Capital and credit market integration and real economic contagion during the global financial crisis

Accepted Manuscript Title: Capital and credit market integration and real economic contagion during the global financial crisis Author: Ju Hyun Pyun, Jiyoun An PII: DOI: Reference:

S0261-5606(16)30028-6 http://dx.doi.org/doi: 10.1016/j.jimonfin.2016.04.004 JIMF 1670

To appear in:

Journal of International Money and Finance

Please cite this article as: Ju Hyun Pyun, Jiyoun An, Capital and credit market integration and real economic contagion during the global financial crisis, Journal of International Money and Finance (2016), http://dx.doi.org/doi: 10.1016/j.jimonfin.2016.04.004. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Capital and Credit Market Integration and Real Economic Contagion during the Global Financial Crisis Ju Hyun Pyun† Korea University Business School

Jiyoun An‡ Kyung Hee University

This version: January 18, 2016 Highlights 

We study the role of financial integration in the transmission of the global financial crisis to real business cycle.



High capital market integration with the United States amplified the transmission of the shock during the global financial crisis.



High credit market integration with the United States buffered against the spread of the global financial crisis to business cycle co-movement.

Abstract This study investigates the role of financial integration in the spread of the global financial crisis. In particular, this study shows how the effect of the crisis on real business cycle co-movement varied for capital and credit market integration, using a sample of 58 countries in 2001–2013. 

We are grateful to Jay M. Chung, Andrew Clare, Stefan Eichler, Stanimira Milcheva, Daehwan Kim, Bokyoung Park, Dyani Poedjioetomo, Alberto Pozzolo, Dong-Eun Rhee, and two anonymous referees for their valuable comments and suggestions. We also thank the participants of the 2013 KIC-IIS-KIEP International Conference, 2014 IFABS conference, and 2014 Asian Finance Association Conference for useful discussion. This research has been financially supported by the Korea Institute for International Economic Policy (KIEP) and was prepared for the 2013 KIEP research paper series (written in Korean). All errors and omissions are our own. † Business School, Korea University, 145 Anam-Ro, Seongbuk-Gu, Seoul 136-701, Tel: 82-2-3290-2610. Email: [email protected] ‡ Corresponding author: Kyung Hee University, 1732 Deogyoung-daero, Giheung-gu, Yongin-si, Gyeonggi-do 446701, Tel: 82-31-201-3884, Email: [email protected]

1 Page 1 of 46

During the global financial crisis, the United States—the epicenter of the crisis—experienced a severe downturn in the real economy, and other countries followed suit. We find that during the global financial crisis, the business cycle co-movements between the United States and the rest of the world were stronger when the level of capital market integration between them was higher. However, the co-movements were weaker when the level of credit market integration was higher. These findings are robust even when including investment channels, local fundamental factors, endogenous policy responses across countries, and alternative measures for financial integration and business cycle co-movements.

JEL classification: E32; F15; F36 Keywords: Capital market integration; Credit market integration; Global financial crisis; Real business cycle co-movements

2 Page 2 of 46

1. Introduction The cross-border holdings of international financial markets have increased remarkably since the 2000s, as shown in Lane and Milesi-Ferretti (2007) and Lane (2013). Other previous literature has confirmed the ongoing process of financial globalization, particularly among advanced economies. However, the global financial crisis (GFC)—the most serious financial crisis since the Great Depression in 1929—has raised doubts about the benefits of financial globalization.1 Negative shocks originating from the US financial market quickly spread globally through various channels during the GFC. Many countries experienced severe recessions following the US economic downturn. Several studies have assessed the existence of contagion during financial crises. 2 Some have emphasized the role of financial linkages in contagion, arguing that financial linkages acted as the main transmission channels of cross-country shocks during financial crises in the 1990s (e.g., Baig and Goldfajn, 1999; Caramazza, Ricci, and Salgado, 2004; Kaminsky and Reinhart, 2000; Van Rijckeghem and Weder, 2003). However, Lane (2013) has provided market-specific views on financial integration during the GFC. He suggests that emerging economies benefited from their “structure of international investment” as holding both liquid foreign currency and safe foreign debt assets during the GFC provided a substantial buffer. However, the portfolio profiles of many advanced economies—such as “long equity, short debt”—proved risky in the face of declining financial markets because of negative valuation effects. According to Lane

1

Baele, Ferrando, Hördahl, Krylova, and Monnet (2004) summarize the benefits of financial integration shown in previous literature: financial integration provides greater risk sharing by diversifying portfolios, improves capital allocation and makes the financial system more efficient by reducing transaction barriers, and creates economic growth via financial development. 2 Please see Claessens, Dornbusch, and Park (2001) for an excellent review of the literature on contagion.

3 Page 3 of 46

(2013), we conjecture that various financial instruments―such as equities, bonds, derivatives, and loans―and their international linkages play different roles in spreading shocks across countries.3 In addition, although extensive studies on finance have focused on contagion across financial markets, relatively few have examined the transmission of negative financial shock to the real economy. Frankel and Saravelos (2012) and Rose and Spiegel (2010, 2012) approach the issue of contagion in the real economy during the GFC from a macroeconomic perspective. Kalemli-Ozcan, Papaioannou, and Perri (2013a) investigate the effect of banking integration on real business cycle synchronization during the GFC. However, previous literature has not fully examined the effect of capital (equity) and credit (debt) market integration on the real economy. Our study investigates how the effect of the GFC on real business cycle co-movement varied for capital and credit market integration for 57 countries with the United States, the origin of the GFC, during 2001–2013. When the crisis is contagious, both the origin and other countries experience economic downturn and their business cycles become more synchronized. Thus, we interpret the effect of the crisis on business cycle co-movement between the United States and the rest of the world as the degree of real economic contagion after controlling for other important determinants of business cycle co-movement. Then, we examine business cycle comovement during the GFC with financial integration in two specific markets: capital (equity) and credit (debt). With a simultaneous equation model that controls for endogeneity, we find that during the GFC, the business cycle co-movement between the United States and each of the other countries

3

For instance, Ahrend and Goujard (2014) find that different types of financial linkages have different influences on asset price co-movement during the GFC. However, the difference is that our main interest is the international business cycle whereas they focused on asset price co-movement.

4 Page 4 of 46

increased more when the United States and its partner had higher capital market integration, and decreased more when the pair had higher credit market integration. The results are robust even when including investment channels, local fundamental factors, endogenous policy responses across countries, and alternative measures for financial integration and business cycle comovements. Our findings provide significant evidence that correlated equity positions caused by extremely “large-scale” negative financial shock during the GFC did not buffer the transmission of the shock and rather increased business cycle co-movement. By contrast, undiminished demand for safe US debt assets during the GFC did not set off a chain reaction of debt collateral and not amplify the negative shock.4 An important concern about the contagion study is the wake-up call hypothesis based on previous research, which suggests that a country’s fundamentals far outweigh financial linkages in terms of the ability to amplify or buffer against negative shock (e.g., Goldstein, 1998). 5 Our work enriches the wake-up call hypothesis literature by focusing on the effects of financial linkages in the “real” market contagion. In so doing, we show that our results on financial linkages and business cycles are robust when controlling for other important local fundamental factors that affect the transmission-of-the-crisis shock. Our study builds on previous studies examining the effect of financial integration on real business cycle co-movement, which were initiated by Kose, Prasad, and Terrones (2003). Subsequent research, such as Imbs (2004) and Kalemli-Ozcan, Sørensen, and Yosha (2003), has

This result is in line with Prasad’s (2014) argument about the “rush to safety,” which shows investors’ special preference for US dollar-denominated assets. 5 Bekaert, Ehrmann, Fratzscher, and Mehl (2014) confirm the wake-up call hypothesis by showing that the quality of countries’ economic fundamentals and policies play substantial roles in the transmission of negative shocks across financial markets, whereas financial linkages have economically small roles. 4

5 Page 5 of 46

debated the true effects of financial integration on the international real business cycle.6 Davis (2014) reconciles the views of Imbs (2004, 2006) and Kalemli-Ozcan et al. (2003) and finds discrepancies between the effects of capital and credit market integration on the synchronization of the international business cycle. Our study is related to Davis (2014), which distinguishes wealth and balance sheet effects on international business cycle transmission through capital and credit markets. While wealth effects tend to dominate the capital market in international business cycle transmission, balance sheet effects tend to dominate the credit market during normal times. However, distinct from Davis (2014), our results imply that during the crisis, the dominance of the effects is possibly different as the coefficients are reversed. Other works closely related to ours are Kalemli-Ozcan et al. (2013a) and Kalemli-Ozcan, Papaioannou, and Peydro (2013b), which investigate the impact of banking integration on the transmission of international business cycles. In particular, Kalemli-Ozcan et al. (2013a) find that higher linkages in the banking sector were associated with more divergent output cycles before 2007; however, during the GFC more integrated banking between countries led to more synchronized output co-movement.7 Our contribution considers aggregate portfolio investment (equity and debt), including stocks, ADRs, bonds, and treasury bills, whereas they focus solely 6

Kose et al. (2003) show that increased trade and financial linkages have no significant effect on output and consumption correlation. However, Kalemli-Ozcan et al. (2003) show that enhanced financial integration creates a greater chance of industrial specialization and results in more exposure to industry- or country-specific shocks and business cycle de-synchronization. Conversely, Imbs (2004) suggests that increased financial integration generates correlated capital flows, which contributes to business cycle synchronization, although increased financial integration indirectly causes business cycle de-synchronization through industrial specialization. Imbs (2004) finds that the positive effect of financial integration on business cycle synchronization is stronger. Subsequent research by Imbs (2006) sheds light on the link between financial integration, output, and consumption correlation. Note that Dées and Zorell (2012) emphasize the direct effect of trade linkages and production structure on business cycle comovement while they find financial linkages increase business cycle co-movement only indirectly through industry specialization. 7 Kalemli-Ozcan et al. (2013a) also provide a theoretical model that included two kinds of shocks—productivity and financial shock—and documented the propagation mechanism of each shock to the real business cycle according to banking integration.

6 Page 6 of 46

on cross-border bank loans. Our work is also linked to subsequent influential studies by Rose and Spiegel (2010, 2012). Rose and Spiegel (2010) ask a very similar question to the one presented here: did financial integration play a significant role in the transmission of the negative shock during the GFC? They do not find strong evidence that international financial linkages played a significant role in the spread of the GFC shock. In this regard, Dées and Zorell (2012), using a simultaneous equation model similar to this study, find the absence of direct link between financial linkages and business cycle co-movement for cross-country data before the GFC. However, our work dissects “different” types of financial integration and provides significant and different roles of capital and credit market linkages in transmitting negative GFC shocks from the United States. Moreover, our work considers the dynamic extent to which financial integration conveyed the shocks of the GFC to the real business cycle by examining both cross-sectional and time-series variations of a 58-country sample over 2001–2013. The remainder of this paper is organized as follows. In Section 2, we provide a theoretical discussion on the channels by which financial integration affects real business cycle comovement during the GFC. Section 3 describes our empirical model and the data used. Sections 4 and 5 present the empirical results and robustness tests, respectively. Concluding remarks follow in Section 6.

2. Theoretical Discussion on Business Cycle Co-movement and Financial Integration during the GFC We first introduce possible theoretical channels through which negative shocks in the integrated financial markets were transmitted to international real business cycle in normal times, 7 Page 7 of 46

as suggested by previous studies (e.g., Baxter and Crucini, 1995; Kollmann, Enders, and, Muller, 2011). Next, we discuss the difference in the effects of financial integration on business cycle comovement between normal times and during the crisis, in particular, the GFC. Prior studies in the international real business cycle model have suggested that international business cycle transmission via financial integration occurs through wealth and balance sheet effects. First, financial integration leads to a diverged business cycle when the wealth effect dominates. Suppose there are two countries, home and foreign, in which the two financial markets are closely integrated. If there is a negative idiosyncratic shock at home, home consumption and investment decline. In addition, such a shock has a negative effect on foreign investors’ wealth (with a form of home equities). Then, the foreign country decreases consumption and increases savings, which leads to an increase in investment. Thus, the asymmetric patterns in the two countries’ investments through the wealth effect result in business cycle divergence. In particular, Davis (2014) shows that the wealth effect dominates when discussing transmission through capital (equity) markets. In this line, previous studies, such as Kalemli-Ozcan et al. (2003), suggest that in an integrated financial market, home and foreign reallocate their capital, which promotes industrial specialization and business cycle divergence. However, we argue that during the GFC, the abovementioned effects of equity market integration on business cycle co-movement can be reversed. If the negative shock at home is no longer a local shock, but a common shock, and spreads through all industries across countries,8 then capital flight (correlated capital movement) rather than capital reallocation in the integrated 8

Previous literature has suggested that the different kinds of negative shocks in the integrated financial market have different consequences for their spillover effects (Pericoli and Sbracia, 2003). A local or idiosyncratic shock at home can be buffered in the integrated market through risk diversification, and thus, it does not lead to significant contagion. On the contrary, a common shock in the integrated market cannot be cushioned; it would have negative effects on both goods and financial markets across countries.

8 Page 8 of 46

equity market can occur, which would not cushion the transmission of the negative shock effectively. Thus, during the “global” financial crisis, capital investment that decreased simultaneously to and from the United States would lead to business cycle synchronization. Indeed, countries that have high equity market integration with the United States, such as Germany, Japan, Netherlands, and the United Kingdom, experienced sudden decreases in equity integration during the GFC (see Figure 2; a more detailed explanation is in Section 3).9 The balance sheet effect introduces a different channel for international business cycle transmission through financial integration. Again, suppose the two financial markets are closely integrated. If the negative shock occurs at home, this would potentially cause some defaults in home debt assets. Furthermore, the balance sheets of foreign financial intermediaries that hold some of the risky home debts become worse off, and the foreign financial intermediaries respond by paying off the home debts in order to decrease their debt to asset ratios. Thus, both countries face a lack of credit supply at the same time, and their real business cycles are more synchronized. Davis (2014) shows that the balance sheet effect tends to dominate when discussing transmission through credit (debt) markets. However, we suggest that during the GFC, the balance sheet effect would be mitigated in the integrated debt market with the United States because the negative shock in the United States did not cause foreign investors to pay off US debt assets and did not result in a chain reaction. In particular, the status of US government debt assets as a safe-haven investment attracted investors during the GFC. Certainly, prior studies provide convincing evidence on undiminished demand for US debt, which also hinted at the “flight to safety.” 10 Forbes (2010) notes that foreign 9

Graphs showing each individual country’s equity and debt market integration over time are available from the authors upon request. 10 Previous studies regarding “flight to safety” have suggested that the increased market uncertainty during crises induced investors to flee stocks in favor of bonds, which are safer assets (e.g, Baele, Bekaert, Inghelbrecht, and Wei,

9 Page 9 of 46

demand for US Treasury debt, in particular, short-term T-bills, increased sharply during the peak of the crisis.11 Prasad (2014) argues that “the dollar remains the preferred refuge from troubled financial markets, even if its home country is the source of those troubles” (Prasad, 2014, p. 19). We also find that our debt integration measure shows a relatively constant pattern compared to the equity integration measure during this period (see Figure 2). Consequently, high debt market integration with the United States during the GFC did not amplify the transmission of the negative shock, unlike the expectations from the previous theory. Therefore, the roles of equity integration and debt integration in delivering the negative shock from the United States to the rest of the world during the GFC can be differentiated from Davis (2014). Countries with high equity market integration with the United States were more vulnerable to the negative shock during the GFC because of capital flight, whereas countries with high debt market integration with the United States were more insulated from the shock because of increased demand for US debt assets, the so-called “rush to safety.”12

3. Empirical Analysis 3.1. Data and measurements To gauge real economic spillover during the GFC, we compare the real economic growth of the United States—the epicenter of the crisis—with that of other countries. Although it is obvious that the US real growth rate decreased during the crisis and recovered later, the extent to which the rest of the world followed a similar pattern is ambiguous. 2013) 11 However, the focus of Forbes (2010) on foreign holdings of US assets is financial underdevelopment for emerging market countries. 12 This is consistent with Naes, Skjeltorp, and Odegaard (2011)’s finding that US stock market liquidity is countercyclical; they argue that during recessions, including the GFC, there is a “flight to quality” in equities, as investors sell stocks, especially riskier stocks, in favor of safer securities.

10 Page 10 of 46

Our study considers the dynamics of real economic contagion from the United States to the rest of the world by examining business cycle co-movement between the United States and 57 other countries. The measurement for year-by-year business cycle synchronization (SYNCHi,us,t) is the negative divergence defined as the negative absolute value of real GDP per capita growth rate differences between countries i and j in year t:

(1) where gi,t and gus,t indicate the log difference of GDP per capita, based on PPP adjusted constant 2005 international dollars of country i and the United States, which are collected from World Development Indicators (WDI) of the World Bank. This simple index is taken directly from Kalemli-Ozcan et al. (2013a), based on Giannone, Lenza, and Reichlin (2010). It does not reflect the volatility of the output growth of each country in a pair but captures only the co-variation of output growth. Business cycle synchronization with the United States became higher when a country was influenced more negatively by the US economy during the GFC. Thus, SYNCH may capture a degree of the real economic contagion during the GFC if we successfully control for other determinants of real business cycle co-movements. A merit of this measure is that we can utilize time-series information for changes in growth rates, as well as intensity of contagion. Figure 1 reports the real growth rate movements of the United States and the selected partner countries. We confirm that the measure for business cycle co-movement reflects well the changes in the growth rates of both the United States and the other countries. For instance, business cycle comovement between the United States and Canada increased during the GFC, as seen by the decrease in Canada’s growth rate beginning with the decline of US growth in 2008. Moreover, 11 Page 11 of 46

we conduct the same analysis with alternative business cycle co-movement measures, SYNCH1, and SYNCH2, which are discussed in Section 5. [Insert Figure 1] The definition of financial integration varies among academics and policymakers. Taking a narrow view, fully integrated markets imply that all potential market participants face a single set of rules, have equal access, and are treated equally (Baele et al., 2004). Broadly, however, financial integration is a state of close relationships between countries or regions created by lowering financial barriers, cutting transaction costs, or reducing regulations on international capital flows. To measure the degree of bilateral financial integration (bilateral financial linkage), we rely on a quantity-based measure, as in Ahrend and Gourjard (2014), Kalemli-Ozcan et al. (2013a, 2013b), and Rose and Spiegel (2010, 2012), following the broad view of financial integration. Formally, the bilateral capital and credit market integration between countries i and the United States (FIEQi,us,t;FIDBi,us,t) is defined as follows: (2) where AssEQi,us,t (AssDBi,us,t) are assets of equity (debt) securities; a country i (resident)’s crossborder holdings of equity (debt) securities issued by the United States (non-resident) at time t, and LibEQi,us,t (LibDBi,us,t) are liabilities of equity (debt) securities; the US cross-border holdings of securities issued by country i at time t.13 These cross-border holding variables are based on market values of positions held and expressed as current US dollars. GDPi,t and GDPus,t are the current gross domestic product of countries i and the United States, respectively. The sum of 13

It is acknowledged that assets data collected from creditor surveys are more reliable detailed data than liabilities data because the holder (creditor) usually knows what securities it holds while the issuer of a security (debtor) may not know the residency of the holder. In fact, the IMF derives its liabilities data, called “derived liabilities,” using the assets data.

12 Page 12 of 46

GDPi,t and GDPus,t standardizes the sum of assets and liabilities and also relieves valuation effects in the numerator. In addition, we introduce an alternative measure of financial integration using the sum of market capitalization instead of GDP to check the robustness of our results. The alternative capital (credit) market integration measure, FIEQ1i,us,t (FIDB1i,us,t), employs capital (credit) market capitalization of country i at year t from the Financial Development and Structure Dataset (see Beck, Demirgüç-Kunt, and Levine, 2010). Higher values of FIEQi,US,t (FIDBi,US,t) indicate higher capital (credit) market integration between country i and the United States. Bilateral equity and debt security holdings data are collected from the Consolidated Portfolio Investment Survey (CPIS) of the International Monetary Fund (IMF). The CPIS reports bilateral portfolio equity and debt holdings for 74 reporting countries and 231 partner countries. The CPIS data include private portfolio holdings and the holdings of quasi-official entities, such as sovereign funds, but exclude official holdings, such as the foreign reserve holdings of central banks. Note that there is a large body of literature that examines the role of reserve holdings in the international transmission of the crisis (e.g., Aizenman and Hutchison, 2012). Panel A of Figure 2 shows the pattern of average level of equity and debt integration with the United States for both advanced and emerging economies before, during, and after the GFC. Panel B of Figure 2 displays aggregate portfolio investment holdings: US equity/debt investment held by foreign residents (equity/debt assets) and foreign equity/debt investment held by US residents (equity/debt liabilities). As Figure 2 highlights, starkly different patterns emerge between equity integration and debt integration during the GFC. In general, equity market integration with the United States plummeted during the GFC for most countries. However, debt market integration with the United States was relatively constant, or even increased, during the GFC. 13 Page 13 of 46

[Insert Figure 2] TIi,us,t is the measure for bilateral trade integration, calculated as (EXi,us,t + IMi,us,t)/(GDPi,t+GDPus,t). EXi,us,t represents exports from country i to the United States at year t, and IMi,US,t represents imports from the United States to country i at year t. Bilateral export– import data are obtained from Direction of Trade Statistics (DOT). SIMi,us,t is the measure for similarity in the production structure or industry specialization, constructed as SIMi,us,t = .

is sector n’s share in total value added in country i.

is sector n’s

share in total value added in the United States. The sectoral value-added shares are computed using agriculture, manufacturing, and service industry value-added data from WDI.14 Exogenous variables are used to identify the system of equations. Geographical variables such as log distance from the US, and border with the US, and common language (English language) are obtained from Centre d’Études Prospectives et d’Informations Internationales (CEPII). The common legal origin variable with the US (common law) is coded as one if both country i and the US share their legal system origin: English (common), as discussed in La Porta, Lopez-de-Silanes, and Shleifer (2008). The US dollar peg variable is based on the classification used in Shambaugh (2004). The time-series variables are the US dollar peg, capital market restriction indices, regional trade agreement, and bilateral GDP per capita. The US dollar peg is coded as one when a currency either stays within 2% bands against the US dollar or has zero volatility in all months except for a one-off devaluation. Bilateral capital control restrictions (equity and debt) are obtained from Fernández, Klein, Rebucci, Schindler and Uribe (2015) 14

Previous studies such as Imbs (2004, 2006) and Dées and Zorell (2012) compute the production similarity using the United Nations Industrial Development Organization (UNIDO) or UN National Accounts country data (UNNA). However, the coverage of these data are limited to manufacturing industry (UNIDO) or to smaller number of countries (UNIDO and UNNA) and when using production similarity from UNNA, the number of our sample countries shrinks from 58 to 40 countries. Nevertheless, our main results using this production similarity measure from UNNA do not change. Please see column (5) of Table 6.

14 Page 14 of 46

which introduce capital control restrictions on both inflows and outflows of ten categories of assets for 100 countries. The bilateral regional trade agreement variable is coded as one when both countries engage in regional trade agreement, otherwise zero. The data is taken from de Sousa (2012). Before we present our benchmark estimates, we report descriptive statistics in Table 1 for the main endogenous variables employed in the empirical analysis. SYNCH ranges from −12% (most divergent) to −0.002% (most synchronized) and its reported mean is −2.4%. The means of equity integration and debt integration over the sum of GDP are 0.56 and 0.43, respectively, but vary across countries (see Appendix Table A.1). [Insert Table 1] 3.2. Empirical model and identification issues We investigate the impact of negative shock from the United States on the international business cycle and the extent to which it varied according to financial integration. We use country-pair panel data of annual observations from 2001 to 2013. Following previous studies on business cycle co-movement and international linkages such as Imbs (2004, 2006), Dées and Zorell (2012) and Davis (2014), we introduce simultaneous equations model as a main specification. This model not only considers direct and indirect channels of financial integration on business cycle co-movement, but also disentangles the interactions between business cycle co-movement, financial integration, trade integration, and other variables. Our simultaneous equation model consists of five equations as follows:

15 Page 15 of 46

(3)

where SYNCHi,us,t is a measurement for business cycle co-movement between country i and the United States at year t. Cus,t is an indicator for the GFC―a dummy variable coded as one for the GFC period of 2008–2009.15 In our robustness tests, we adopt an additional continuous variable as a proxy for the GFC, using the Chicago Board Options Exchange Market Volatility Index (VIX). FIEQi,us,t (FIDBi,us,t) is the financial integration variable for the capital (credit) market between country i and the United States at year t. Financial integration variables are adjusted by subtracting mean values for ease of interpretation on the interaction terms.16 TIi,us,t is the measure for bilateral trade integration. SIMi,us,t is the measure for similarities in the production structure or industry specialization. Our five endogenous variables are SYNCHi,us,t, FIEQi,us,t, FIDBi,us,t, TIi,us,t, and SIMi,us,t. We include the interaction terms of the crisis and each financial integration measure in the first equation to examine the role of financial integration in determining business cycle co-movement 15

Broadly, the GFC occurred between 2007 and 2009. In November 2007, Lehman Brothers Inc. collapsed, and then, the crisis quickly spread across the world. Within 2009, the US GDP growth rates began to recover, growing close to pre-crisis levels. However, our analysis excludes 2007 and considers only the period from 2008 to 2009 as the GFC period because annual observations may not reflect time-to-time changes quickly. 16 The coefficients in the interaction terms of financial integration and the crisis need to be interpreted by comparing the levels of financial integration and their mean values.

16 Page 16 of 46

especially during the crisis. The interaction of each financial integration and the crisis variable gives two more nonlinear endogenous variables, which makes the number of endogenous variables exceed the number of equations. Wooldridge (2010) states that in the simultaneous equation model nonlinear in endogenous variables, it does not require to include additional equations for the nonlinear endogenous interaction terms. A simple approach of the identification is to add an appropriate set of exogenous variables such as the interaction terms between the crisis variable and other instrumented exogenous variables once the rank condition holds.17 XEi,us,t, XDi,us,t, XTi,us,t, and XSi,us,t are vectors of exogenous variables for each equation.18 As shown in the previous studies such as Davis (2014), Dées and Zorell (2012) and Imbs (2004, 2006), the included exogenous variables help to describe bilateral financial and trade integration and production similarity between country i and the United States. First, a common set of exogenous variables in the equations (FIEQi,us,t, FIDBi,us,t, and TIi,us,t) are physical distance (log distance from the United States) and border with the United States which influence transactions of real goods and financial assets, and English language which captures linguistic and cultural proximity between countries. We also include the time-varying exchange rate peg (US dollar peg) to control for the impact of exchange rate volatility on financial and trade integration with the United States and, furthermore, the transmission of negative shock through the peg during the GFC. To identify the whole system, we include different sets of exogenous variables in each equation. For two financial integration equations, XEi,us,t and XDi,us,t contain common legal origin 17

The rank condition can safely be assumed to hold in a model of this size. Please see the chapter 9 of Wooldridge (2010) for detailed discussion of the identification of simultaneous equation models such as order and rank conditions. Wooldridge (2010) also mentions that technically, the rank condition in nonlinear interaction term models is not necessary for identification if the rank condition of the original system without the interaction term holds. 18 The formal conditions for identification include order and rank conditions. Basically, we need to exclude as many exogenous variables from each equation as we include endogenous variables (Wooldridge, 2010).

17 Page 17 of 46

(common law) because it is likely to lead to similar institutions, regulations, and customs for financial transactions between countries. We add additional time-varying exogenous variables into XEi,us,t and XDi,us,t separately. Fernández et al. (2015) construct capital control measures for various categories of assets. Given their contribution, XEi,us,t contains both the sum and the absolute difference of aggregate capital control restriction indices in equities transaction for countries i and the United States, whereas XDi,us,t includes both the sum and absolute difference of capital control restriction indices in debts (with all maturities) transaction for both countries. For the identification of trade equation, XTi,us,t contains additional time-varying institutional variable, regional trade agreement dummy, which is widely used in gravity literature in trade. Lastly, in the production similarity equation, we add the (log) product of and the absolute value of difference between GDPs per capita of countries into XSi,us,t. Additional exogenous variables for all equations are year dummies. Hence, if the system is well identified by the simultaneous equations model, then we successfully determine the effects of the crisis and its interaction with equity and debt market integration on business cycle correlation through the coefficients βi, for i = 1,..,7. Note that Fries and Kappler (2015) discuss the identification issue of the panel structure simultaneous equation models for the business cycle co-movement thoroughly.19 We mainly focus on country pairs with the United States. Of course, the negative shock from the United States can be transmitted to France via Germany as well as directly to France and there might be inter-linkage transmission channels, such as United States–Germany–France. However, we clarify more the direct impact of the negative shock from the United States on the 19

Fries and Kappler (2015) focus on the first business cycle equation in their system using two stage least squares, rather than reporting the result of the whole system because it is certainly hard to find appropriate exogenous instruments that identify the relationship between trade and FDI.

18 Page 18 of 46

real economies of other countries in order to address investors’ special preference for US assets. In Subsection 4.5, we discuss whether our main findings on business cycle and financial integration between the United States and the rest of the world are generalized. Our main analysis focuses on the effect of the crisis on the international business cycle (i.e., contagion to the real economy), which is the partial derivative with respect to the crisis variable, ∂SYNCHi,us,t/∂Cus,t.. This partial derivative represents how strongly the crisis shock affected business cycle co-movement. Thus, the effect of the crisis on the business cycle is computed as follows, ∂SYNCHi,us,t/∂Cus,t = β1 + β3FIEQi,us,t + β5FIDBi,us,t . However, our main interest is to examine not only the real economic contagion effect but also the extent to which the real contagion depends separately on equity and debt market integration, as indicated by the coefficients of β3= ∂(∂SYNCHi,us,t/∂Cus,t)/∂FIEQi,us,t, and β5=∂(∂SYNCHi,us,t/∂Cus,t)/∂FIDBi,us,t. If β3 is positive (negative), then real economic contagion marginally increases (decreases) with equity market integration. If β5 is positive (negative), then real economic contagion marginally increases (decreases) with debt market integration. For instance, Kalemli-Ozcan et al. (2013a) present a positive coefficient on the interaction term between banking integration and the GFC variable, showing that higher banking integration is associated with more synchronized business cycle during the GFC period.

4. Empirical Results 4.1. Main model with simultaneous equations Table 2 presents the results from the simultaneous equations model in (3) for business cycle co-movement, each proxy of financial integration, trade integration and production similarity, estimated using three-stage least squares (3SLS) analysis. Column (1) reports the 19 Page 19 of 46

results for the first equation in the model, in which bilateral business cycle correlation is the dependent variable. First, the estimated coefficient of equity integration is significantly negative, whereas the coefficient of debt market integration is significantly positive. These two coefficients of financial integration on the international business cycle confirm that different types of bilateral financial integration have the opposite effects on business cycle co-movement, as suggested in Davis (2014). In addition, we show that trade integration has a positive effect on business cycle co-movement and countries with similar in production structure (low values in SIMi,us,t) have more synchronized business cycle. The findings in Table 2 are consistent with those of Davis (2014), validating our system of equations indirectly. [Insert Table 2] Our analytical model of real economic contagion includes the crisis variable and the interaction terms of the crisis with each form of financial integration. The interaction terms with equity and debt market integration are shown to be statistically significant. Specifically, the coefficient of the interaction term with equity market integration is significantly positive (0.04, s.e. = 0.02), corresponding to ∂(∂SYNCHi,us,t/∂Cus,t)/∂FIEQi,us,t, and that of the interaction with debt market integration is significantly negative (−0.04, s.e. = 0.02), corresponding to ∂(∂SYNCHi,us,t/∂Cus,t)/∂FIDBi,us,t. Hence, the positive interaction term with equity integration indicates that a higher level of equity market integration with the United States amplified the effect of the GFC shock originating from the United States on business cycle co-movement. Conversely, the negative coefficient of the interaction term with debt integration means that higher debt market integration with the United States buffered the effect of the GFC on business cycle co-movement. Again, during the GFC, the wealth effect through equity market integration and the balance sheet effect through debt market integration were mitigated, as discussed in 20 Page 20 of 46

Section 2. However, our empirical findings require additional explanation compared to Davis (2014). Although our results for equity and debt market integration on the international business cycle “during the GFC” are opposite to the findings in Davis (2014), the estimated coefficients of both financial integration types on business cycle co-movement over the whole sample period are consistent with those in Davis (2014).

4.2. Including macroeconomic policy responses Macroeconomic policy responds to the internal and external shocks directly and influences the transmission of the shock. For instance, many advanced countries implemented expansionary fiscal and monetary policies in light of the GFC. To control for these policy responses “endogenously” and the effect of policy lags, we include the correlation of macroeconomic policy variables and their lagged terms (predetermined instrument variables) in the system of equations. In this analysis, business cycle co-movement is a function of policy correlation between countries and that the policy correlation is a function of business cycle co-movement, the crisis dummy, peg dummy, and its lagged term. The year-by year fiscal and monetary policy co-movement variables are computed similar to output correlation measure. The fiscal policy variable is calculated as the absolute value of differences in changes in government spending between the United States and individual countries. The monetary policy variable is calculated in a similar way except that it uses changes in M2. In addition, we use real interest rates collected from the WDI as a proxy for monetary policy. The results with real interest rates are also consistent with those with changes in M2. The results with the policy correlations in Table 3 support the robustness of our main results. Two monetary and fiscal policy variables show 21 Page 21 of 46

statistically positive signs on business cycle co-movement. Moreover, the results including policy co-movement variables are robust with alternative identification: We consider policy variables as exogenous ones but the main results are preserved, which are available from the authors upon request. [Insert Table 3] 4.3. Tracing the theoretical channels: Investment correlation Our theory emphasizes the investment channels by which financial integration transmits shock across real economies because equity and debt market integration directly influences changes in investment. Thus, we provide a natural experiment for tracing the theoretical channels from financial markets to real investment, and from the investment to real output. In Table 4, we add investment correlation measure which is constructed similar to output correlation measure into the system. Here, we allow financial integration and the similarity in production structure to affect investment directly while investment and output co-movement are determined endogenously. The results using the measurement for investment co-movement are consistent with the main results in Table 2. In particular, the coefficients on each type of financial integration and the interaction terms with the crisis echo the main results in Table 2 although the interaction term for debt market integration turns out to be slightly insignificant. Note that the results with two alternative SYNCH measures show consistent and significant effects of financial integration on investment correlation as in Table 4. [Insert Table 4] 4.4. Testing the wake-up call hypothesis

22 Page 22 of 46

We test the wake-up call hypothesis that country’s fundamentals have a stronger influence than financial linkages in amplifying or buffering negative shock—that is, macroeconomic conditions of the country play a more important role in contagion than does financial integration (Bekaert et al., 2014). Moreover, it may be questioned whether our results can be driven by confounding factors that cause the synchronization of output. To address these issues, we include country fundamental factors that may influence contagion in the main simultaneous equation model. Domestic macroeconomic fundamental variables are foreign exchange reserves, current account balance, government budget balance, and unemployment rate which are similar to those used by Bekaert et al. (2014) All domestic variables are ratios to GDP except for unemployment rate. In addition to them, we include domestic financial condition variable such as liquid liabilities (currency plus demand and interest-bearing liabilities of banks and non-bank financial intermediaries) to GDP, indicating financial depth. Table 5 shows the results with local fundamental factors of individual countries. The interaction terms with the equity and debt integration variables show significant coefficients, consistent with those in the results in Table 2, even as more local fundamental factors are included from columns (1) to (4). Thus, we confirm the significant and different roles of capital and credit market integration in the transmission of the negative shock during the GFC. Furthermore, we find some evidence that high levels of reserves (in column 2) buffered against the spillover of the negative shock from the United States and induced business cycle divergence during the GFC. In column (4), which includes all domestic controls, the results on local fundamental variables are statistically insignificant while the coefficients on the interaction term of the crisis with financial integration remain significant. [Insert Table 5] 23 Page 23 of 46

4.5. Generalizing the results The main results in Tables 2 to 5 shed light on the spillover of negative shock from the origin of the GFC, the United States, but it may be questioned whether the main results are specific to financial integration defined vis-à-vis the United States. In other words, is it a general observation that higher capital market integration leads to more synchronized business cycle and higher credit market integration results in more divergent business cycle during crises? Generalizing our results requires more thorough investigation, and so, we save this question for future research for the following reasons. It may be suggested that we employ the same empirical framework to investigate the world country pair sample. However, such a sample has a more heterogeneous property than our US– country pair sample in various aspects. Thus, the world country pair sample would require careful control for appropriate heterogeneity in the sample. Previous studies on financial integration and business cycles address country heterogeneity. For instance, Kose et al. (2003) notice different risk-sharing patterns between advanced and developing economies. KalemliOzcan et al. (2013a) focus on only developed countries’ banking integration and business-cycle co-movement before and after the GFC. In fact, we have implemented the same empirical analysis to investigate the world country pair sample that excludes any country pairs with the United States as a reference. However, the results are quite sensitive to the choice of sample country (i.e., world vs. advanced country sample/ world vs. world without tax-haven countries). In addition, given that the negative shock is transmitted from the United States to other countries, our system of equations captures the effects of the crisis emanating from the Unites States on business cycle co-movement through different forms of financial integration. However, when expanding our international real business cycle framework to a world–country pair sample, 24 Page 24 of 46

the identification of the transmission of the shock would be more complex, in particular, during the period of the crisis. Specifically, we would need to identify which country is the source of the negative shock and where the shock comes from.

5. Robustness Tests 5.1. Alternative measures and other tests To check the robustness of the results, we first repeat the main regressions in Table 2 by replacing our crisis dummy for the GFC with the alternative crisis index, the VIX. The VIX is a popular measure of the implied volatility of the S&P 500 index options; high VIX values indicate a high level of expected market volatility. The VIX was recorded at 11.56 on the last day of the year in 2006, 22.50 in 2007, and 40.00 in 2008. Thus, the VIX as a continuous variable shows the progress of the crisis. In the analysis, we use the standardized VIX measures because of different scales compared to other variables. Column (1) of Table 6 shows the results of the main business cycle equation using the VIX and supports our main message in Table 2. [Insert Table 6] We consider alternative measures of output co-movement: First, the alternative measure for business cycle co-movement (SYNCH1) is taken from Morgan, Rime, and Strahan (2004). We regress GDP per capita growth on country fixed effect and year fixed effect as follows: gi,t = αi + αt + νi,t for all countries i. Then the residuals (νi,t and νus,t) represents how much output growth of country i and the United States deviates from average growth over the estimation. We then construct the business cycle synchronization proxy as the negative absolute value of difference of residuals as follows, SYNCH1i,us,t = –| νi,t – νus,t |. This index measures how similarly output growth moves between two countries in any given year. 25 Page 25 of 46

Another output correlation measure, SYNCH2i,us,t is constructed by computing output correlation coeffcients between countries with a five-year rolling window for comparative purpose with previous cross-section studies.20 SYNCH2i,us,t is based on the growth rate of real GDP per capita, adjusted for purchasing power parity. But notice that SYNCH2 i,us,t computed for the years after the GFC inevitably includes the effect of the GFC because it is constructed as a five year backward looking variable. So the co-movement effect of the financial crisis is included in all of the second half of the sample for the relevant endogenous variable, which could bias the results. Columns (2) and (3) of Table 6 report the results with the alternative measures (SYNCH1i,us,t and SYNCH2i,us,t). The results are consistent with the results using SYNCHi,us,t. Since the dependent variable in column (3) is an estimate, we report bootstrapped standard errors to improve the accuracy of statistical inference. Column (4) of Table 6 introduces the alternative financial integration measure using the market capitalization variable. Column (5) reports the result with the alternative production similarity measure (SIM1). 21 We confirm that the effect of the crisis on business cycle comovement between the United States and other countries increased with higher levels of equity market integration between them, whereas the effect of the crisis on business cycle co-movement decreased with higher debt market integration. In addition, our main variables of financial integration are bounded at zero and skewed to the right. To check whether extreme values of financial integration on a long right tail drive the results, we winsorize financial integration measures at the 1% and 99% level and exclude outlier

20

Previous literature employs correlation coefficients to measure bilateral business cycle co-movements (Baxter and Kouparitsas, 2005; Davis, 2014; Dées and Zorell, 2012; Imbs, 2004, 2006). 21 For the alternative SIM (SIM1), gross value added for six broad sectors is obtained from UNNA data. The six sectors based on ISIC revision 4 are in the followings: agriculture, forestry, and fishing (ISIC A); mining, manufacturing, utilities (ISIC B-E); construction (ISIC F); wholesale, retail trade, transportation, and accommodation (ISIC G-I); Information and communication (ISIC J); other activities (ISIC K-R)

26 Page 26 of 46

information in column (6). The results on financial integration are preserved and confirm that the main results are not driven by the extreme values. Previous studies of Dées and Zorell (2012) and Imbs (2004) consider that financial integration is exogenous to trade integration and production similarity in their system. We follow their specifications and examine whether different identification strategy influences our results in column (7). The results support our explanation of the role of each type of financial integration in the transmission of negative shock to the real economy.

5.2. Dissecting financial integration It may be argued that the measurements of each type of financial market integration may not be detailed enough to capture the heterogeneous roles of financial instruments in the transmission of the shock. For instance, holding government bonds is different from holding corporate bonds in terms of the transmission of the shock. However, because of the limitations of the data, we are unable to separate different categories of debts. Another concern is that our financial integration measures are also biased because assets and liabilities of equity and debt securities could be asymmetric (i.e., country i ’s capital market is highly integrated with the US market because US investors own too much of country i’s equities (liabilities) and country i’s investors own relatively few US equities (assets)). To deal with this issue of asymmetry, we split the financial integration measures into assets and liabilities. Table 7 shows the results with separate assets and liabilities integration. We show that the separate integration measures (the shares of domestic assets owned by US investors and of US assets owned by domestic investors) do not display asymmetric effects but work in the same direction for business cycle synchronization during the crisis: both assets and liabilities 27 Page 27 of 46

integration in the capital market increased business cycle co-movement during the crisis, and both types of integration in the credit market decreased it during the GFC. [Insert Table 7] 6. Conclusions We study the role of capital (equity) and credit (debt) integration in the transmission of the negative shock of the GFC from the US financial markets—the epicenter of the GFC—to the rest of the world. Our empirical analysis reveals that a higher level of equity market integration during the GFC amplified the transmission of the US negative shock to many other counties and led to higher business cycle synchronization between the United States and the rest of the world. However, higher debt market integration with the United States during the GFC cushioned the transmission of the negative shock and led to lower business cycle co-movement. We find that the effects of equity and debt market integration on business cycle co-movement during the GFC were opposite to the results of seminal research by Davis (2014). However, when computing the total effects of financial integration in the whole sample period, which includes analysis before and after the GFC, we confirm Davis (2014)’s findings. Our study contributes to the literature by linking the financial market with the real market and examining the spillover from financial market to real market during the crisis period. Future work will extend our analysis to a world country pair sample and other crisis events, as well as identify financial linkages that are more valid as buffers against real economic contagion, thereby providing important guidance.

28 Page 28 of 46

References Ahrend, R., Goujard, A., 2014. Are all forms of financial integration equally risky? Asset price contagion during the global financial crisis. J. Financ. Stab. 14, 35–53. Aizenman, J., Hutchison, M. M., 2012. Exchange market pressure and absorption by international reserves: Emerging markets and fear of reserve loss during the 2008–2009 crisis. J. Int. Money Financ. 31, 1076–1091. Baele, L., Bekaert, G., Inghelbrecht, K., Wei, M., 2013. Flights to Safety. NBER working paper, No. w19095. Baele, L., Ferrando, A., Hördahl, P., Krylova, E., Monnet, C., 2004. Measuring European financial integration. Oxf. Rev. Econ. Policy 20, 509–530. Baig, T., Goldfajn, I., 1999. Financial Market Contagion in the Asian Crisis. IMF Staff Papers, 46. Baxter, M., Crucini, M. J., 1995. Business cycles and the asset structure of foreign trade. Int. Econ. Rev. 36, 821–854. Baxter, M., Kouparitsas, M. A., 2005. Determinants of business cycle comovement: A robust analysis. J. Monet. Econ. 52, 113–157. Beck, T., Demirgüç-Kunt, A., Levine, R., 2010. Financial institutions and markets across countries and over time: The updated financial development and structure database. World Bank Econ. Rev. 24 (1), 77–92. Bekaert, G., Ehrmann, M., Fratzscher, M., Mehl, A. J., 2014. The global crisis and equity market contagion. J. Financ. 69, 2597–2649. Caramazza, F., Ricci, L., Salgado, R., 2004. International financial contagion in currency crises. J. Int. Money Financ. 23, 51–70. CEPII database, http://www.cepii.fr/anglaisgraph/bdd/distances.htm, last accessed on January 30, 2015. Claessens, S., Dornbusch, R. W., Park, Y. C., 2001. Contagion: Why Crises Spread and How this Can Be Stopped. In: Claessens, S., Forbes, K. J. (Eds.), International Financial Contagion. Kluwer Academic Publishers, Norwell, MA. Coordinated Portfolio Investment Survey (CPIS) database, http://cpis.imf.org, IMF, last accessed on January 30, 2015. Davis, J. S., 2014. Financial integration and international business cycle co-movement. J. Monet. Econ., http://dx.doi.org/10.1016/j.jmoneco.2014.01.007. de Sousa, J. 2012. The currency union effect on trade is decreasing over time, Econ. Lett., 117, 917-920. Dées, S., Zorell, N., 2012. Business cycle synchronisation: Disentangling trade and financial linkages. Open Econ. Rev. 23, 623-643. Fernández, A., Klein, M. W., Rebucci, A., Schindler, M., Uribe, M., 2015. Capital Control Measures: A New Dataset. NBER working paper, No. w20970. Financial Development and Structure, World Bank, http://go.worldbank.org/X23UD9QUX0, last accessed on January 30, 2015. Forbes, K. J., 2010. Why do foreigners invest in the United States?. J. Int. Econ., 80, 3-21. Frankel, J. A., Saravelos, G., 2012. Are leading indicators useful for country vulnerability? Evidence from the 2008–09 global financial crisis. J. Int. Econ. 87, 215–231.

29 Page 29 of 46

Fries, C., Kappler, M., 2015. Does foreign direct investment synchronise business cycles? Results from a panel approach. ZEW-Centre for European Economic Research Discussion Paper (15-031). Giannone, D., Lenza, M., Reichlin, L., 2010. Business Cycles in the Euro Area. In: Alesina, A., Giavazzi, F. (Eds.), Europe and the Euro. University of Chicago Press, Chicago, Illinois, pp. 141–167. Goldstein, M. 1998. The Asian Financial Crisis. Peterson Institute for International Economics, Washington DC (No. PB98-1). Imbs, J., 2004. Trade, finance, specialization, and synchronization. Rev. Econ. Stat. 86, 723–734. Imbs, J., 2006. The real effects of financial integration. J. Int. Econ. 68 (2), 296–324. Kalemli-Ozcan, S., Papaioannou, E., Perri, F., 2013a. Global banks and crisis transmission. J. Int. Econ. 89, 495–510. Kalemli-Ozcan, S., Papaioannou E., Peydro, J. L., 2013b. Financial regulation, financial globalization, and the synchronization of economic activity. J. Financ. 68, 1179–1228. Kalemli-Ozcan, S., Sørensen, B. E., Yosha, O., 2003. Risk sharing and industrial specialization: Regional and international evidence. Am. Econ. Rev. 93, 903–918. Kaminsky, G. L., Reinhart, C. M., 2000. On crises, contagion, and confusion. J. Int. Econ. 51, 145–168. Kollmann, R., Enders, Z., Muller, G. J., 2011. Global banking and international business cycles. Eur. Econ. Rev. 55, 407–426. Kose, M. A., Prasad, E. S., Terrones, M. E., 2003. How does globalization affect the synchronization of business cycles? Am. Econ. Rev. 93, 57–62. Lane, P. R., 2013. Financial globalization and the crisis. Open Econ. Rev. 24, 555–580. Lane, P. R., Milesi-Ferretti, G. M., 2007. The external wealth of nations mark II: Revised and extended estimates of foreign assets and liabilities, 1970–2004. J. Int. Econ. 73, 223–250. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., 2008. The economic consequences of legal origins. J. Econ. Lit. 46, 285–332. Morgan, D. P., Rime, B., Strahan, P., 2004. Bank Integration and State Business Cycles. Q. J. Econ., 119, 1555–85. Naes, R., Skjeltorp, J., and Odegaard, B., 2011. Stock market liquidity and the business cycle. J. Financ. 66, 139–176. Pericoli, M., Sbracia, M., 2003. A primer on financial contagion. J. Econ. Surv. 17, 571–608. Prasad, E. S., 2014. The Dollar Trap: How the US Dollar Tightened its Grip on Global Finance. Princeton University Press. Rose, A. K., Spiegel, M. M., 2010. Cross-country causes and consequences of the 2008 crisis: International linkages and American exposure. Pac. Econ. Rev. 15, 340–363. Rose, A., Spiegel, M. M., 2012. The causes and consequences of the 2008 crisis: Early warning. Jpn. World Econ. 24, 1–16. Shambaugh, J. C., 2004. The effect of fixed exchange rates on monetary policy. Q. J. Econ. 119, 301–352. United Nations Statistics, http://data.un.org, last accessed on December 12, 2015 Van Rijckeghem, C., Weder, B., 2003. Spillovers through banking centers: A panel data analysis. J. Int. Money Financ. 22, 483–509. Wooldridge, J.M., 2010. Econometric analysis of cross section and panel data. MIT press. World Development Indicators, World Bank, http://data.worldbank.org/, last accessed on January 30, 2015. 30 Page 30 of 46

Appendix Table A.1 Country average of financial integration and synchronization measures Panel A. Advanced economies

Major currency areas (6) Canada France Germany Italy Japan United Kingdom

Capital markets (FIEQ) Total ASS. LIB.

Credit markets (FIDB) Total ASS. LIB.

SYNCH

3.484 1.727 1.438 0.542 3.172 5.093

1.529 0.392 0.414 0.182 1.112 1.643

1.955 1.336 1.024 0.360 2.060 3.450

2.045 1.302 1.147 0.578 3.614 4.435

0.595 0.819 0.663 0.430 3.353 2.499

1.451 0.483 0.485 0.148 0.261 1.936

-0.008 -0.017 -0.016 -0.008 -0.009 -0.008

0.118 0.263 0.006 0.367 0.067 1.273 0.000 2.188 0.053 0.006 0.494

0.055 0.101 0.001 0.071 0.013 0.808 0.000 1.282 0.018 0.005 0.067

0.063 0.162 0.005 0.295 0.054 0.465 0.000 0.905 0.035 0.001 0.427

0.156 0.241 0.012 0.075 0.035 2.268 0.006 1.315 0.048 0.009 0.314

0.114 0.192 0.010 0.034 0.024 2.010 0.005 0.711 0.040 0.003 0.181

0.042 0.049 0.002 0.041 0.011 0.258 0.001 0.605 0.008 0.005 0.134

-0.010 -0.008 -0.026 -0.019 -0.035 -0.019 -0.022 -0.015 -0.017 -0.029 -0.012

0.648 0.687 0.622

0.102 0.088 0.354

0.545 0.600 0.268

0.387 0.192 0.399

0.368 0.093 0.342

0.018 0.099 0.057

-0.026 -0.025 -0.035

Euro area (11) Austria Belgium Cyprus Finland Greece Ireland Malta Netherlands Portugal Slovenia Spain Newly industrialized Asian economies (3) Hong Kong SAR, China Korea, Rep. Singapore Other advanced economies (8)

31 Page 31 of 46

Australia Czech Republic Denmark Iceland New Zealand Norway Sweden Switzerland Avg. of advanced economies (28)

1.241 0.023 0.377 0.010 0.078 0.551 0.739 1.996 0.974

0.608 0.004 0.210 0.009 0.053 0.405 0.421 0.443 0.371

0.634 0.019 0.167 0.000 0.025 0.147 0.319 1.552 0.603

0.933 0.009 0.215 0.019 0.082 0.465 0.445 0.494 0.759

0.284 0.007 0.145 0.004 0.022 0.319 0.175 0.448 0.496

0.649 0.002 0.070 0.015 0.060 0.147 0.269 0.046 0.263

-0.011 -0.025 -0.010 -0.021 -0.016 -0.008 -0.015 -0.011 -0.017

32 Page 32 of 46

Panel B. Emerging and developing economies

Commonwealth of independent states (2) Kazakhstan Russian Federation Developing Asia (7) China India Indonesia Malaysia Pakistan Philippines Thailand Latin America and the Caribbean (10) Argentina Bolivia Brazil Chile Colombia Costa Rica Mexico Panama Uruguay Venezuela, RB Middle East and North Africa (3) Egypt, Arab Rep. Kuwait Lebanon Sub-Saharan Africa (2) Mauritius South Africa Central and Eastern Europe (5)

Capital markets (FIEQ) Total ASS. LIB.

Total

Credit markets (FIDB) ASS. LIB.

0.015 0.214

0.012 0.001

0.003 0.213

0.096 0.075

0.088 0.019

0.008 0.056

-0.051 -0.044

0.320 0.359 0.103 0.109 0.005 0.043 0.096

0.012 0.000 0.001 0.025 0.000 0.001 0.005

0.308 0.359 0.101 0.085 0.005 0.042 0.090

0.031 0.015 0.042 0.057 0.001 0.051 0.019

0.023 0.000 0.003 0.006 0.000 0.016 0.007

0.008 0.015 0.040 0.051 0.001 0.035 0.011

-0.081 -0.049 -0.034 -0.023 -0.011 -0.023 -0.028

0.062 0.000 0.640 0.158 0.029 0.001 0.414 0.125 0.001 0.004

0.053 0.000 0.013 0.115 0.016 0.001 0.010 0.001 0.001 0.001

0.009 0.000 0.628 0.043 0.013 0.000 0.404 0.124 0.000 0.003

0.086 0.002 0.196 0.118 0.074 0.004 0.357 0.048 0.017 0.076

0.046 0.001 0.018 0.056 0.030 0.002 0.117 0.018 0.006 0.036

0.041 0.000 0.178 0.062 0.044 0.002 0.240 0.030 0.011 0.040

-0.056 -0.021 -0.019 -0.023 -0.020 -0.021 -0.015 -0.046 -0.047 -0.042

0.019 0.010 0.005

0.000 0.010 0.004

0.019 0.000 0.001

0.019 0.005 0.006

0.009 0.005 0.004

0.010 0.000 0.002

-0.021 -0.015 -0.028

0.021 0.342

0.009 0.073

0.012 0.268

0.003 0.051

0.003 0.014

0.001 0.037

-0.027 -0.014

SYNCH

33 Page 33 of 46

Bulgaria Hungary Latvia Poland Turkey Avg. of emerging economies (29)

0.001 0.032 0.000 0.039 0.094 0.112

0.000 0.005 0.000 0.002 0.000 0.013

0.001 0.027 0.000 0.036 0.093 0.100

0.006 0.023 0.004 0.053 0.046 0.055

0.002 0.002 0.002 0.007 0.002 0.019

0.005 0.021 0.002 0.046 0.044 0.036

-0.038 -0.020 -0.042 -0.030 -0.035 -0.032

Note: Country classification follows the IMF World Economic Outlook (2013). Equity-market and debt-market columns report the average of equity and debt market integration measures, respectively, from 2001 to 2013. Total, ASS., and LIB. indicate the sum of assets and liabilities, of assets, and of liabilities, respectively. These measures (Total, ASS., and LIB.) are (total, assets, and liabilities) holdings between the United States and individual country divided by the sum of the current GDPs of both countries (percentage term).

34 Page 34 of 46

Figure 1: Trend of real GDP growth rates and business cycle synchronization Panel A. Major countries

35 Page 35 of 46

Panel B. Asian countries

Note: GDPG and US GDPG indicate real GDP growth rates of each country and the United States, respectively. We select the major (G6) and several Asian countries for space constraints; trends for other countries are available upon request.

36 Page 36 of 46

Figure 2: Trend of bilateral financial integration with the United States

Panel A. Levels of financial integration with the United States Advanced economies 1.8 1.5 1.2 0.9 0.6 0.3 0 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 FIEQ

FIDB

Emerging and developing economies 0.5

0.4

0.3

0.2

0.1

0 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 FIEQ

FIDB

Note: FIEQ and FIDB are the equal-weighted average of the equity and debt market integration measures across countries. Advanced economies include 28 countries while emerging and developing economies include 29 countries from our sample.

37 Page 37 of 46

Panel B. Aggregate portfolio investment holdings with the United States

(current USD millions)

Partner: Advanced economies

12,000,000 10,000,000 8,000,000 6,000,000 4,000,000 2,000,000 0 2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

US equity investment held by foreign residents

Foreign equity investment held by US residents

US debt investment held by foreign residents

Foreign debt investment held by US residents

(current USD millions)

2013

Partner: Emerging economies

1,400,000 1,200,000 1,000,000 800,000 600,000 400,000 200,000 0 2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

US equity investment held by foreign residents

Foreign equity investment held by US residents

US debt investment held by foreign residents

Foreign debt investment held by US residents

2013

Note: Portfolio investment (equity and debt) holdings are aggregated across countries. Advanced economies include 28 countries and emerging and developing economies include 29 countries from our sample.

38 Page 38 of 46

Table 1 Descriptive statistics. N

Mean

S.D.

Min

Max

p1

p5

p25

p50

p75

p95

p99

SYNCH

646

-0.024

0.021

-0.123 -0.00002 -0.091 -0.067 -0.036 -0.018 -0.008 -0.001 -0.0003

FIEQ

646

0.557

0.996

0.000

7.310

0.000

0.001

0.020

0.141

0.596

2.519

5.166

FIDB

646

0.425

0.895

0.000

5.514

0.001

0.002

0.020

0.074

0.342

2.664

4.600

TI

646

0.255

0.473

0.001

3.629

0.001

0.003

0.025

0.094

0.279

1.114

2.646

SIM

646

0.270

0.188

0.008

0.970

0.014

0.044

0.127

0.227

0.358

0.675

0.741

Note: SYNCH is a proxy for real business cycle co-movement, defined as absolute value of real GDP growth rate differences between country i and the United States in year t. FIEQ and FIDB are the capital and credit market integration measures, respectively, calculated as the sum of assets and liabilities divided by the sum of the current GDP of the United States and partner country. TI is the trade integration measure, the sum of exports and imports divided by the sum of the current GDP of the United States and partner country. These are quantity-based measures. SIM is a measure for the similarity in the production structure or industry specialization. Mean, S.D., Min, and Max are the summary statistics of country-year observations. FIEQ, FIDB, and TI are multiplied by 100 and represented as a percentage term for simple presentation.

39 Page 39 of 46

Table 2 Main results. (1) (2) SYNCH FIEQ 0.042** (0.020) FIDB x CRISIS -0.041** -0.113** (0.020) (0.053) FIEQ -0.048*** (0.010) FIDB 0.053*** 1.114*** (0.012) (0.056) TI 0.009*** 0.284* (0.003) (0.169) SIM -0.072*** -0.216 (0.010) (0.185) CRISIS -0.010*** -0.105** (0.003) (0.045) Log distance from the US -0.002 (0.053) Border with the US -0.426 (0.449) English language -0.047 (0.053) Common law -0.02 (0.052) US dollar peg 0.049 (0.054) Sum of bilateral capital control restrictions (equity) 0.019 (0.021) Abs. diff. of bilateral capital control restrictions (equity) -0.014 (0.027) Sum of bilateral capital control restrictions (debt) Dependent variables FIEQ x CRISIS

Abs. diff. of bilateral capital control restrictions (debt)

(3) FIDB 0.101** (0.046)

(4) TI

(5) SIM

0.780*** (0.041)

0.155* (0.083) 0.156 (0.096)

0.028 (0.054) -0.121** (0.062) 0.049*** (0.017)

0.153 (0.143) -0.051 (0.171) 0.088** (0.040) -0.068 (0.046) -0.629* (0.376) 0.114** (0.046) 0.011 (0.045) -0.090* (0.048)

0.631*** (0.116) 0.018 (0.028) 0.154*** (0.029) 2.469*** (0.084) -0.171*** (0.030)

0.013 (0.017)

0.110*** (0.031)

-0.027 (0.020) 0.010 (0.020)

Regional trade agreement

0.006 (0.023)

Log product of bilateral GDPs per capita Abs. diff. of bilateral GDPs per capita Observations

646

646

646

646

-0.106*** (0.013) -0.000002* (0.000001) 646

Note: Standard errors are reported in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Financial integration variables are adjusted by subtracting mean values for the interaction terms. Additional exogenous variables are the year dummy variables and the interaction terms between global financial crisis dummy and timevarying exogenous variables (US dollar peg, sum of equity/debt market restrictions, absolute difference in equity/debt market restriction, log product of bilateral GDPs per capita, and absolute difference in bilateral GDPs per capita)

40 Page 40 of 46

Table 3 Including macroeconomic policy responses. Dependent variables FIEQ x CRISIS FIDB x CRISIS FIEQ FIDB TI SIM CRISIS Monetary policy Fiscal policy

(1) SYNCH 0.052** (0.021) -0.041** (0.020) -0.029*** (0.008) 0.032*** (0.009) 0.002 (0.003) -0.039*** (0.010) -0.011*** (0.003) 0.124*** (0.025) 0.289*** (0.065)

(2) FIEQ

(3) FIDB 0.149** (0.059)

(4) TI

(5) SIM

0.849*** (0.040)

0.192** (0.083) 0.081 (0.089)

-0.034 (0.044) -0.046 (0.045) 0.067*** (0.016)

(6) Monetary

(7) Fiscal

0.005 (0.007)

0.004 (0.003)

-0.140** (0.059)

1.062*** (0.049) 0.197 0.154 (0.172) (0.154) -0.556*** 0.335* (0.171) (0.175) -0.103** 0.094** (0.045) (0.042)

0.569*** (0.116) 0.031 (0.029)

-0.002 (0.016)

0.322*** (0.030)

Lag of monetary policy Lag of fiscal policy Observations

569

569

569

569

569

569

0.317*** (0.039) 569

Note: See Table 2. Table 4 introduces the fiscal and monetary policy variables. The fiscal policy variable is calculated as the negative absolute value of government final consumption expenditure growth rate differences between countries i and the United States in year t. The monetary policy variable is calculated in a similar way except that it uses average annual M2 growth rates. For simple presentation, this table does not report the coefficients of the constant and exogenous variables in the main model, which are available from the authors upon request.

41 Page 41 of 46

Table 4 Investment correlations. Dependent variables FIEQ x CRISIS

(1) SYNCH

FIDB x CRISIS FIEQ FIDB TI

CRISIS Observations

0.392*** (0.024) 0.008*** (0.002) 636

(3) FIEQ

(4) FIDB

(5) TI

(6) SIM

0.756*** (0.045)

0.150** (0.076) 0.131 (0.088)

0.054 (0.050) -0.129** (0.056) 0.037** (0.016)

-0.078*** (0.025)

1.121*** (0.063) 0.210 (0.184) -0.123 (0.196)

0.320** (0.150) -0.147 (0.177)

0.475*** (0.113)

-0.046*** (0.008) 636

-0.114** (0.046) 636

0.087** (0.041) 636

0.015 (0.028) 636

-0.0003 (0.002)

SIM INVESTMENT

(2) INV. 0.090* (0.052) -0.074 (0.054) -0.080*** (0.026) 0.107*** (0.031)

0.016 (0.016) 636

Note: See Table 2. Investment is the negative absolute value of real investment growth rate differences between countries i and the United States in year t. For simple presentation, this table only reports the coefficients of endogenous variables and excludes those of the constant and exogenous variables in the main model, which are available from the authors upon request.

42 Page 42 of 46

Table 5 Wake-up call hypothesis. Dependent variables FIEQ x CRISIS FIDB x CRISIS FIEQ FIDB TI SIM CRISIS Current account Current account x CRISIS Unemployment Unemployment x CRISIS

(1) SYNCH 0.043** (0.019) -0.037** (0.019) -0.032*** (0.010) 0.037*** (0.011) 0.003 (0.003) -0.055*** (0.010) -0.008*** (0.003) 0.066*** (0.022) -0.038 (0.032) -0.046* (0.023) 0.020 (0.070)

(2) SYNCH 0.038** (0.018) -0.036** (0.018) -0.033*** (0.010) 0.039*** (0.010) 0.003 (0.003) -0.051*** (0.009) -0.009*** (0.003) 0.070*** (0.021) -0.015 (0.033) -0.054** (0.024) -0.017 (0.071) -0.008 (0.006) -0.028* (0.015)

(3) SYNCH 0.035** (0.017) -0.031* (0.017) -0.051*** (0.011) 0.056*** (0.011) 0.005 (0.003) -0.048*** (0.010) -0.010*** (0.003) 0.093*** (0.026) -0.067 (0.041) -0.042 (0.026) 0.014 (0.074) -0.015 (0.009) 0.001 (0.026) 0.000 (0.003) 0.001 (0.009)

566

566

503

Reserves Reserves x CRISIS Liquid liabilities Liquid liabilities x CRISIS Fiscal balance Fiscal balance x CRISIS Observations

(4) SYNCH 0.032** (0.015) -0.027** (0.013) -0.018* (0.010) 0.020** (0.009) -0.001 (0.007) -0.036*** (0.009) -0.006** (0.003) 0.064*** (0.023) -0.054 (0.038) -0.056** (0.024) -0.020 (0.068) -0.028*** (0.008) 0.008 (0.021) 0.001 (0.003) 0.006 (0.008) 0.015 (0.027) -0.068 (0.066) 412

Note: See Table 2. Current account, unemployment, reserves, liquidity liabilities, and fiscal balance variables are country i’s fundamental variables. In the analysis, these variables are adjusted by subtracting mean values for the interaction terms. For simple presentation, this table only reports the first equation and excludes the coefficients on the constant and exogenous variables in the main model, which are available from the authors upon request.

43 Page 43 of 46

Table 6 Alternative measures and other tests.

Robustness tests Dependent variables FIEQ x CRISIS FIDB x CRISIS FIEQ FIDB TI SIM CRISIS Observations

(1)

(2)

(3)

(4)

Alternative crisis variable (VIX)

Alternative SYNCH (SYNCH1)

Alternative SYNCH (SYNCH2)

SYNCH 0.023** (0.011) -0.036*** (0.013) -0.085*** (0.013) 0.096*** (0.015) 0.021*** (0.005) -0.085*** (0.012) -0.008*** (0.002) 646

SYNCH1 0.043** (0.019) -0.039** (0.019) -0.027*** (0.010) 0.036*** (0.011) 0.003 (0.003) -0.016* (0.009) -0.007*** (0.002) 646

SYNCH2 1.973*** (0.734) -1.976*** (0.762) -0.894 (0.774) 1.068 (0.883) 0.146 (0.130) -0.658 (0.417) 0.190* (0.103) 644

(5)

Alternative Alternative SIM FIEQ/FIDB (SIM1) (FIEQ1/FIDB1) SYNCH 0.046*** (0.017) -0.034** (0.016) -0.092*** (0.013) 0.067*** (0.011) 0.017*** (0.003) -0.115*** (0.011) -0.009** (0.004) 403

SYNCH 0.023* (0.014) -0.020 (0.014) -0.015* (0.008) 0.010 (0.008) 0.010*** (0.002) -0.107*** (0.015) -0.006** (0.003) 421

(6)

(7)

Winsorizing FIEQ/FIDB at 1% and 99% level

FIEQ/FIDB depend on only exogenous variables

SYNCH 0.042** (0.020) -0.040* (0.021) -0.052*** (0.011) 0.057*** (0.012) 0.010*** (0.003) -0.073*** (0.010) -0.010*** (0.003) 646

SYNCH 0.076*** (0.019) -0.072*** (0.019) -0.050*** (0.010) 0.058*** (0.012) 0.008*** (0.003) -0.065*** (0.010) -0.009*** (0.003) 646

Note: See Table 2. Column (4) reports bootstrapped standard errors because alternative SYNCH2 is an estimated measure. Column (1) uses the standardized VIX to indicate the GFC. Column (2) uses the synchronization measure with SYNCH1, which is the negative absolute value of the residual GDP growth rate differences and column (3) employs other alternative measure, SYNCH2, which is the correlation coefficients between country i and the United States with a fiveyear rolling window. Column (4) employs alternative FIEQ/FIDB that are divided by market capitalization instead of GDP. Column (5) employs alternative SIM (the similarity in production structures) using more detailed industry categories from UN data. Column (6) winsorizes FIEQ/FIDB at the 1% and 99%. Column (7) introduces different system that FIEQ/FIDB only depend on exogenous variables following Dees and Zorell (2012). For simple presentation, this table only reports the first equation and excludes the coefficients of the constant and exogenous variables in the main model, which are available from the authors upon request.

44 Page 44 of 46

Table 7 Dissecting financial integration. Dependent variables FIEQASS x CRISIS FIEQLIB x CRISIS FIDBASS x CRISIS FIDBLIB x CRISIS FIEQASS FIEQLIB FIDBASS FIDBLIB TI SIM CRISIS

(1) SYNCH

(2) FIEQASS

0.130** (0.061) 0.02 (0.032) -0.023 -0.037 (0.021) (0.051) -0.145*** -0.067 (0.051) (0.099) -0.02 (0.026) -0.064*** (0.013) 0.043*** 0.115*** (0.014) (0.038) 0.069*** 0.981*** (0.022) (0.078) 0.010*** 0.311*** (0.003) (0.070) -0.071*** -0.371*** (0.011) (0.088) -0.007** -0.03 (0.003) (0.019)

(3) FIEQLIB

(4) FIDBASS

(5) FIDBLIB

0.217 (0.365) 0.035 (0.268)

0.052 (0.188) 0.047 (0.140)

0.087 (0.189) 0.913*** (0.108)

0.927*** (0.088) -0.076 (0.053)

-0.02 (0.173) -0.085 (0.201) 0.085** (0.043)

-0.242*** (0.082) 0.381*** (0.101) 0.023 (0.020)

(6) TI

(7) SIM

0.815*** (0.281) 0.197 (0.122) 0.016 (0.132) -0.648*** (0.221)

0.076 (0.149) -0.159** (0.072) 0.129* (0.068) -0.214** (0.107) 0.089*** (0.019)

-0.059 (0.110) -0.214 (0.219)

0.957*** (0.098) 0.063 (0.180) 0.086 (0.163) -0.036 (0.170) -0.074* (0.039)

0.734*** (0.137) 0.041 (0.029)

-0.002 (0.017)

Observations 645 645 645 645 645 645 645 Note: See Table 2. FIEQ assets (FIEQASS), FIEQ liabilities (FIEQLIB), FIDB assets (FIDBASS), and FIDB liabilities (FIDBLIB) are the financial integration measures for equity assets, equity liabilities, debt assets, and debt liabilities from/to the United States. In the analysis, these variables are adjusted by subtracting mean values for the interaction terms. For simple presentation, this table does not report the coefficients of the constant and exogenous variables in the main model, which are available from the authors upon request.

45 Page 45 of 46

46 Page 46 of 46