 (0.092) (0.000) (0.000) (0.000) (0.000) (0.000) CAPijt 0.117* -0.118* 0.134* 0.504* 0.240* 0.278* -0.358* 1 (0.002) (0.002) (0.000) (0.000) (0.000) (0.000) (0.000) RERijt -0.359* 0.361* -0.244* -0.061 -0.115* -0.118* 0.033 -0.101* 1 (0.000) (0.000) (0.000) (0.110) (0.003) (0.002) (0.387) (0.008) * indicates correlation is significant at 5% level. P-values are in parentheses. 64 Table 1.4. Regression results (a) Country-specific and country-pair time fixed effects? Main countries: United States, Japan, and Germany. Partner countries TRAijt (1) All partner countries (2) Developed countries (3) Less developed countries Gravity DISij -0.643*** (0.038) -0.483*** (0.053) -0.794*** (0.053) LCBij 0.023 (0.113) 0.570*** (0.133) 0.116 (0.201) DEVijt -1.321*** (0.120) --- --- External Funds LNAijt 0.735*** (0.205) 0.808** (0.315) 1.097*** (0.258) EQAijt 0.425** (0.173) -1.839*** (0.504) 0.453** (0.196) International Finance CCRijt -0.883*** (0.157) -1.800*** (0.531) -0.622*** (0.181) CAPijt 0.548*** (0.111) -1.209*** (0.382) 0.588*** (0.123) RERijt -0.292 (0.286) 2.779*** (0.686) -0.776** (0.327) Intercept -3.349*** (0.665) 8.927*** (2.320) -3.267*** (0.782) Observations 690 318 372 Adjusted R2 0.937 0.958 0.936 ? Anderson and van Wincoop (2003) and Baier and Bergstrand (2007), selected model. ***,**,* indicate variable is significant at 1%, 5%, and 10% levels, respectively. Standard errors are in parentheses. --- indicates variable is omitted. 65 (b) Country-specific time fixed effects?? Main countries: United States, Japan, and Germany. Partner countries TRAijt (1) All partner countries (2) Developed countries (3) Less developed countries Gravity DISij -1.100*** (0.051) -0.903*** (0.070) -1.314*** (0.068) LCBij -0.335* (0.172) -0.092 (0.197) 0.467 (0.313) DEVijt -0.525** (0.214) --- --- External Funds LNAijt 0.216 (0.222) 0.02 (0.297) 0.395 (0.300) EQAijt 1.129*** (0.294) 2.852*** (0.000) 0.924*** (0.328) International Finance CCRijt 0.442* (0.245) 0.912 (0.721) 0.311 (0.268) CAPijt 0.264 (0.205) 0.240 (0.564) 0.385* (0.229) RERijt -0.465 (0.365) -0.279 (0.585) -0.867* (0.458) Intercept -0.683 (1.110) -8.313** (3.765) 1.098 (1.246) Observations 690 318 372 Adjusted R2 0.700 0.781 0.698 ? ?Anderson and van Wincoop (2003), ignoring country-pair fixed effects. ***,**,* indicate variable is significant at 1%, 5%, and 10% levels, respectively. Standard errors are in parentheses. --- indicates variable is omitted. 66 (c) OLS??? Main countries: United States, Japan, and Germany. Partner countries TRAijt (1) All partner countries (2) Developed countries (3) Less developed countries Gravity DISij -0.937*** (0.050) -0.763*** (0.068) -1.129*** (0.072) LCBij -0.030 (0.185) 0.263 (0.215) 0.307 (0.348) DEVijt -0.531*** (0.115) --- --- External Funds LNAijt 0.029 (0.132) -0.046 (0.183) 0.078 (0.186) EQAijt 1.222*** (0.203) 1.693*** (0.448) 1.187*** (0.239) International Finance CCRijt 0.272* (0.157) 0.220 (0.485) 0.332* (0.179) CAPijt 0.321*** (0.107) 0.302 (0.191) 0.357*** (0.136) RERijt -0.648*** (0.224) -0.614 (0.374) -0.856*** (0.293) Intercept -2.147** (0.862) -5.545*** (1.750) -0.608 (1.075) Observations 690 318 372 Adjusted R2 0.537 0.621 0.499 ?? ?Ordinary Least Squares, ignoring fixed effects. ***,**,* indicate variable is significant at 1%, 5%, and 10% levels, respectively. Standard errors are in parentheses. --- indicates variable is omitted. 67 Table 1.5. Fixed effects regressions without pair-country interactions for the financial development variables Main countries: United States, Japan, and Germany. Partner countries TRAijt (1) All partner countries (2) Developed countries (3) Less developed countries Gravity DISij -0.548*** (0.034) -0.419*** (0.061) -0.585*** (0.046) LCBij 0.095 (0.099) 0.519*** (0.140) 0.144 (0.147) DEVijt -1.253*** (0.108) --- --- External Funds LNAit LNAjt 0.990 (0.999) 0.074 (0.236) 6.867*** (0.692) 0.849** (0.352) 1.056 (0.916) 0.423 (0.326) EQAit EQAjt -3.402 (2.156) 0.320* (0.169) --- -1.871*** (0.591) -3.068 (1.977) 0.455*** (0.167) International Finance CCRit CCRjt 6.921* (3.607) 1.280*** (0.149) --- 0.642 (0.522) 7.466** (3.340) 1.259*** (0.165) CAPit CAPjt 3.641*** (0.632) 0.154 (0.120) --- -0.990** (0.391) 3.602*** (0.595) 0.099 (0.125) RERit RERjt -2.291 (1.598) 1.761*** (0.222) --- -0.772 (0.619) -2.730* (1.506) 2.268*** (0.249) Intercept -36.46*** (11.75) -4.69 (3.114) -39.824*** (10.827) Observations 690 318 372 Adjusted R2 0.955 0.954 0.967 ***,**,* indicate variable is significant at 1%, 5%, and 10% levels, respectively. Standard errors are in parentheses. --- indicates variable is omitted. 68 APPENDIX 2 Tables and Figures for Chapter 2 69 Table 2.1. Commodity price data description Category ID Commodity IMF Code Metals 1 Aluminum, LME standard grade, minimum purity, CIF UK PALUM 2 Copper, LME, grade A cathodes, CIF Europe PCOPP 3 Iron Ore Carajas PIORECR 4 Lead, LME, 99.97 percent pure, CIF European PLEAD 5 Nickel, LME, melting grade, CIF N Europe PNICK 6 Tin, LME, standard grade, CIF European PTIN 7 Zinc, LME, high grade, CIF UK PZINC 8 Uranium, NUEXCO, Restricted Price, US$ per pound PURAN Fuels 9 Coal thermal for export, Australia PCOALAU 10 Oil, Average of U.K. Brent, Dubai, and West Texas Intermediate POILAPSP 11 Oil, UK Brent, light blend 38 API, fob U.K. POILBRE 12 Oil, Dubai, medium, Fateh 32 API, fob Dubai POILDUB 13 Oil, West Texas Intermediate, 40 API, Midland Texas POILWTI 14 Natural Gas, BEA Food 15 Bananas, avg of Chiquita, Del Monte, Dole, U.S. Gulf delivery PBANSOP 16 Barley, Canadian Western No. 1 Spot PBARL 17 Beef, Australia/New Zealand frozen, U.S. import price PBEEF 18 Cocoa, ICO price, CIF U.S. & European ports PCOCO 19 Coconut Oil, Philippines/Indonesia, CIF Rotterdam PROIL 20 Fishmeal, 64/65 percent, any orig, CIF Rotterdam PFISH 21 Groundnut, U.S. runners, CIF European PGNUTS 22 Lamb, New Zealand, PL frozen, London price PLAMB 23 Maize, U.S. number 2 yellow, fob Gulf of Mexico PMAIZMT 24 Olive Oil, less that 1.5% FFA POLVOIL 25 Orange Brazilian, CIF France PORANG 26 Palm Oil, Malaysia and Indonesian, CIF NW Europe PPOIL 27 Hogs, 51-52% lean, 170-191 lbs, IL, IN, OH, MI, KY PPORK 28 Chicken, Ready-to-cook, whole, iced, FOB Georgia Docks PPOULT 29 Rice, 5 percent broken, nominal price quote, fob Bangkok PRICENPQ 30 Norwegian Fresh Salmon, farm bred, export price PSALM 31 Shrimp, U.S., frozen 26/30 count, wholesale NY PSHRI 32 Soybean Meal, 44 percent, CIF Rotterdam PSMEA 33 Soybean Oil, Dutch, fob ex-mill PSOIL 34 Soybean, U.S., CIF Rotterdam PSOYB 35 Sugar, EC import price, CIF European PSUGAEEC 36 Sugar, International Sugar Agreement price PSUGAISA 37 Sugar, U.S., import price contract number 14 CIF PSUGAUSA 38 Sunflower Oil, any origin, ex-tank Rotterdam PSUNO 39 Wheat, U.S. number 1 HRW, fob Gulf of Mexico PWHEAMT Beverages 40 Coffee, Other Milds, El Salvdor and Guatemala, ex-dock New York PCOFFOTM 41 Coffee, Robusta, Uganda and Cote dIvoire, ex-dock New York PCOFFROB 42 Tea, From July 1998,Kenya auctions, Best Pekoe Fannings. Prior, London auctions, CIF U.K. warehouses PTEA Raw Materials 43 Cotton, Liverpool Index A, CIF Liverpool PCOTTIND 44 Wool Coarse, 23 micron, AWEX PWOOLC 45 Wool Fine, 19 micron, AWEX PWOOLF Continue to next page. 70 Continued from previous page. Category ID Commodity IMF Code Industrial Inputs 46 Hides, U.S., Chicago, fob Shipping Point PHIDE 47 Log, soft, export from U.S. Pacific coast PLOGORE 48 Log, hard, Sarawak, import price Japan PLOGSK 49 Rubber, Malaysian, fob Malaysia and Singapore PRUBB 50 Sawnwood, dark red meranti, select quality PSAWMAL 51 Sawnwood, average of softwoods, U.S. West coast PSAWORE Note: i) All data is obtained from IMF Primary Commodity Prices database with an exception of natural gas (ID#14). The U.S. wellhead natural gas data is obtained from the U.S. Energy Information Administration. 71 Table 2.2. Augmented Dickey-Fuller test and Cross-section Dependence test results: Nominal commodity prices ID ADF p-value ID ADF p-value ID ADF p-value 1 -3.062* 0.026 18 -1.616 0.472 35 -1.641 0.456 2 -1.157 0.690 19 -1.826 0.359 36 -2.706 0.067 3 0.215 0.977 20 -1.212 0.666 37 -7.394* 0.000 4 -0.847 0.803 21 -4.047* 0.001 38 -2.699 0.069 5 -1.878 0.334 22 -2.207 0.197 39 -2.505 0.108 6 -2.078 0.246 23 -2.877* 0.043 40 -2.561 0.093 7 -2.104 0.229 24 -1.660 0.448 41 -2.061 0.246 8 -1.182 0.682 25 -1.998 0.278 42 -4.194* 0.000 9 -1.879 0.334 26 -2.996* 0.031 43 -3.487* 0.007 10 -0.912 0.787 27 -3.354* 0.011 44 -2.051 0.254 11 -0.999 0.754 28 -0.842 0.811 45 -3.412* 0.010 12 -0.818 0.819 29 -1.855 0.343 46 -3.181* 0.018 13 -1.048 0.738 30 -1.962 0.294 47 -1.550 0.504 14 -1.823 0.359 31 -2.608 0.085 48 -1.612 0.472 15 -2.940* 0.036 32 -3.043* 0.027 49 -1.367 0.601 16 -2.086 0.237 33 -2.654 0.077 50 -1.384 0.593 17 -1.961 0.294 34 -2.849* 0.048 51 -1.229 0.658 CD Statistic: 53.978, p-value: 0.000 Note: i) ADF denotes the augmented Dickey-Fuller t-statistic with an intercept. ii) Superscript * refers the case when the null of nonstationarity is rejected at the 5% significance level. iii) CD statistic is a cross-section dependence test statistic by Pesaran (2004) with the hypothesis of no cross-section dependence. 72 Table 2.3. Augmented Dickey-Fuller test and Cross-section Dependence test results: Relative commodity prices ID ADF p-value ID ADF p-value ID ADF p-value 1 -3.569* 0.006 18 -2.689 0.071 35 -2.492 0.108 2 -2.087 0.237 19 -2.807 0.053 36 -3.274* 0.014 3 -0.969 0.763 20 -2.232 0.189 37 -2.946* 0.035 4 -1.843 0.351 21 -3.226* 0.016 38 -3.200* 0.017 5 -2.663 0.075 22 -3.805* 0.002 39 -2.706 0.067 6 -2.501 0.108 23 -2.636 0.080 40 -2.360 0.141 7 -2.740 0.063 24 -2.669 0.074 41 -2.035 0.262 8 -1.879 0.334 25 -3.751* 0.003 42 -3.036* 0.028 9 -2.410 0.133 26 -3.231* 0.016 43 -2.445 0.116 10 -1.864 0.343 27 -1.493 0.536 44 -2.590 0.088 11 -1.952 0.294 28 -5.692* 0.000 45 -2.364 0.141 12 -1.766 0.391 29 -2.540 0.097 46 -2.487 0.108 13 -2.080 0.246 30 -2.090 0.237 47 -1.735 0.407 14 -2.284 0.165 31 -0.709 0.843 48 -2.876* 0.043 15 -3.813* 0.002 32 -2.915* 0.040 49 -2.631 0.081 16 -3.870* 0.002 33 -2.705 0.068 50 -2.336 0.149 17 -2.146 0.213 34 -2.688 0.071 51 -2.237 0.181 CD Statistic: 48.313, p-value: 0.000 Note: i) ADF denotes the augmented Dickey-Fuller t-statistic with an intercept with the null of nonstationarity. ii) Superscript * refers the case when the null hypothesis is rejected at the 5% significance level. iii) CD statistic is a cross-section dependence test statistic by Pesaran (2004) with the null hypothesis of no cross-section dependence. iv) Each commodity price is deflated by the U.S. consumer price index to obtain the relative price. 73 Table 2.4. PANIC test results: Nominal commodity prices Idiosyncratic Components ID ADF p-value ID ADF p-value ID ADF p-value 1 -2.220* 0.024 18 -1.047 0.270 35 -1.083 0.254 2 -2.839* 0.004 19 -2.995* 0.003 36 -1.558 0.108 3 -0.952 0.302 20 -1.084 0.254 37 -4.079* 0.000 4 -1.324 0.173 21 -4.856* 0.000 38 -4.688* 0.000 5 -2.194* 0.026 22 -2.225* 0.024 39 -3.410* 0.001 6 -1.237 0.197 23 -2.372* 0.016 40 -2.258* 0.022 7 -2.298* 0.019 24 -1.438 0.141 41 -1.842 0.060 8 -0.845 0.351 25 -3.851* 0.000 42 -2.699* 0.006 9 -3.326* 0.001 26 -3.448* 0.001 43 -3.406* 0.001 10 -2.117* 0.030 27 -3.075* 0.002 44 -2.158* 0.027 11 -2.205* 0.025 28 -5.730* 0.000 45 -2.171* 0.027 12 -1.986* 0.042 29 -1.288 0.181 46 -1.414 0.149 13 -2.336* 0.018 30 -2.078* 0.033 47 -1.566 0.108 14 -1.577 0.108 31 -1.507 0.124 48 -2.240* 0.023 15 -5.327* 0.000 32 -2.719* 0.006 49 -1.448 0.141 16 -1.191 0.213 33 -1.708 0.082 50 -2.057* 0.035 17 -1.578 0.108 34 -1.860 0.057 51 -1.220 0.205 Panel Test Statistics: 21.445*, p-value: 0.000 Common Factor Components ADF (Factor 1): -2.283, p-value: 0.165 ADF (Factor 2): -2.901*, p-value: 0.042 Note: i) ADF denotes the augmented Dickey-Fuller t-statistic with no deterministic terms (idiosyncratic components) and with an intercept (common factors) with the null hypothesis of nonstationarity. ii) Superscript * refers the case when the null hypothesis is rejected at the 5% significance level. 74 Table 2.5. PANIC test results: Relative commodity prices Idiosyncratic Components ID ADF p-value ID ADF p-value ID ADF p-value 1 -2.089* 0.032 18 -0.978 0.294 35 -1.311 0.173 2 -2.890* 0.004 19 -2.865* 0.004 36 -1.518 0.124 3 -0.798 0.367 20 -0.990 0.294 37 -3.429* 0.001 4 -1.396 0.157 21 -4.626* 0.000 38 -4.648* 0.000 5 -1.928* 0.048 22 -2.335* 0.018 39 -3.385* 0.001 6 -1.224 0.205 23 -2.233* 0.023 40 -2.078* 0.033 7 -2.201* 0.025 24 -1.658 0.091 41 -1.773 0.070 8 -0.807 0.367 25 -8.250* 0.000 42 -2.259* 0.022 9 -3.090* 0.002 26 -3.282* 0.001 43 -3.578* 0.001 10 -1.910* 0.050 27 -3.560* 0.001 44 -1.802 0.066 11 -1.974* 0.043 28 -2.269* 0.021 45 -2.316* 0.019 12 -2.062* 0.035 29 -1.114 0.246 46 -1.536 0.116 13 -2.197* 0.025 30 -2.038* 0.037 47 -1.666 0.090 14 -1.761 0.072 31 -1.868 0.056 48 -2.095* 0.031 15 -3.870* 0.000 32 -2.293* 0.020 49 -1.528 0.116 16 -1.071 0.262 33 -1.357 0.165 50 -2.214* 0.024 17 -1.170 0.221 34 -1.471 0.133 51 -1.671 0.089 Panel Test Statistics: 19.497*, p-value: 0.000 Common Factor Components ADF (Factor 1): -1.887, p-value; 0.326 ADF (Factor 2): -2.912*, p-value: 0.040 Note: i) ADF denotes the augmented Dickey-Fuller t-statistic with no deterministic terms (idiosyncratic components) and with an intercept (common factors) with the null hypothesis of nonstationarity. ii) Superscript * refers the case when the null hypothesis is rejected at the 5% significance level. iii) Each commodity price is deflated by the U.S. consumer price index to obtain the relative price. 75 Table 2.6. Out-of-sample forecast performance: Nominal commodity price factors K RRMSPE DMW 1 1.0122 0.7039** 2 1.0026 0.2607 3 1.0054 0.5352* 4 1.0134 1.1514** Note: i) Out-of-sample forecasting was recursively implemented by sequentially adding one additional observation from 180 initial observations toward 360 total observations. ii) k denotes the forecast horizon. iii) RRMSPE denotes the ratio of the root mean squared prediction error of the random walk hypothesis to the common factor model. iv) DMW denotes the test statistics of Diebold and Mariano (1995) and West (1996). v) * and ** denote rejection of the null hypothesis of equal predictability at the 10% and 5% significance levels, respectively. Critical values were obtained from McCracken (2007). Table 2.7. Out-of-sample forecast performance: Relative commodity price factors K RRMSPE DMW 1 1.0169 0.9804** 2 1.0062 0.5834* 3 1.0082 0.7753** 4 1.0167 1.3815*** Note: i) Out-of-sample forecasting was recursively implemented by sequentially adding one additional observation from 180 initial observations toward 360 total observations. ii) k denotes the forecast horizon. iii) RRMSPE denotes the ratio of the root mean squared prediction error of the random walk hypothesis to the common factor model. iv) DMW denotes the test statistics of Diebold and Mariano (1995) and West (1996). v) *, **, and ** denote rejection of the null hypothesis of equal predictability at the 10%, 5%, and 1% significance levels, respectively. Critical values were obtained from McCracken (2007). vi) Each commodity price is deflated by the U.S. consumer price index to obtain the relative price. 76 Figure 2.1. Number of factors estimation: Nominal commodity prices 77 Figure 2.2. Number of factors estimation: Relative commodity prices 78 Figure 2.3. Relative importance of common factors: Nominal commodity prices 79 Figure 2.4. Relative importance of common factors: Relative commodity prices 80 Figure 2.5. Factor loadings estimates: Nominal commodity prices 81 Figure 2.6. Factor loadings estimates: Relative commodity prices 82 Figure 2.7. U.S. nominal exchange rates vs. common factor estimates 83 Figure 2.8. Crude oil prices 84 APPENDIX 3 Tables and Figures for Chapter 3 85 Table 3.1. Import quantity and price data Small size car import Medium size car import Japan Korea Japan Korea 1995-2005 (monthly) g1855g3046g3037g3035 g1855g3046g3038g3039 g1855g3040g3037g3039 g1855g3040g3038g3035 Import quantity Average 9,122 6,007 199,235 30,752 Standard deviation 5,706 3,644 64,342 16,582 Price ($ per unit) Average 11,154 6,916 8,495 9,646 Standard deviation 2,050 653 1,932 916 1995-2008 (monthly) Import quantity Average 14,231 - 190,819 36,989 Standard deviation 11,349 - 59,653 19,603 Price ($ per unit) Average 11,522 - 8,284 9,082 Standard deviation 1,960 - 1,784 1,386 Note: (1) Small size cars: Motor cars and other motor vehicles for transport of persons with spark-ignition internal combustion reciprocating piston engine of a cylinder capacity exceeding 1,000 cc but not exceeding 1,500 cc (HTS 8703.22). (2) Medium size cars: Motor cars and other motor vehicles for transport of persons with spark-ignition internal combustion reciprocating piston engine of a cylinder capacity exceeding 1,500 cc but not exceeding 3,000 cc (HTS 8703.23). (3) Small size car imports from Korea discontinued after 2005. (4) Average price is calculated by dividing total Custom value by quantity of imports.The Customs value is the value of imports as appraised by the U.S. Customs. This value is defined as the price actually paid or payable for merchandise, excluding U.S. import duties, freight, insurance, and other charges. (5) Price data are deflated using Consumer Price Index. Data source: United States International Trade Commission (2009) and the Bureau of Labor Statistics (2009). 86 Figure 3.1. Import quantity data series 0 50 00 10 00 0 15 00 0 20 00 0 Ko re a 0 10 00 0 20 00 0 30 00 0 40 00 0 Ja pa n 1995m1 2000m1 2005m1 2010m1 Period Japan Korea U.S. Imports of Small Size Cars from Japan and Korea 0 20 00 0 40 00 0 60 00 0 80 00 0 Ko re a 10 00 00 20 00 00 30 00 00 40 00 00 Ja pa n 1995m1 2000m1 2005m1 2010m1 Period Japan Korea U.S. Imports of Medium Size Cars from Japan and Korea Data source: United States International Trade Commission (2009). 87 Figure 3.2. Car price data series 55 00 60 00 65 00 70 00 75 00 80 00 Ko rea 60 00 80 00 10 00 0 12 00 0 14 00 0 16 00 0 Ja pa n 1995m1 2000m1 2005m1 2010m1 Period Japan Korea (USD Custom value; CPI adjusted) Unit Price of Small Size Cars from Japan and Korea 60 00 70 00 80 00 90 00 10 00 0 11 00 0 Ko rea 40 00 60 00 80 00 10 00 0 12 00 0 14 00 0 Ja pa n 1995m1 2000m1 2005m1 2010m1 Period Japan Korea (USD Custom value; CPI adjusted) Unit Price of Medium Size Cars from Japan and Korea Data source: United States International Trade Commission (2009). 88 Figure 3.3. Actual shipping cost data series 20 0 30 0 40 0 50 0 Ko re a 20 0 40 0 60 0 80 0 Ja pa n 1995m1 2000m1 2005m1 2010m1 Period Japan Korea Actual CIF Charges for Small Size Cars (USD/per unit) 10 0 20 0 30 0 40 0 50 0 Ko re a 10 0 20 0 30 0 40 0 50 0 Ja pa n 1995m1 2000m1 2005m1 2010m1 Period Japan Korea Actual CIF Charges for Medium Size Cars (USD/per unit) Data source: Calculated based on related data from the United States International Trade Commission (2010). 89 Figure 3.4. Other data series 20 40 60 80 10 0 W TI_ Cr ud e_ Oi l_P ric e 0 50 00 10 00 0 15 00 0 Ba ltic _D ry_ Ind ex 1995m1 2000m1 2005m1 2010m1 Period Baltic_Dry_Index WTI_Crude_Oil_Price Baltic Dry Index and WTI Crude Oil Price (USD/barrel) Data source: U.S. Bureau of Energy Administration (2009) and FactSet Research Systems (2010). .00 06 .00 08 .00 1 .00 12 .00 14 KR W .00 7 .00 8 .00 9 .01 .01 1 .01 2 JP Y 1995m1 2000m1 2005m1 2010m1 Period JPY KRW USD Exchange Rate vs. JPY and KRW Data source: Federal Reserve Bank of New York (2009). 90 Table 3.2. Augmented Dickey-Fuller (ADF) test results Variable ADF test 1995m1-2005m12 1995m1-2008m12 Import quantity g1855g3046g3037g3035 -3.14* -2.06 g1855g3046g3038g3039 -2.32 g1855g3040g3037g3039 -3.09* g1855g3040g3038g3035 -2.76 -3.03* Car price g1868g3046g3037g3035 -2.12 -2.54 g1868g3046g3038g3039 -2.045 g1868g3040g3037g3039 -4.22* g1868g3040g3038g3035 -2.78 -2.37 Freight rate g1858 -0.50 -1.28 Actual shipping costs (cost, insurance and freight) g1872g3046g3037g3035 -5.85* -6.58* g1872g3046g3038g3039 -5.12* g1872g3040g3037g3039 -5.35* g1872g3040g3038g3035 -5.78* -4.08* USD/JPY g1857g3037 -2.25 -1.92 USD/KRW g1857g3038 -1.93 -1.99 * indicates the result is significant at a 5% level of significance compared to the MacKinnon critical value of 2.89. 91 Table 3.3. Regressions and cointegration test results (1995m1-2005m12) (a) Regressions using freight rate Small size car import Medium size car import Japan Korea Japan Korea 1995-2005 (monthly) g1855g3046g3037g3035 g1855g3046g3038g3039 g1855g3040g3037g3039 g1855g3040g3038g3035 Car price,g1868g3036 0.537* (0.305) 4.530*** (0.856) -0.384*** (0.083) -0.115 (0.458) Freight rate,g1858 0.291* (0.153) -0.865*** (0.222) -0.616*** (0.050) 1.054*** (0.124) Exchange rate,g1857g3036 2.954*** (0.422) 3.976*** (0.871) 0.724*** (0.193) -1.149** (0.479) Intercept 15.78*** (2.565) -25.33*** (7.369) 23.60*** (1.171) 3.398 (4.221) Observation Adjusted R2 Breusch-Godfrey LM ARCH 131 0.415 66.54 13.56 131 0.397 65.77 8.11 131 0.587 74.85 39.27 131 0.413 82.00 14.54 Engle-Granger test EG? = -3.8 -4.62 -4.42 -4.21 -3.93 ***,**,* indicate that the coefficient is significant at 1%, 5%, and 10% level, respectively; standard errors are in parentheses. The freight rate is based on Baltic Dry Index as an increasing function of oil price (with a one-month lag). (b) Regressions using actual shipping costs (cost, insurance, and freight) Small size car import Medium size car import Japan Korea Japan Korea 1995-2005 (monthly) g1855g3046g3037g3035 g1855g3046g3038g3039 g1855g3040g3037g3039 g1855g3040g3038g3035 Car price,g1868g3036 1.013*** (0.258) 5.884*** (1.095) -0.208 (0.191) 0.277 (0.435) Shipping costs,g1872g3036 0.142 (0.179) -1.527*** (0.484) -0.278* (0.146) -1.735*** (0.251) Exchange rate,g1857g3036 3.374*** (0.389) -0.703 (0.630) 0.248 (0.282) -1.946*** (0.226) Intercept 14.713*** (3.584) -39.95*** (14.05) 16.66*** (1.499) 4.182 (4.323) Observation Adjusted R2 Breusch-Godfrey LM ARCH 132 0.399 69.21 20.47 132 0.272 55.19 7.63 132 0.057 104.88 77.38 132 0.463 55.19 7.630 Engle-Granger test EG? = -3.8 -4.45 -4.02 -2.94 -5.44 ***,**,* indicate that the coefficient is significant at 1%, 5%, and 10% level, respectively; standard errors are in parentheses. Actual shipping cost is calculated by deducting the Custom value from the CIF (cost, insurance, and freight) value, and the result is divided by the total quantity of car imports. The Custom value is defined by the USITC (2010) as ?the price actually paid or payable for merchandise, excluding U.S. import duties, freight, insurance, and other charges.? The CIF value excludes U.S. import duties. 92 Table 3.4. Regressions and cointegration test results for higher quality cars, extending the period of observations (1995m1-2008m12) (a) Regressions using freight rate Small size car import Medium size car import Japan Korea Japan Korea 1995-2008 (monthly) g1855g3046g3037g3035 g1855g3046g3038g3039 g1855g3040g3037g3039 g1855g3040g3038g3035 Car price,g1868g3036 -0.184 (1.054) - - 0.424 (0.299) Freight rate,g1858 1.054*** (0.096) - - 0.883*** (0.092) Exchange rate,g1857g3036 1.870*** (0.418) - - -1.039*** (0.388) Intercept 11.79*** (2.716) - - -0.269 (3.180) Observation Adjusted R2 Breusch-Godfrey LM ARCH 167 0.631 102.58 26.53 - - 167 0.506 100.21 19.58 Engle-Granger test EG? = -3.8 -4.27 - - -4.64 ***,**,* indicate that the coefficient is significant at 1%, 5%, and 10% level, respectively; standard errors are in parentheses. The freight rate is based on Baltic Dry Index as an increasing function of oil price (with a one-month lag). (b) Regressions using actual shipping costs (cost, insurance, and freight) Small size car import Medium size car import Japan Korea Japan Korea 1995-2008 (monthly) g1855g3046g3037g3035 g1855g3046g3038g3039 g1855g3040g3037g3039 g1855g3040g3038g3035 Car price,g1868g3036 2.177*** (0.335) - - 0.363 (0.369) Shipping costs,g1872g3036 0.293 (0.246) - - -1.457*** (0.229) Exchange rate,g1857g3036 3.502*** (0.516) - - -1.468*** (0.425) Intercept 3.879 (4.887) - - 15.39*** (2.657) Observation Adjusted R2 Breusch-Godfrey LM ARCH 168 0.359 126.15 74.97 - - 168 0.383 99.98 14.58 Engle-Granger test EG? = -3.8 -3.53 - - -4.92 ***,**,* indicate that the coefficient is significant at 1%, 5%, and 10% level, respectively; standard errors are in parentheses. Actual shipping cost is calculated by deducting the Custom value from the CIF (cost, insurance, and freight) value, and the result is divided by the total quantity of car imports. The Custom value is defined by the USITC (2010) as ?the price actually paid or payable for merchandise, excluding U.S. import duties, freight, insurance, and other charges.? The CIF value excludes U.S. import duties. 93 Table 3.5. Stock-Watson dynamic OLS regression results (1995m1-2005m12) (a) Regressions using freight rate Small size car import Medium size car import Japan Korea Japan Korea 1995-2005 (monthly) g1855g3046g3037g3035 g1855g3046g3038g3039 g1855g3040g3037g3039 g1855g3040g3038g3035 Car price,g1868g3036 -3.021*** (0.951) 4.97*** (1.480) -0.486*** (0.062) -1.916* (0.962) Freight rate,g1858 2.676*** (0.603) -0.278 (0.421) -1.122*** (0.083) 2.628*** (0.378) Exchange rate,g1857g3036 -4.395** (1.824) -5.635** (2.602) -0.288 (0.717) -5.381*** (1.357) Intercept -3.931 (10.05) -33.16*** (11.01) 23.37*** (3.566) 8.545 (7.326) F value Prob > F 63.91 0.000 64.53 0.000 813.3 0.000 66.83 0.000 ***,**,* indicate that the coefficient is significant at 1%, 5%, and 10% level, respectively; Newey-West standard errors are in parentheses. The freight rate is based on Baltic Dry Index as an increasing function of oil price (with a one-month lag). (b) Regressions using actual shipping costs (cost, insurance, and freight) Small size car import Medium size car import Japan Korea Japan Korea 1995-2005 (monthly) g1855g3046g3037g3035 g1855g3046g3038g3039 g1855g3040g3037g3039 g1855g3040g3038g3035 Car price,g1868g3036 0.774 (0.487) 4.940*** (1.100) -0.167 (0.245) 1.591* (0.832) Shipping costs,g1872g3036 0.061 (0.248) 1.122** (0.490) -0.445** (0.217) -1.750*** (0.642) Exchange rate,g1857g3036 4.753*** (0.975) -12.97*** (3.415) -1.172** (0.476) -1.868 (2.730) Intercept 23.859*** (7.878) -41.19*** (9.566) 10.42*** (2.776) 5.897 (8.293) F value Prob > F 4.03 0.000 34.18 0.000 15.68 0.000 3.46 0.000 ***,**,* indicate that the coefficient is significant at 1%, 5%, and 10% level, respectively; Newey-West standard errors are in parentheses. Note: g1855g3040g3037g3039 is not cointegrated (see Table 3.3b). Actual shipping cost is calculated by deducting the Custom value from the CIF (cost, insurance, and freight) value, and the result is divided by the total quantity of car imports. The Custom value is defined by the USITC (2010) as ?the price actually paid or payable for merchandise, excluding U.S. import duties, freight, insurance, and other charges.? The CIF value excludes U.S. import duties. 94 Table 3.6. Stock-Watson dynamic OLS regression results for higher quality cars, extending the period of observations (1995m1-2008m12) (a) Regressions using freight rate Small size car import Medium size car import Japan Korea Japan Korea 1995-2008 (monthly) g1855g3046g3037g3035 g1855g3046g3038g3039 g1855g3040g3037g3039 g1855g3040g3038g3035 Car price,g1868g3036 -1.091*** (0.417) - - 1.619*** (0.480) Freight rate,g1858 1.584*** (0.218) - - 1.022*** (0.115) Exchange rate,g1857g3036 1.189 (1.589) - - -7.577*** (1.413) Intercept 12.87 (7.756) - - -12.24** (4.919) F value Prob > F 24.04 0.000 - - 65.10 0.000 ***,**,* indicate that the coefficient is significant at 1%, 5%, and 10% level, respectively; Newey-West standard errors are in parentheses. The freight rate is based on Baltic Dry Index as an increasing function of oil price (with a one-month lag). (b) Regressions using actual shipping costs (cost, insurance, and freight) Small size car import Medium size car import Japan Korea Japan Korea 1995-2008 (monthly) g1855g3046g3037g3035 g1855g3046g3038g3039 g1855g3040g3037g3039 g1855g3040g3038g3035 Car price,g1868g3036 1.654* (0.941) - - 1.175** (0.468) Shipping costs,g1872g3036 -0.413 (0.587) - - -1.551*** (0.407) Exchange rate,g1857g3036 10.23*** (1.829) - - -3.481* (1.831) Intercept 44.55*** (15.63) - - 8.599** (3.352) F value Prob > F 3.61 0.000 - - 5.07 0.000 ***,**,* indicate that the coefficient is significant at 1%, 5%, and 10% level, respectively; Newey-West standard errors are in parentheses. Actual shipping cost is calculated by deducting the Custom value from the CIF (cost, insurance, and freight) value, and the result is divided by the total quantity of car imports. The Custom value is defined by the USITC (2010) as ?the price actually paid or payable for merchandise, excluding U.S. import duties, freight, insurance, and other charges.? The CIF value excludes U.S. import duties.