Foreign direct investment and business cycle co-movements: The panel data evidence

Journal of Macroeconomics 33 (2011) 770–783

Contents lists available at SciVerse ScienceDirect

Journal of Macroeconomics journal homepage: www.elsevier.com/locate/jmacro

Foreign direct investment and business cycle co-movements: The panel data evidence Chih-Chiang Hsu, Jyun-Yi Wu, Ruey Yau ⇑ Department of Economics, National Central University, Taoyuan 32001, Taiwan

a r t i c l e

i n f o

Article history: Received 21 October 2010 Accepted 6 June 2011 Available online 17 August 2011 JEL classification: E32 F10 F21 Keywords: Business cycle co-movements Foreign direct investment Trade Industrial dissimilarity

a b s t r a c t The previous literature has largely overlooked the possible channels through which foreign direct investment (FDI) might influence business cycle synchronization. In this study we analyze the linkages that exist among FDI, trade and industrial dissimilarity in relation to business cycle co-movements using a panel data set taken from 77 pairs of developed countries. The error component three-stage least squares (EC3SLS) estimates from a simultaneous equations model with panel data are shown to be superior to the estimates obtained from single equation models or simultaneous equations models with cross-sectional data. Our results indicate that FDI serves as a channel of international business cycle transmission that is equally important as the channels of trade and monetary policy. On the contrary, industrial dissimilarity is identified as having an indirect impact on the business cycle correlation through trade and FDI. Furthermore, our findings suggest that in our sample FDI is of the horizontal type and tends to substitute for trade. Ó 2011 Elsevier Inc. All rights reserved.

1. Introduction The channels of international business cycle co-movements have been debated in the literature, with this debate focusing on how such co-movements have been propagated and transmitted from one country to another. There are several possible channels, the most prominent channel being trade. Frankel and Rose (1998) state that countries with closer trade links tend to have more tightly correlated business cycles. Baxter and Kouparitsas (2005) also arrive at similar conclusions whereby intense bilateral trade tends to result in a high degree of synchronization among business cycles. By contrast, Gruben et al. (2002) and Inklaar et al. (2008) find that the trade effect is smaller than previously reported. Crosby (2003) even states that trade does not explain the correlation. Another transmission channel is dissimilarity in industrial structures. Imbs (2004) points out that industrial dissimilarity (or specialization) patterns have sizable effects on business cycle co-movements. However, Otto et al. (2001) and Baxter and Kouparitsas (2005) do not confirm this result. More recently, the role of financial integration in business cycle synchronization has been stressed by Imbs (2004, 2006). He finds that economic regions with strong financial linkages are more synchronized, an argument that is however not supported by Inklaar et al. (2008). Other possible channels can be summarized as follows: (i) monetary integration (Schiavo, 2008); (ii) economic integration (Kalemli-Ozcan et al., 2001); (iii) similarity of fiscal policies (Clark and van Wincoop, 2001); (iv) exchange rate volatility (Inklaar et al., 2008); and others: see de Haan et al. (2008) for a recent survey. In this paper we focus on the role of foreign direct investment (FDI) in technology diffusion and financial investment and argue that FDI might be another important channel for the international transmission of disturbances. FDI is a category of ⇑ Corresponding author. Tel.: +886 03 4227151; fax: +886 03 4222876. E-mail address: [email protected] (R. Yau). 0164-0704/$ - see front matter Ó 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.jmacro.2011.06.001

771

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783

1980

1990

2000

8 4 0

Canada

France Germany

Italy

Japan

UK

1980

20

FDI/GDP (%)

FDI/GDP (%)

12

USA

World

1990

2000

16 12 8 4 0

Canada

France Germany

Italy

Japan

UK

USA

World

Fig. 1. FDI flows to GDP ratio.

cross-border investment made by a resident in one economy (source economy) to acquire a lasting interest in an enterprise operating in another economy (host economy).1 FDI has increased dramatically since the 1980s. Fig. 1 shows the FDI inflows and outflows as a share of GDP among the G7 countries (Canada, France, Germany, Italy, Japan, the UK and the US) and the world in different years.2 For the world, we find that inward FDI as a share of GDP increased from 0.50% in 1980 to 4.39% in 2000. The share of GDP accounted for by outward FDI for the world increased from 0.51% in 1980 to 3.53% in 2000. For the G7 countries, both inward and outward of FDI have grown more than five times over these two decades. Although Japan’s inward and outward FDI were very low in 2000, being only 0.17% and 0.67%, respectively, its FDI is in a rising trend and could play an important role in cross-border business cycle co-movements. In contrast to FDI, exports and imports as a share of GDP are more stable, but trade (imports+exports) still accounts for a large share of GDP. This is shown in Fig. 2. Apparently, the role of FDI has become increasingly important during this period. However, only a few papers indicate that the business cycles are more highly correlated for those countries which are more involved in bilateral FDI. Jansen and Stockman (2004) use aggregated data on bilateral FDI among OECD countries and suggest that countries with tighter FDI linkages also have more correlated business cycles. Similar results are found by Otto et al. (2001), but they state that the effect of FDI is smaller than that of bilateral trade. Levy Yeyati et al. (2007) find FDI flows to be countercyclical with the business cycle of the source country. How does bilateral FDI contribute to the business cycle synchronization? Otto et al. (2001) and Jansen and Stockman (2004) suggest some possible channels. We summarize them below. First, if foreign firms introduce new processes into the domestic market, then domestic firms may benefit from the accelerated diffusion of new technology. Second, if a deterioration in the economic conditions in the foreign investor’s home country weakens the financial health of the parent companies, inward FDI may lead to a cutback in hiring and a decrease in the wage and investment in the host countries. Hence, international rent sharing within multinational companies may cause the spread of local macroeconomic shocks from one country to another. Third, in an outward FDI position, unfavorable disturbances in the host foreign countries may reduce the net worth of the domestic investing firms, which may further hurt domestic investment via the balance sheet channel and the stock market channel. It is also likely to induce an adverse impact on domestic consumption via the wealth effect. Fourth, if capital is relatively mobile between two countries, then a change in the saving and investment decision in one country is likely to affect the price and availability of financial assets. This will lead to more closely synchronized business cycles. These outlined channels of business cycle transmission through FDI are related to the notions of technology spillover, activities of multinationals, and financial integration in a more general sense. In this study, we argue that there has been a trend in the past two decades whereby firms have substituted FDI for trades in their decision to serve foreign markets.3 In the existing literature, the interaction between trades and specialization patterns and their linkages with business cycle synchronization have been investigated empirically.4 Given the fact that FDI’s role is increasing, an interesting question that remains to be explored is whether the shift in trades and FDI composition will change the interaction patterns. The purpose of this paper is to investigate the impact of bilateral FDI, trade and industrial dissimilarity on business cycle correlation in a simultaneous equations system. Our empirical methodology is similar in spirit to that introduced in Imbs (2004), but our approach diverges from his in two respects. First, unlike most empirical studies on synchronization that use cross-sectional data, this paper estimates a simultaneous equations system using a panel data set of 77 country pairs. Estimating the model with panel data allows us to take into account how FDI development over time for a given country pair may have affected synchronization and other endogenous variables in the system. Any unobserved country pair-specific effect can be controlled for in a panel estimation, while it could lead to biased and inconsistent estimates in a pure cross-sectional regression. The second major difference between this paper and Imbs (2004) is the inclusion of the FDI variable in our empirical model. FDI cannot only transmit technological shocks but also link itself to the international financial market through 1

See OECD (2008). Data are obtained from the UNCTAD (2008) World Investment Report. According to Helpman et al. (2004), larger transport costs and small plant-level returns to scale are potential explanations for such a trend. This is because exporting involves lower fixed costs while FDI entails lower variable costs. 4 For instance, Frankel and Rose (1998) and Otto et al. (2001) use the single equation method to estimate the relationship, whereas Imbs (2004), de Haan et al. (2008) and Schiavo (2008) adopt the simultaneous equations framework. 2 3

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783

1980

40

1990

2000

30 20 10 0

Canada France Germany Italy

Japan

UK

USA

1980

50

Export/GDP (%)

Import/GDP (%)

772

World

1990

2000

40 30 20 10 0

Canada

France Germany

Italy

Japan

UK

USA

World

Fig. 2. Import (export) to GDP ratio.

the reallocation of capital. The notion of inducing capital flows is akin to the financial integration variable first considered in Imbs (2004) and also in Inklaar et al. (2008). In this paper, we choose to focus on the channel through FDI rather than through financial integration. Our model specification is not designed to compete with the novel concept introduced by Imbs (2004), but to offer new evidence on whether this theoretically plausible channel of FDI is desirable. We apply both of the single equation and simultaneous equations estimations using the panel data. The single equation estimation involves both the fixed effects (FE) and random effects (RE) estimations. However, since the single equation procedure might ignore the endogeneity problem and the indirect effects of each variable, we also employ the estimation method which involves the use of the error component three-stage least squares (EC3SLS) approach proposed by Baltagi (1981). An advantage of the EC3SLS approach is that it can tackle the problem of endogeneity and has the characteristics of simultaneous equations. Therefore, EC3SLS is more reliable than either the FE or RE estimations. In the empirical section, the estimates from the single equation approach are demonstrated to be biased because a number of key coefficients are insignificant. On the contrary, the EC3SLS estimation yields the most favorable results.5 The panel data approach is also shown to be superior to the cross-sectional data one for that the 3SLS estimates from the latter approach give fewer significant coefficients. The main contribution of our empirical analysis is that we introduce FDI into a simultaneous equations system to study for business cycle correlation. Our EC3SLS estimation results suggest that trade and FDI have positive effects on the business cycle co-movements, separately. The similarity in monetary policies is another important variable that can explain business cycle synchronization. To our surprise, in contrast to the finding in Imbs (2004, 2006), the coefficient for the industrial dissimilarity is statistically insignificant, although with the expected sign. The positive influence of industrial dissimilarity (or production specialization) on output correlation is however identified through indirect linkages via trade and FDI. Note that such an inference cannot be drawn simply with the single equation estimation method. This is another merit by adopting the simultaneous equations approach. More specifically, in our sample, trade occurs more intensively between countries that obey the principle of comparative advantage by concentrating on different industries, whereas FDI flows are more active between countries with similar sectoral structures. In addition, we find that FDI is of the horizontal type among these developed countries and that FDI activities tend to substitute for trade activities. The remainder of this paper is organized as follows. The next section describes the model specification and econometric methodology employed. Section 3 reports the empirical results, while the final section provides the main conclusions. 2. The econometric model 2.1. Model specification To investigate the relationships that exist among foreign direct investment (F), trade intensity (T), dissimilarity in the industrial structure (DS) and business cycle correlation (q), this paper estimates the following system of equations:

qi;j;t ¼ a0 þ a1 F i;j;t þ a2 T i;j;t þ a3 DSi;j;t þ a4 Z 1;i;j;t þ e1;i;j;t ; F i;j;t ¼ b0 þ b1 T i;j;t þ b2 DSi;j;t þ b3 Z 2;i;j;t þ e2;i;j;t ; T i;j;t ¼ c0 þ c1 F i;j;t þ c2 DSi;j;t þ c3 Z 3;i;j;t þ e3;i;j;t ; DSi;j;t ¼ d0 þ d1 F i;j;t þ d2 T i;j;t þ d3 Z 4;i;j;t þ e4;i;j;t ;

ð1Þ ð2Þ ð3Þ ð4Þ

where i, j, and t are index country pairs (i, j) in period t, and e is the disturbance term. Vectors Z1, Z2, Z3, and Z4 contain exogenous variables that are employed in the system to achieve identification. In a panel framework, it is practical to allow for individual heterogeneity. We specify the disturbance term as the sum of a time-invariant pair-specific term (l) and an idiosyncratic random error (e):

ek;i;j;t ¼ lk;i;j þ ek;i;j;t ; for k ¼ 1; 2; 3; 4;

5

Imbs (2004, 2006) and Schiavo (2008) also adopt the simultaneous equations approach and stress its advantages over the single equation approach.

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783

773

where l1,i,j, l2,i,j, l3,i,j, and l4,i,j enter the model to capture the individual effects that are specific to country pair (i, j) in the four equations, respectively. For these pair-specific terms, we consider both the fixed effects (where the differences across units are constant) and the random effects (where the terms are randomly drawn from a large population) models. The Hausman specification test is adopted later to detect appropriate model specification.6 We first explain the measures of the key variables in this system. A detailed description of the variables and sources of data is included in the Appendix. In Eq. (1), two different measures of output are used to construct the business cycle synchronization (correlation) variable. One measure is the natural logarithm of real GDP, detrended with a Hodrick and Prescott (1997) filter. We denote this measure of output as ‘‘HP-filtered output’’. The other measure is the annual growth rate of real GDP and is therefore referred to as ‘‘first-differenced output’’. The annual data for real GDP are taken from the World Bank’s World Development Indicators. These two measures have become standard ones and are also used by Clark and van Wincoop (2001) and many other studies for cross-border business cycle studies. Our measure of the bilateral trade intensity, Ti,j,t, is defined as

T i;j;t ¼

xi;j;t þ mi;j;t þ xj;i;t þ mj;i;t ; xi;t þ mi;t þ xj;t þ mj;t

ð5Þ

where xi,j,t is the value of exports from country i to country j at time t, mi,j,t is the value of imports from i to country j at time t, xi,t is the value of country i’s exports to all countries at time t, and mi,t is the value of country i’s imports from all countries at time t. This measure has been used by Baxter and Kouparitsas (2005) to measure bilateral trade intensity.7 The trade data are obtained from the International Monetary Fund’s Direction of Trade Statistics. Since there are no standard measures of bilateral FDI intensity, we construct this index using similar methodology in constructing the bilateral trade intensity. Specifically,

F i;j;t ¼

fi;j;t þ fj;i;t ; fi;t þ fj;t

ð6Þ

where fi,j,t denotes the total FDI (both inward and outward) from country i to country j in year t, and fi,t is the total FDI for country i. The data on bilateral FDI are available from the OECD’s International Direct Investment Statistics. Following Imbs (2004), we build up the dissimilarity measurement as

DSi;j;t ¼

K X

jsk;i;t  sk;j;t j;

ð7Þ

k

where sk,i,t is the GDP share of industry k in country i at time t.8 A larger value of DSi,j,t indicates a greater degree of dissimilarity in industrial structure. This indicator uses the 1-digit level manufacturing sector taken from the OECD’s Stan Indicators Database. Eq. (1) illustrates the major determinants of output synchronization and is the main focus of this paper. To be consistent with our intuition discussed in the Introduction that FDI and business cycle co-movements have a positive relationship, it should be the case that a1 > 0. As for the sign of a2, Frankel and Rose (1998) find that closer trade ties among industrialized countries result in tighter business cycle co-movements because common shocks are more prevalent and intra-industry trade dominates. We therefore expect the sign of a2 to be positive. For a3, if business cycles are driven by industry-specific shocks, then countries with greater similarities in industrial structure tend to move together (Imbs, 2004). We should thus expect that a3 < 0. The exogenous variable set Z1t includes a measure for the similarity of the monetary polices of the two countries. Similar monetary policy behavior is likely to be the outcome of similar economic structures and underlying shocks (Otto et al., 2001). In this case, Z1t controls for the possibility of a common shock to both economies from an external source, while FDI and trade in Eq. (1) catch the channels of transmitting shocks from one country to another.9 Eqs. (2) and (3) are for bilateral FDI and bilateral trade, respectively. The FDI equation is the key equation that has been omitted in the previous studies for business cycle co-movements. Because the bilateral FDI and trade flows are likely to be motivated by common characteristics that are specific to the country-pair, we expect the sign of b1 to be positive. Such a relationship is supported by our empirical results from using the EC3SLS estimation method. Even though there is no theoretical guidance, the industrial dissimilarity variable is included in the FDI equation because it is a crucial variable for trade and might be important for FDI as well. The vector of the exogenous variables, Z2, includes a measure of monetary policy 6 In the empirical section, models with or without a deterministic time trend in each equation are both considered to control for the likely trend component in some of the dependent variables. 7 One may question the double counting in this measure as the exports from i to j are counted as imports from j to i. However, xi,j,t – mj,i,t is a problem associated with the bilateral trade data. An alternative measure, (xi,j,t + mi,j,t)/(xi,t + mi,t + xj,t + mj,t), is used in earlier work. Nevertheless, it may understate the actual trade flows, a concern acknowledged by Frankel and Rose (1998). This concern explains why the measure defined in (5) becomes a standard one in more recent empirical studies. 8 Baxter and Kouparitsas (2005) define ‘‘similarity’’ in industrial structure as 1  DSi,j,t. 9 Other variables that might be able to control for the possibility of a common shock to both countries are the similarity in fiscal policy and exchange rates (Inklaar et al., 2008), and financial integration (Imbs, 2004). Including all these variables may render insensible results due to a high degree of collinearity between the variables (Otto et al., 2001). We therefore include only the similarity in monetary policy in Z1t for it is a standard one used in most of the prior studies for output co-movements. In a sensitivity test, which is available upon request, we find that once the monetary policy indicator is included in the output correlation equation, it is enough to obtain robust estimation results. Similar conclusion has been reached in Imbs (2004).

774

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783

closeness and a dummy variable for common legal origin. According to Frankel and Rose (1998), similar monetary policies might explain in part the high output correlation among European countries. In this equation, the explanation for the monetary policy coordination is based on its role through the FDI transmission channel. As has been discussed in the Introduction, the channels of business cycle transmission through FDI are related to the notion of technology spillover and financial integration in a general sense. If two countries share similar monetary policy objectives and are highly integrated through FDI, then an idiosyncratic shock to one country can be transmitted to the other country’s real activities through this channel. Another exogenous variable in Z2 used to achieve identification is an indicator for common legal origin in the two countries. Legal origin, as an important factor in forming government infrastructure and financial development, is emphasized by La Porta et al. (1998). Countries with similar legal systems may have more integrated financial markets and corporate regulations, which could produce the propagation of financial shocks between countries. In this paper, the monetary policy closeness is calculated as the correlation between the short-term interest rates of two countries and the legal origin data are taken from La Porta et al. (1998). In Eq. (3) for trade flows, the relationship between trade and FDI depends on the nature of FDI. Horizontal FDI, where multiple-plant firms undertake the same production activities in multiple countries, has been distinguished from vertical FDI, where firms locate different stages of production in different countries (Markusen and Maskus, 2002; Amiti and Wakelin, 2003). If c1 > 0, this indicates that FDI is mostly of the vertical type since this type of FDI is conducted according to relative factor prices and could boost trade. Conversely, if the type of FDI is horizontal, then c1 < 0. This is because the production of a homogeneous good in the host country is meant to seek the overseas market and avoid high transportation costs. Therefore, horizontal FDI is more likely to substitute for trade. It is well perceived that industrial differences might generate trade, which implies that c2 > 0 (Balassa, 1986). For the gravity variables in Z3, we include exogenous variables such as a dummy for a common official language, a dummy for land adjacency, the log of geographic distance between the two countries’ capitals, and the log of the ratio of the two countries’ GDPs. These exogenous variables have been used by Imbs (2004) and Baxter and Kouparitsas (2005) in analyzing the association between trade flows and business cycle co-movements. They are believed to contain a clear exogeneity property and have strong predictive power for trade flows.10 Eq. (4) is for industrial dissimilarity. If FDI and/or trade lead to industry fragmentation, we shall observe d1 > 0 and/or d2 > 0. On the other hand, if FDI and/or trade lead to industry concentration, d1 < 0 and/or d2 < 0. Classical Ricardian theory, such as Dornbusch et al. (1977), predict a positive linkage between trade and specialization, which implies a positive d2. On the contrary, there are no rigorous theories giving rise to implications about the relationship between specialization and FDI. Hence, the sign of d1 is ambiguous. The exogenous determinants of the industrial dissimilarity are collected in vector Z4, which contains the log of the ratio of GDP and the log of the product of GDP for different countries. These variables are included following Imbs’ (2004) recommendations. They are believed to be legitimate exogenous variables and can affect the patterns of specialization (Imbs and Wacziarg, 2003).11 2.2. Econometric methodology In this paper, we use two methods to estimate the system of multiple equations for (1)–(4). The first method involves the single equation estimation. For comparison purposes, we apply the random effects (RE) and fixed effects (FE) models for pairspecific effects. The Hausman test is used to conduct a specification test for the RE null hypothesis against the FE alternative. The second method involves the simultaneous equations model with the error component three-stage least squares (EC3SLS) estimates proposed by Baltagi (1981). Both estimation methods adopt the panel data approach, which embeds the characteristics of the cross-sectional and time-series data. The panel data approach considers the unobserved country-specific effects and offers advantages to the cross-sectional regression on examining the FDI effects; see Levine et al. (2000) and Carkovic and Levine (2005). Such advantages are also identified from the empirical evidences provided in this paper. The error component three-stage least squares estimator cannot only deal with the endogeneity problem, but it also shares the features of simultaneous equations. We can represent the system of Eqs. (1)–(4) by

3 1 6 F 7 7 6 7 2 3 36 0 6 T 7 e1 7 6 7 7 6 6 0 76 DS 7 6 e2 7 7 7 þ 6 7: 76 4 e3 5 Z 0 56 1 7 7 6 7 d3 6 e4 6 Z2 7 7 6 4 Z3 5 2

2

q

3

2

a0 a1 a2 a3 a4

6 F 7 6b 7 6 0 6 7¼6 6 4 T 5 4 c0 DS

d0

0

0

0

b1

b2

0

b3

0

c1

0

c2

0

0

c3

d1

d2

0

0

0

0

ð8Þ

Z4 10

For more details, see the arguments in Imbs (2004, 2006). Imbs and Wacziarg (2003) identify a non-monotonic relationship between per capita income and the stages of specialization. They show that, as income per capita grows, economies initially diversify (in the sense of spreading economic activities more equally across sectors) but start specializing after a higher level of income per capita is reached. 11

775

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783 Table 1 Summary statistics.

qD

qHP

FDI

Trade

Dissimilarity

0.211 0.116 0.287

1.000 0.198 0.113 0.311

1.000 0.401 0.334

1.000 0.087

1.000

0.465 0.517

0.244 0.691

0.008 0.012

0.055 0.065

0.202 0.095

Sample correlation between variables qD 1.000

qHP FDI Trade Dissimilarity Summary statistics Mean Std. dev.

For the business cycle synchronization measure, qD is calculated based on the annual growth rate of real GDP and qHP is calculated based on the HP-filtered real GDP. Dissimilarity is a measure for industrial difference as defined in Eq. (7).

Without causing confusion, we can write Eq. (8) in a more concise matrix form as,

Y ¼ kX þ :

ð9Þ

Denote S as a set of instrumental variables, and V as the variance-covariance matrix of e. Then, the EC3SLS estimator of k is

^kEC3SLS ¼ ðX 0 PS X  Þ1 X 0 PS Y  ;

ð10Þ

b 12 Y; X  ¼ V b 12 X; P S ¼ S ðS0 S Þ1 S0 , and S ¼ V b 12 ðI4  SÞ, where V b is a consistent estimator of V. A more detailed with Y ¼ V discussion can be found in Baltagi (2005). This approach provides more information than the pure cross-sectional approach and achieves higher precision for coefficient estimates. 

3. Empirical results Due to data availability, we collect annual observations over the period 1988–2002.12 We focus on the bilateral linkages of the G7 countries, among themselves and with eight other developed countries that are part of the OECD. The G7 countries are Canada, France, Germany, Italy, Japan, the UK, and the US. The remaining eight OECD countries are Australia, Denmark, Korea, the Netherlands, Norway, Spain, Sweden, and Switzerland. From this we create 77 country pairs.13 To construct our panel data, we group the data into three time periods: 1988–1992, 1993–1997, and 1998–2002.14 To sum up, we have 231 observations in our sample. Table 1 reports the unconditional correlations and summary statistics for the four endogenous variables in the system. We use two measures for the co-movements of the business cycles; one is based on the HP-filtered output data (qHP) and the other one is based on the first-differenced output data (qD). We find that FDI and trade are positively correlated with both cycle synchronization measures. Interestingly, the correlation of cycle synchronization with FDI is larger than that with trade. This might imply that the cross-border shock transmission mechanism from FDI is more influential than that from the channel of trade over the sample period 1988–2002. It is noted that industrial dissimilarity is negatively correlated with both correlation measures. 3.1. The single equation estimation For comparison, we first consider the cross-sectional regression for the time-averaged data over the period 1988–2002.15 Column (1) of Table 2 shows the results of the ordinary least squares (OLS) estimation when the FDI variable is not included. Similar to Frankel and Rose (1998), Kose and Yi (2002) and Baxter and Kouparitsas (2005), we find that higher bilateral trade between two countries is associated with more correlated business cycles. We also find that industrial similarity can enhance business cycle co-movement, as shown by Imbs (2004). When the FDI variable is included, Column (2) of Table 2 indicates that the FDI coefficient is not significant. Monetary policy closeness is not a significant regressor in the correlation equation either. However, the OLS estimates are not reliable, because the OLS method ignores the endogeneity problem and may lead to 12 The data for bilateral FDI after 2002 consist of many missing values. Therefore, we are prevented from expanding our sample to include data for more recent years. 13 With 15 countries in our sample, however, we are unable to use a full set of bilateral country pairs (which will produce N  (N  1)/2 = 105 pairs) because the bilateral FDI data between non-G7 countries are not available over this period. Hence, in our sample, each country pair consists of at least a G7 country. This limitation leads to 77 country pairs. 14 All the variables, except the business cycle correlation, are measured as averages over the periods 1988–1992, 1993–1997, and 1998–2002. Constructing panel data in this fashion is not rare. For instance, Levine et al. (2000), Barro and Lee (2001), Carkovic and Levine (2005), Rose (2005) and Schiavo (2008) have used this method to construct panel data. The limitation of our panel data is that the output co-movements cannot capture cycles longer than four years. Longer business cycle co-movements can be studied once more FDI data become available. 15 One exception is the output correlation measure, which is calculated pair-wise by the output observations over the whole sample period.

776

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783 Table 2 Single equation estimation with cross-sectional data. (1) (excl. FDI)

(2)

Panel A1 : Correlation (HP-filtered) FDI Trade Dissimilarity Monetary policy

1.117 (0.602)⁄⁄ 2.135(0.387)⁄⁄ 0.020(0.100)

0.500(2.792) 1.174 (0.656)⁄⁄ 2.160 (0.414)⁄⁄ 0.015 (0.090)

Panel A2: Correlation (first-differenced) FDI Trade Dissimilarity Monetary policy

0.849 (0.490)⁄⁄ 1.544 (0.315)⁄⁄ 0.597 (0.068)

0.582 (2.273) 0.915 (0.558) 1.573 (0.734)⁄⁄ 0.0663 (0.073)

Panel B: FDI Trade Dissimilarity Monetary policy Legal origins

0.113 0.050 0.011 0.001

(0.026)⁄⁄ (0.016)⁄⁄ (0.004)⁄⁄ (0.004)

Panel C: Trade FDI Dissimilarity Language Distance Adjacency GDP gap

0.037 0.011 0.031 0.085 0.013

(0.068) (0.020) (0.013) (0.021)⁄⁄ (0.013)

1.366 0.044 0.018 0.018 0.081 0.009

(0.356)⁄⁄ (0.066) (0.019) (0.013) (0.019)⁄⁄ (0.012)

Panel D: Dissimilarity FDI Trade GDP gap GDP product

0.128 (0.128) 0.068 (0.027)⁄⁄ 0.015 (0.060)

2.011 0.143 0.065 0.006

(0.661)⁄⁄ (0.182) (0.024)⁄⁄ (0.057)

Standard errors are in parentheses. Constant estimates are not reported. ⁄ 10% significance. ⁄⁄ 5% significance.

estimation bias and inconsistency. The bias in the OLS estimates is further supported by the results in Panels (C) and (D), where trade and dissimilarity are not significantly related to each other, an indication hard to reconcile with trade theories and the main findings documented in many earlier studies. Next, we use panel data to investigate the relationship between FDI and business cycle co-movements, first with the single equation approach (reported in Table 3), then with the simultaneous equations approach (reported in Table 4) in Section 3.2. The purpose of presenting the single equation estimates is to compare them with those from models free of endogeneity bias. In Table 3, the results are for single equations that consider fixed effects (FE) or random effects (RE), with or without a time trend. Columns (1) and (2) present the results for models in which the bilateral FDI variable is omitted. In the specifications of Columns (3) and (4), FDI is allowed to enter the equations. To choose the appropriate model between the FE and RE specifications, we rely on the Hausman test. A statistically significant test statistic indicates that the RE specification is rejected against the FE alternative.16 Panels A1 and A2 are results for the single equation of output correlation, using either HPfiltered or first-differenced output measures. The results in Panel A2 with first-differenced output are basically similar with those in Panel A1 with HP-filtered output. In Panels A1 and A2, the Hausman tests cannot reject the null hypothesis at the 5% significance level in 7 out of 8 cases, and therefore suggest that the RE specification is preferred. Furthermore, it is noted that the coefficients of industrial dissimilarity are negatively significant in all specifications with the RE effects. In other words, the homogeneity in industrial structure is associated with highly-correlated business cycles between pairs of countries. This is compatible with the perception that business cycle disturbances are mainly industry-specific. On the other hand, the evidence that a closer bilateral trade relationship would contribute to business cycle synchronization is rather weak. Among these 8 REeffect models for output correlation, the trade variable is significant in three models at the 10% level and insignificant in the remaining five models. To our surprise, when FDI is included in the equation for correlation (Columns (3) and (4) in Panels A1 and A2 of Table 3), FDI displays no significant impact in 7 out of 8 cases. In addition, in no RE-effect specifications can monetary policy closeness help explain business cycle correlation. Panel B of Table 3 displays the results for the FDI equation. The estimates from the FE models, as suggested by the Hausman tests, indicate that trade tends to reduce FDI but that dissimilarity has no significant effect on FDI. The former result is contrary to the intuition that closer trade ties could bring in more FDI when these two activities are likely to be affected by

16 The Hausman test is based on testing the difference between the FE and RE estimates. When the individual effect (i.e., the country-pair-specific effect in our model) and other explanatory variables are correlated, the FE estimates are consistent whereas the RE estimates are not. This is because RE models assume that the individual effect is random with no correlation with other regressors. The Hausman test statistic is asymptotically Chi-squared distributed.

Table 3 Single equation estimation with panel data. Time trend

N (1) (excl. FDI) FE

Panel A2: Correlation (first-differenced) FDI Trade 7.165 (3.299)⁄⁄ Dissimilarity 1.263 (1.170) Monetary policy 0.021 (0.056) Hausman test(prob > v2) 0.02

N (3) FE

Y (4)

RE

FE

RE

0.882 (0.661) 1.927 (0.417)⁄⁄ 0.065 (0.054)

4.016 (4.850) 1.325 (1.474) 0.102 (0.070) 0.19

1.171 (0.642)⁄ 1.828 (0.404)⁄⁄ 0.035 (0.053)

8.884 (5.320)⁄ 3.509 (4.604) 0.894 (1.534) 0.103 (0.073) 0.10

2.305 0.620 1.804 0.055

(2.519) (0.720) (0.439)⁄⁄ (0.056)

2.924 4.463 1.368 0.104 0.26

(5.365) (4.931) (1.480) (0.070)⁄⁄

0.035 1.175 1.830 0.035

(2.505) (0.710)⁄ (0.424)⁄⁄ (0.054)

0.730 (0.516) 1.373 (0.326)⁄⁄ 0.439 (0.042)⁄⁄

1.432 (3.780) 1.627 (1.149) 0.024 (0.055) 0.10

0.912 (0.508)⁄ 1.312 (0.320)⁄⁄ 0.024 (0.042)

6.622 (4.049) 5.154 (3.504) 1.425 (1.167) 0.026 (0.056) 0.31

2.563 0.439 1.237 0.032

(1.964) (0.562) (0.342)⁄⁄ (0.043)

3.446 (4.176) 0.906 (3.838) 1.678 (1.152) 0.026 (0.055) 0.15

1.158 0.773 1.253 0.020

(1.981) (0.562) (0.336)⁄⁄ (0.043)

0.067 0.054 0.002 0.005

(0.026)⁄⁄ (0.014)⁄⁄ (0.001) (0.004)

0.153 (0.074)⁄⁄ 0.015 (0.022) 0.000 (0.001)

0.096 0.044 0.000 0.004

(0.025)⁄⁄ (0.014)⁄⁄ (0.001) (0.004)

0.079 0.039 0.005 0.028 0.088 0.002

(0.091) (0.023)⁄ (0.019) (0.012)⁄⁄ (0.020)⁄⁄ (0.010)

0.685 0.087 0.068 0.080

(0.269)⁄⁄ (0.141) (0.023) (0.054)

Panel B: FDI Trade Dissimilarity Monetary policy Legal origins Hausman test(prob > v2)

0.304 (0.066)⁄⁄ 0.024 (0.023) 0.001 (0.001)

RE

FE

0.00

Panel C: Trade FDI Dissimilarity Language Distance Adjacency GDP gap Hausman test(prob > v2)

0.013 (0.018) 0.20

Panel D: Dissimilarity FDI Trade GDP gap GDP product Hausman test(prob > v2)

0.657 (0.250)⁄⁄ 0.055 (0.051) 0.034 (0.042) 0.00

0.097 (0.028)⁄⁄

0.076 0.003 0.030 0.090 0.010

(0.026)⁄⁄ (0.020) (0.013)⁄⁄ (0.021)⁄⁄ (0.011)

0.132 (0.141) 0.044 (0.022)⁄⁄ 0.033 (0.031)

0.045 (0.025)⁄

0.022 (0.016) 0.29

0.331 (0.262) 0.055 (0.060) 0.383 (0.133)⁄⁄ 0.00

0.042 0.006 0.027 0.088 0.004

(0.023)⁄ (0.020) (0.013)⁄⁄ (0.021)⁄⁄ (0.010)

0.063 (0.145) 0.064 (0.024)⁄⁄ 0.075 (0.056)

0.420 (0.091)⁄⁄ 0.076 (0.027)⁄⁄

0.014 (0.017) 0.00 0.206 (0.312) 0.627 (0.255)⁄⁄ 0.046 (0.053) 0.022 (0.046) 0.00

RE

0.00 0.310 0.060 0.002 0.033 0.091 0.010

(0.093)⁄⁄ (0.026)⁄⁄ (0.019) (0.013)⁄⁄ (0.020)⁄⁄ (0.010)

0.790 0.141 0.055 0.005

(0.266)⁄⁄ (0.136) (0.021)⁄⁄ (0.033)

0.165 (0.091)⁄ 0.043 (0.025)⁄

0.017 (0.016) 0.00 0.297 (0.303) 0.279 (0.267) 0.072 (0.062) 0.412 (0.136) 0.00

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783

Panel A1: Correlation (HP-filtered) FDI Trade 6.207 (4.336) Dissimilarity 0.677 (1.537) Monetary policy 0.097 (0.737) Hausman test(prob > v2) 0.13

Y (2) (excl. FDI)

Standard errors are in parentheses. Constant terms and time trend terms are not reported. p-Values are reported for the Hausman test (for the null hypothesis of the RE specification against the FE specification). 10% significance. ⁄⁄ 5% significance. ⁄

777

778

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783

common institutional factors or infrastructure attributes given a country-pair. Panel C is for the trade equation. When FDI is excluded from the equation, the RE specification is preferred over the FE specification; when FDI is included, the FE specification is preferred. In the RE models, the gravity variables, such as the distance between capitals and adjacency, all have the expected signs. A shared finding, in any of the eight specifications for the trade equations, is that industrial differences are able to promote trade (Balassa, 1986). However, more FDI discourages trade. This result implies that the type of FDI is horizontal. Horizontal FDI often occurs when multinational firms produce final goods in multiple locations, as a result increasing access to the overseas markets and avoiding high trade costs. Therefore, horizontal FDI is likely to serve as a substitute for trade. Panel D presents results for the dissimilarity equation. The results of all our Hausman tests favor FE specifications over RE ones. In these FE-effect dissimilarity equations, the coefficients of trade are significantly positive in the no-time-trend models. That is, trade and dissimilarity complement each other. The main results from the single equation approach reported in Tables 2 and 3 can be summarized as follows. First, monetary policy has no direct impact on the business cycle synchronization. Second, trade and industrial dissimilarity are not statistically related to each other in many model specifications. Basically, these two findings are inconsistent with the predictions of economic theories. Third, the industrial dissimilarity is better able to explain the business cycles synchronization than trade or FDI. However, with the single equation approach, the FE and RE estimates in Table 3 may be subject to the endogeneity bias. Later in Section 3.2 we find the EC3SLS estimates totally reverse this finding, which is another proof that the OLS estimates are inconsistent. Last, the industrial differences can promote trade and FDI is of the horizontal type.

3.2. The simultaneous equations estimation with panel data Next, in Table 4, we estimate the error component three-stage least squares (EC3SLS) model, which provides consistent estimates even when the dependent variables are endogenous. For each output measure, eight model specifications are estimated depending on whether a time trend or FDI is included in the simultaneous equations system. In Panel A we report the results for the correlation equation. We first focus on the model specification where the FDI variable is omitted from the system (Columns (1)–(4)). It is shown that a high degree of bilateral trade is associated with high cross-border business cycle correlation in the two models without a trend. This outcome is similar to the findings in previous studies that ignore the role of FDI in output correlation, such as Frankel and Rose (1998) and Baxter and Kouparitsas (2005). On the contrary, the coefficients of industrial dissimilarity in the synchronization equation in the system become insignificant in each of the model specifications. This is in sharp contrast to the significant findings in the single equation RE-effect estimates, as reported in Table 3. The problem of endogeneity arising from the single equation estimation method might be responsible for the different results. Moreover, similarity in monetary policy is a significant explanatory variable for the output correlation only in the Column (1) specification. Given the above evidence that trade is positively linked with business cycle correlation, the main question we raise in this paper is whether FDI provides another crucial channel on transmitting shocks between countries. In Columns (5)–(8) of Table 4, we estimate the correlation equation when FDI is included in the system. With the EC3SLS estimates, we find that the channels of FDI, trade, and monetary policy closeness are more or less equally important in explaining the business cycle correlation, all with positive relations. The results confirm our intuition that intensive FDI activities could contribute to output co-movements similar to the way in which trade activities could. During our sample period 1988–2002, an average of 74% of world FDI, including inflows and outflows, is produced by these 77 country pairs. Over time, as a consequence of the continuous growth in FDI flows among the 15 developed countries, shocks are easily transmitted from one country to another. Unlike the single equation estimates that suggest that monetary policy plays only a trivial role in determining output correlation, with the EC3SLS estimates, the monetary policy closeness is found to contribute to output co-movements as a separate channel for transmitting shocks. On the other hand, the coefficients of dissimilarity remain insignificant in all 4 cases. In the FDI equation, the coefficient of trade, b1, is expected to be positive if bilateral trade is able to promote FDI between two countries. This proposition is supported by our estimation results in Panel B Columns (5)–(7). The sharp contrast to the ^1 estimates in Table 3 is apparently caused by the endogeneity bias in the single equation approach. While monnegative b etary policy closeness affects FDI positively, legal origins do not have any significant impact on FDI. Hence, an idiosyncratic shock to one country can be transmitted to another country if these two countries share similar monetary policy objectives and are highly associated through FDI. The total effect from monetary policy on output correlation is then the sum of a direct effect and an indirect effect as the multiplication of technology spillover with financial integration. Although dissimilarity (or production specialization) is proved to be tightly associated with trade in the literature, there is little guidance (theoretically and empirically) to suggest that there is a relationship between industrial dissimilarity and FDI. Interestingly, from the results in Table 4, dissimilarity is demonstrated to be a crucial determinant not only for trade but also for FDI. All negative estimates of b2 in the FDI equation are strongly significant at the 5% level, whereas the estimates of c2 in all four trade model specifications are strongly positively significant. The interpretation is that countries with similar industrial structures would trade less, but engage in more FDI. Put differently, trade occurs more intensively between countries that concentrate on different industries, while FDI emerges more heavily for countries that share similar sectoral structures. This observation suggests that for the OECD economies intra-industry FDI activities are more prevalent,

Table 4 Simultaneous equations estimation with panel data: EC3SLS estimates. HP-filtered (excl. FDI)

Time trend

N (1)

4.151 (1.628)⁄⁄ 0.883 (1.244) 0.169 (0.090)⁄

Panel A: Correlation FDI Trade Dissimilarity Monetary policy

First-differenced (excl. FDI)

HP-filtered

Y (2)

N (3)

Y (4)

N (5)

Y (6)

6.155 (5.902) 1.616 (4.537) 1.001 (0.354)

4.521 (1.306)⁄⁄ 0.785 (0.998) 0.091 (0.073)

5.185 (4.149) 0.861 (3.112) 0.416 (0.363)

58.420(16.512)⁄⁄ 14.611(3.228)⁄⁄ 0.261 (2.120) 0.351 (0.157)⁄⁄

37.054 17.005 1.001 0.327

(14.654)⁄⁄ (2.762)⁄⁄ (1.763) (0.128)⁄⁄

30.624 3.178 0.994 0.180

(13.062)⁄⁄ (2.595) (1.712) (0.118)

0.545 0.069 0.010 0.001

(0.029)⁄⁄ (0.020)⁄⁄ (0.002)⁄⁄ (0.002)

0.543 0.068 0.010 0.001

(0.027)⁄⁄ (0.019)⁄⁄ (0.002)⁄⁄ (0.002)

0.155 0.074 0.004 0.008

(0.048)⁄⁄ (0.164) (0.021) (0.154) (0.011)

0.770 0.380 0.027 0.035 0.107 0.003

(0.287)⁄⁄ (0.134)⁄⁄ (0.025) (0.016)⁄⁄ (0.025)⁄⁄ (0.013)

0.898 0.364 0.010 0.016 0.097 0.014

(0.849) (0.131)⁄⁄ (0.027) (0.018) (0.026)⁄⁄ (0.015)

5.538 (0.451)⁄⁄ 0.023 (0.026) 0.003 (0.013)

4.106 1.230 0.142 0.278

(1.780)⁄⁄ (0.618)⁄⁄ (0.064)⁄⁄ (0.126)⁄⁄

4.169 1.287 0.149 0.303

(2.933) (0.772)⁄ (0.082)⁄ (0.146)⁄⁄

Panel B: FDI Trade Dissimilarity Monetary policy Legal origins Panel C: Trade FDI Dissimilarity Language Distance Adjacency GDP gap

0.844 0.005 0.034 0.089 0.029

(0.138)⁄⁄ (0.053) (0.034) (0.055) (0.021)

Panel D: Dissimilarity FDI Trade 2.271 (0.757)⁄⁄ GDP gap 0.013 (0.051) GDP product 0.101 (0.070)

0.646 0.020 0.029 0.080 0.004

(0.075)⁄⁄ (0.283) (0.184) (0.302) (0.018)

2.006 (0.700)⁄⁄ 0.033 (0.052) 0.177 (0.131)

0.804 0.007 0.037 0.085 0.019

(0.138)⁄⁄ (0.073) (0.030) (0.065) (0.015)

2.571 (0.857)⁄⁄ 0.014 (0.061) 0.131 (0.092)

0.660 0.014 0.007 0.151 0.005

First-differenced N (7)

Y (8) 15.002 5.562 0.686 0.303

(8.827)⁄ (2.116)⁄⁄ (1.195) (0.19)⁄⁄

(0.085)⁄ (0.025)⁄⁄ (0.001)⁄⁄ (0.005)

0.123 0.047 0.014 0.009

(0.256) (0.154) (0.004)⁄⁄ (0.016)

0.741 0.434 0.044 0.038 0.082 0.004

(0.315)⁄⁄ (0.141)⁄⁄ (0.025)⁄ (0.015)⁄⁄ (0.031)⁄⁄ (0.013)

0.057 0.626 0.040 0.033 0.099 0.000

(0.796) (0.106)⁄⁄ (0.033) (0.030) (0.056)⁄ (0.025)

5.318 2.267 0.232 0.457

(5.662) (1.005)⁄⁄ (0.241) (0.357)

5.098 2.309 0.231 0.468

(5.640) (0.985)⁄ (0.251) (0.416)

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783

Output measure

Standard errors are in parentheses. Constant and time trend estimates are not reported. ⁄ 10% significance. ⁄⁄ 5% significance.

779

780

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783

which is consistent with the horizontal FDI behavior implied from the estimates in the trade equation (discussed in the next paragraph). In the trade equation, a negative (positive) coefficient of FDI would indicate that FDI is horizontal (vertical). According to the significantly negative coefficients in Panel C Columns (5) and (7) of Table 4, FDI is likely to replace trade and is therefore of the horizontal type, a common finding shared in Table 3. Because the OECD countries in our sample have similar economic size and relative endowments, it makes sense to identify horizontal-type FDI among them (Markusen and Maskus, 2002). The positive coefficient on dissimilarity supports the argument that economies with different industrial structures will enjoy an abundance of Ricardian-type trade (or inter-industry trade).17 In addition, no gravity variables are significant when FDI is omitted from the system. In the model specifications with FDI, the gravity variables that are statistically significant at least at the 10% level all have the expected signs in the trade equation. For instance, adjacency boosts trade because the transportation cost is limited. For the same reason, more distance would distress trade. Finally, Panel D of Table 4 displays the results for the dissimilarity equation. In the simultaneous equations system that excludes the bilateral FDI variable, dissimilarity and trade are found to be positively related to each other (Panels C and D, Columns (1)–(4)), while no other explanatory variables are statistically significant at the conventional levels. In the model specification of Column (5), when FDI is included as a regressor, apart from trade flows, FDI and the two proxy variables determining the development stages of industrial specialization become statistically significant. In the model specification of Columns (6)–(8), the impact of FDI on industrial dissimilarity is still negative, but insignificantly.

3.3. Robustness analysis To evaluate the sensitivity of our results, we have estimated models that employ various sets of exogenous variables. Basically, using different sets of exogenous variables does not change our main conclusions drawn from our earlier discussions. For instance, we estimate a model in which the gravity variables in the trade equation (language, distance, adjacency, and the GDP gap) are also included in the FDI equation. The single equation estimates are quite similar to those that have been reported in Table 3.18 That is, the unreliable estimates from the single equation method suggest that FDI and trade do not have a significant association with the output correlation while dissimilarity does. On the contrary, the EC3SLS estimates from the simultaneous equations system, presented in Table 5, support the main findings drawn from Table 4; namely, FDI, trade flows, and monetary policy closeness all have power in explaining business cycle co-movements while dissimilarity has none. Industrial dissimilarity (or production specialization) still has an impact on output synchronization, but rather indirectly through the channels of FDI and trade. The estimated coefficients in the correlation equation are shown to be robust in terms of significance and signs with one exception, i.e. the effect of monetary policy closeness on output co-movements turns negative in the model specification of Column (2) Table 5. Similar EC3SLS estimates for the FDI equation are also identified in Tables 4 and 5, which include that stronger ties in trade relations lead to more FDI activities, countries with more similar sectoral structure engage more FDI, similarity in monetary policy helps FDI, and that the legal origin variable does not have explanatory power for FDI. Among those newly-added exogenous variables in the FDI equation, only the adjacency dummy explains FDI negatively at the conventional levels. Adjacent countries will engage in less FDI since low transportation costs keep trade from being overwhelmingly replaced by FDI. The other key implications from Tables 4 and 5 for the trade and FDI equations are similar too. Basically, we can conclude that the main results remain valid in these alternative models with various exogenous variable sets, and that our results are robust to different model specifications. In brief, the simultaneous equations and the single equation approaches lead to quite different conclusions, and while the former approach provides us with more reliable estimates, the latter potentially suffers from the endogeneity problem. The growing importance of FDI in the past two decades and its channel for international business cycle transmission have been neglected in previous empirical studies. In this paper, the simultaneous equations estimates imply a strong positive relationship between the bilateral FDI and business cycle co-movements. We have demonstrated the crucial role played by FDI in cross-border business cycle shock transmission and confirm that the FDI in our sample is mostly of the horizontal type. To demonstrate the merits of employing the panel data approach proposed in this paper, we estimate a simultaneous equations model with the cross-sectional data and report the 3SLS estimation results in Table 6. The 3SLS estimations yield fewer statistically significant coefficients than the EC3SLS ones. For example, with the HP-filtered output measure, only the FDI variable has a significant impact on output correlation (Column (1)); and with the first-differenced output measure, only the trade variable is significantly related to output correlation (Column (2)). In particular, trade and dissimilarity fail to have a significant association with each other, a result that is hard to reconcile with the main findings documented in many previous studies. This evidence suggests that the panel data method gives more efficient estimates and is superior to the crosssectional data method in analyzing the subject which we are interested in this paper.

17 With an earlier sample (1980–1999) constructed from 24 countries, Imbs (2004) obtains a negative coefficient on the specialization in the equation. Because FDI had not begun its dramatic growth until 1995, Imbs’ results basically capture the effect of intra-industry trade. 18 To save space, the results for the single equation approach are not reported here, but are available from the authors upon request.

781

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783 Table 5 Simultaneous equations estimation with panel data: EC3SLS estimates. Output measure

HP-filtered

Time trend

N (1)

First-differenced Y (2)

N (3)

Y (4)

Panel A: Correlation FDI Trade Dissimilarity Monetary policy

54.275 14.780 0.123 0.338

(15.791)⁄⁄ (3.074)⁄⁄ (2.016) (0.152)⁄⁄

43.638 16.086 0.692 0.323

(13.205)⁄⁄ (2.387)⁄⁄ (1.547) (0.160)⁄⁄

35.893 0.057 0.800 0.411

(18.626)⁄ (3.726) (2.461) (0.166)⁄⁄

21.828 9.551 0.661 0.121

(10.152)⁄⁄ (1.860)⁄⁄ (1.200) (0.111)

Panel B: FDI Trade Dissimilarity Monetary policy Legal origins Language Distance Adjacency

1.556 0.097 0.004 0.016 0.010 0.026 0.092

(0.811)⁄ (0.102) (0.005) (0.018) (0.016) (0.027) (0.067)

0.788 0.083 0.008 0.002 0.006 0.002 0.032

(0.190)⁄⁄ (0.041)⁄⁄ (0.004)⁄⁄ (0.005) (0.006) (0.007) (0.016)⁄⁄

0.790 0.098 0.007 0.002 0.005 0.002 0.032

(0.165)⁄⁄ (0.034)⁄⁄ (0.003)⁄⁄ (0.004) (0.005) (0.005) (0.014)⁄⁄

0.789 0.079 0.006 0.002 0.006 0.002 0.035

(0.186)⁄⁄ (0.052) (0.004)⁄⁄ (0.005) (0.007) (0.008) (0.015)⁄⁄

Panel C: Trade FDI Dissimilarity Language Distance Adjacency GDP gap

0.753 0.391 0.038 0.039 0.083 0.006

(0.309)⁄⁄ (0.138)⁄⁄ (0.027) (0.016)⁄⁄ (0.032)⁄⁄ (0.013)

0.946 0.407 0.013 0.015 0.098 0.017

(0.860) (0.133)⁄⁄ (0.028) (0.018) (0.026)⁄⁄ (0.015)

0.819 0.510 0.037 0.034 0.112 0.001

(0.292)⁄⁄ (0.139)⁄⁄ (0.027) (0.017)⁄⁄ (0.028)⁄⁄ (0.014)

1.240 0.431 0.015 0.014 0.085 0.015

(0.891) (0.133)⁄⁄ (0.036) (0.021) (0.039)⁄⁄ (0.014)

Panel D: Dissimilarity FDI Trade GDP gap GDP product

5.776 1.622 0.171 0.398

(3.255)⁄ (0.814)⁄⁄ (0.108)⁄ (0.218)

4.257 2.190 0.226 0.451

(5.064) (1.066)⁄⁄ (0.248) (0.399)

4.504 2.152 0.219 0.401

(5.243) (1.117)⁄⁄ (0.251) (0.362)

4.888 2.341 0.227 0.464

(5.981) (1.008)⁄⁄ (0.274) (0.454)

Standard errors are in parentheses. Constant terms and time trend are not reported. ⁄ 10% significance. ⁄⁄ 5% significance.

Table 6 Simultaneous equations estimation with cross-sectional data: 3SLS estimates. Output measure

(2) First-differenced

Panel A: Correlation FDI Trade Dissimilarity Monetary policy

17.723 1.093 0.871 0.201

(8.794)⁄⁄ (1.626) (1.111) (0.126)

4.814 2.348 0.188 0.024

(7.154) (1.295)⁄ (0.887) (0.101)

Panel B: FDI Trade Dissimilarity Monetary policy Legal origins

0.126 0.072 0.008 0.002

(0.042)⁄⁄ (0.036)⁄ (0.003) (0.003)

0.125 0.071 0.008 0.002

(0.042)⁄⁄ (0.036)⁄ (0.003)⁄⁄ (0.003)

Panel C: Trade FDI Dissimilarity Language Distance Adjacency GDP gap

0.654 0.526 0.037 0.044 0.047 0.048

(1.855) (1.421) (0.089) (0.045) (0.056) (0.099)

0.644 1.578 0.103 0.079 0.005 0.121

(1.782) (1.343) (0.084) (0.043)⁄ (0.053) (0.093)

Panel D: Dissimilarity FDI Trade GDP gap GDP product

2.293 0.161 0.040 0.049

(1.374) (0.336) (0.022)⁄ (0.050)

2.207 0.149 0.041 0.055

(1.378) (0.337) (0.022)⁄ (0.052)

Standard errors are in parentheses. Constant terms are not reported. 10% significance. ⁄⁄ 5% significance. ⁄

(1) HP-filtered

782

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783

4. Conclusions This paper aims to investigate the role of bilateral foreign direct investment (FDI) in influencing business cycle co-movements. The dataset we use has a panel structure and is composed of 77 OECD country-pairs over the period 1988–2002. In a departure from the previous studies that estimate single equations, we adopt the simultaneous equations method with panel data to obtain the error component three-stage least squares (EC3SLS) estimates. The EC3SLS estimates are demonstrated to be superior to the OLS estimates of single equation models and the 3SLS estimates of cross-sectional regressions, for the OLS method entails endogeneity bias and the 3SLS method is less efficient when it fails to use the dynamic information embedded in the panel data. We find that, apart from the conventionally-known trade channel, FDI provides another important channel for transmitting shocks from one country to another, a perception that has been ignored in the previous studies on business cycle synchronization. Specialization in production (or industrial dissimilarity) only has an indirect impact on the business cycle synchronization through the channels of trade and FDI. Interestingly, industrial dissimilarity is positively related to trade, but negatively related to FDI. This observation further verifies our conclusion that among these developed countries, FDI is identified mostly to be of the horizontal type and tends to replace trade when transportation cost is a concern. Given the strong growth of FDI in the past two decades, we expect to observe an increasing influence of FDI compared to trade, particularly when the source of disturbances is sector-specific. An immediate implication from the evidence presented in this paper is to call for more research effort that incorporates the role of FDI into macroeconomic modeling. Appendix The details of the data sources and variable definitions are described below: Correlation: Bilateral correlation of detrended or first-differenced real GDP. Source: World Bank, World Development Indicators. FDI: Bilateral FDI is constructed using Eq. (6). Source: OECD, International Direct Investment Statistics. Trade: Bilateral trade is constructed using Eq. (5). Source: International Monetary Fund, Direction of Trade Statistics. Dissimilarity: Manufacturing sector dissimilarity, absolute difference. Source: OECD STAN Indicators Database. Monetary policy: The correlation of short-term interest rates between two countries using money market rates. Source: International Monetary Fund, International Financial Statistics. Legal origins: Dummy variable that equals unity when both countries share the same legal origins, 0 otherwise. Source: La Porta et al. (1998). Language: Dummy variable that equals unity when both countries share a common language, 0 otherwise. Source: Andrew Rose’s website at http://faculty.haas.berkeley.edu/arose. Distance: The log mile distance between the countries’ capitals. Source: Andrew Rose’s website at http://faculty.haas.berkeley.edu/arose. Adjacency: Dummy variable that equals unity when both countries are adjacent to one another, 0 otherwise. Source: Andrew Rose’s website at http://faculty.haas.berkeley.edu/arose. GDP gap: The log of the ratio of each country’s real GDP. Source: World Bank, World Development Indicators. GDP product: The log of the product of each country’s real GDP. Source: World Bank, World Development Indicators.

References Amiti, M., Wakelin, K., 2003. Investment liberalization and international trade. Journal of International Economics 61, 101–126. Balassa, B., 1986. The determinants of intra-industry specialization in United States trade. Oxford Economic Papers 38, 220–233. Baltagi, B.H., 1981. Simultaneous equations with error components. Journal of Econometrics 17, 189–200. Baltagi, B.H., 2005. Econometric Analysis of Panel Data, third ed. John Wiley and Sons Press. Barro, R.J., Lee, J.W., 2001. International data on educational attainment: updates and implications. Oxford Economic Papers 53, 541–563. Baxter, M., Kouparitsas, M.A., 2005. Determinants of business cycle comovement: a robust analysis. Journal of Monetary Economics 52, 113–157. Carkovic, M., Levine, R., 2005. Does foreign direct investment accelerate economic growth? In: Moran, T.H., Graham, E.M., Blomström, M. (Eds.), Does Foreign Direct Investment Promote Development. Institute for International Economics Press, Washington, D.C, pp. 195–220. Clark, T., van Wincoop, E., 2001. Borders and business cycles. Journal of International Economics 55, 56–85. Crosby, M., 2003. Business cycle correlations in Asia–Pacific. Economics Letters 80, 35–44. de Haan, J., Inklaar, R., Jong-A-Pin, R., 2008. Will business cycles in the Euro area converge? A critical survey of empirical research. Journal of Economic Surveys 22, 234–273. Dornbusch, R., Fisher, S., Samuelson, P., 1977. Comparative advantage, trade and payments in a Ricardian model with a continuum of goods. American Economic Review 67, 823–839. Frankel, J.A., Rose, A.K., 1998. The endogeneity of the optimum currency area criteria. Economic Journal 108, 1009–1025. Gruben, W.C., Koo, J., Millis, E., 2002. How much does international trade affect business cycle synchronization? Working Paper. Helpman, E., Melitz, M.J., Yeaple, S.R., 2004. Export versus FDI with heterogeneous firms. American Economic Review 94, 300–316. Hodrick, R., Prescott, E., 1997. Postwar business cycles. Journal of Money, Credit and Banking 29, 1–16. Imbs, J., 2004. Trade, finance, specialization, and synchronization. Review of Economics and Statistics 86, 723–734. Imbs, J., 2006. The real effects of financial integration. Journal of International Economics 62, 296–324. Imbs, J., Wacziarg, R., 2003. Stages of diversification. American Economic Review 93, 63–86.

C.-C. Hsu et al. / Journal of Macroeconomics 33 (2011) 770–783

783

Inklaar, R., Jong-A-Pin, R., de Haan, J., 2008. Trade and business cycle synchronization in OECD countries – a re-examination. European Economic Review 52, 646–666. Jansen, W.J., Stockman, A.C.J., 2004. Foreign direct investment and international business cycle comovement. Working Paper. Kalemli-Ozcan, S., Sørensen, B., Yosha, O., 2001. Economic integration, industrial specialization, and the asymmetry of macroeconomic fluctuations. Journal of International Economics 55, 107–137. Kose, A., Yi, K.M., 2002. The trade-comovement problem in international macroeconomics. Working Paper. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 1998. Law and finance. Journal of Political Economy 106, 1113–1155. Levine, R., Loayza, N., Beck, T., 2000. Financial intermediation and growth: causality and causes. Journal of Monetary Economics 43, 31–77. Levy Yeyati, E., Panizza, U., Stein, E., 2007. The cyclical nature of North–South FDI flows. Journal of International Money and Finance 26, 104–130. Markusen, J.R., Maskus, K.E., 2002. General equilibrium approaches to the multinational firm: a review of theory and evidence. In: Harrigan, J. (Ed.), Handbook of Empirical International Trade. Blackwell, London. OECD, 2008. OECD Benchmark Definition of Foreign Direct Investment, fourth ed. OECD Publishing. Otto, G., Voss, G., Willard, L., 2001. Understanding OECD output correlations. Working Paper. Rose, A.K., 2005. Does the WTO make trade more stable? Open Economies Review 16, 7–22. UNCTAD, 2008. World Investment Report 2008. New York and Geneva, United Nations. Schiavo, S., 2008. Financial integration, GDP correlation and the endogeneity of optimum currency areas. Economica 75, 165–189.