Has Globalization Really Increased Business Cycle Synchronization?
  • 1 https://isni.org/isni/0000000404811396, International Monetary Fund
  • | 2 https://isni.org/isni/0000000404811396, International Monetary Fund

Contributor Notes

Author’s E-Mail Address: eric.monnet@banque-france.fr; dpuy@imf.org;

This paper assesses the strength of business cycle synchronization between 1950 and 2014 in a sample of 21 countries using a new quarterly dataset based on IMF archival data. Contrary to the common wisdom, we find that the globalization period is not associated with more output synchronization at the global level. The world business cycle was as strong during Bretton Woods (1950-1971) than during the Globalization period (1984-2006). Although globalization did not affect the average level of co-movement, trade and financial integration strongly affect the way countries co-move with the rest of the world. We find that financial integration de-synchronizes national outputs from the world cycle, although the magnitude of this effect depends crucially on the type of shocks hitting the world economy. This de-synchronizing effect has offset the synchronizing impact of other forces, such as increased trade integration.

Abstract

This paper assesses the strength of business cycle synchronization between 1950 and 2014 in a sample of 21 countries using a new quarterly dataset based on IMF archival data. Contrary to the common wisdom, we find that the globalization period is not associated with more output synchronization at the global level. The world business cycle was as strong during Bretton Woods (1950-1971) than during the Globalization period (1984-2006). Although globalization did not affect the average level of co-movement, trade and financial integration strongly affect the way countries co-move with the rest of the world. We find that financial integration de-synchronizes national outputs from the world cycle, although the magnitude of this effect depends crucially on the type of shocks hitting the world economy. This de-synchronizing effect has offset the synchronizing impact of other forces, such as increased trade integration.

I- Introduction

The synchronized collapse in output in a number of advanced and emerging markets in the wake of the global financial crisis has reinforced the view that economic and financial globalization generates intense co-movement across countries. Although economic theory is still inconclusive as to the exact effect of integration on the properties of business cycle co-movement, a number of empirical and policy contributions have argued that intense integration – i.e. before the World War I and after the mid 1980s - is associated with a strong degree of business cycle synchronization.2 Interestingly, an implicit corollary of that view is that the Bretton Woods era (1946-1971) – a period during which both trade and financial integration were substantially below both pre-WWI and current levels (Bordo, Eichengreen, Irwin, 1999) – is presented as the low point of business cycle synchronization.

There are several reasons behind this popular view. First, at least in theory, the Bretton Woods system was explicitly designed to avoid the international transmission of business cycles (Williamson 1985, Bordo 1993). In an effort to avoid the international spread of depression that affected world economies in the 1930s, the negotiators of Bretton Woods strived to set up an institutional machinery that would not only temper the global business cycle - in particular by avoiding currency wars - but also prevent small shocks or short-term imbalances to propagate.3 Second, the channels behind business cycle propagation were de facto shut down under the Bretton Woods agreement. In particular, although they were progressively lifted, the extensive use of capital controls and trade barriers imposed extreme restrictions on the potential contagion mechanisms. As a result, the recent empirical contributions which found more co-movement in the recent period only confirmed the common wisdom.

Still, robust empirical evidence on changes in the nature and strength of global co-movement over the postwar period is still missing. Although a number of empirical contributions have argued that macroeconomic fluctuations have become more closely linked since the collapse of Bretton Woods (e.g Kose, Otrok and Whiteman (2008), Lumsdaine and Prasad (2003), Kose, Prasad and Terrones (2003) and Kose et al. (2003b), Bordo and Helbling (2003)), these studies share two important data limitations. First, because quarterly GDP statistics computed by the OECD - which have been extensively used in empirical contributions - are not available before 1960, these studies cannot identify a world cycle between 1950 and 1960. This implies that the sample used for estimation always excludes two important global cycles of the Bretton Woods era (1952-1953 and 1957-1958). Second, even when they are reported, quarterly GDP data in 1960s and 1970s are usually the result of linear interpolations. As we discuss below, we find that this method introduces a bias, which underestimates severely the amount of co-movement across OECD countries between 1960 and 1973.4 As a result, the common wisdom builds, at this stage, on empirical studies which consistently underestimated the amount of co-movement during Bretton Woods.

The objective of this paper is to test empirically whether the world co-moves more in the recent “globalization” period than it used to. To do so, we first address the data issue described above. Instead of relying on GDP data, we assemble a new panel of quarterly Industrial Production (IP) indexes covering the whole post-war period. Because IP data are not available in a digital format between 1950 and 1960, we collect missing data directly from International Financial Statistics (IFS) paper volumes at the IMF archives. With this procedure, we are able to assemble a final panel of 21 countries - seventeen advanced and four emerging - between 1950 Q1 and 2014 Q4. After showing that IP indexes are very accurate trackers of real activity in the past (and in the present), we then estimate a dynamic latent factor model as in Kose, Otrok and Whiteman (2003, 2008) and use variance decompositions to measure the strength of co-movement at the world level. Building on the higher frequency and longer coverage of the database, we estimate the factor model on the full sample and on four different sub-periods: (i) the Bretton Woods period (1951 Q1-1971 Q2), the first common shock period (1972 Q1-1983 Q4), the Globalization period (1984 Q1-2006 Q4) and the second common shock period (2007 Q1-2014 Q4).5 Factors and Variance decompositions are then compared across samples.

Our key findings are as follows. First, we estimate a very precise world factor over the full sample, including the Bretton Woods period. To support our statistical analysis, we conduct a narrative analysis based on IMF annual reports published between 1950 and 2014, and find that the world cycles described in IMF reports match almost exactly the turning points and phases identified by our estimation procedure. Second, we find that the quantitative impact of the world cycle varies significantly across periods. As expected, the share of variance in output accounted for by the global dynamic increases dramatically during common shocks periods (1972-1983 and 2007-2014) for almost all countries. Contrary to the common wisdom however, we do not find any difference in co-movement between Bretton Woods (1950-1971) and the Globalization period (1984-2006). On average, the global dynamic accounts for 20% of output volatility in both periods. Third, we find that, although the average co-movement in both periods is unchanged, the way individual countries co-move with the rest of the world has changed substantially over time. Countries such as the UK, Belgium or Sweden, who were strongly co-moving with the rest of the world during Bretton Woods, have been significantly disconnected from the world dynamic during the Globalization period. In contrast, countries such as France, Italy and to a minor extent the US, who were dominated by idiosyncratic dynamics under Bretton Woods, have been re-synchronizing with the world cycle. A significant number of countries are also found to display constantly low (or high) degree of synchronization with the rest of the world in both periods.

Why do some countries “de-synchronize” while others “re-synchronize”? In the second part of the paper, we study the role of trade and financial integration in rationalizing these findings. Using a simple panel regression controlling for both country and time fixed effects, we show that the changes in synchronization can be explained, to a great extent, by trade and financial integration patterns. We find that financial integration has, on average, a negative impact on the synchronization with the world cycle, although this average effect conceals an important asymmetry. In normal times, we find that increasing financial integration by 10% – measured by the ratio of Foreign Assets plus Foreign Liabilities to GDP - decreases co-movement with the rest of the world by 1%. However, this “de-synchronizing” effect is compensated by an opposite force in financial crisis periods, during which integrated economies tend to co-move more than others. Although the net effect remains negative, this asymmetry shows that if financial integration de-synchronizes national outputs from the world cycle, the magnitude of this effect depends crucially on the type of shock hitting the world economy. For instance, financial integration did not seem to have an (additional) impact during the oil shocks period.

We also find that if trade integration tends to increase co-movement with the rest of the world, this effect disappears when considering only advanced economies and is not restored by the use of interaction terms (to uncover potential asymmetries) or other proxies of trade integration (e.g. value-added trade rather gross trade). This suggests that trade integration only affects the strength of co-movement with the global at early stages of the process. Finally, we find that the other variables that have been discussed in the literature, such as the degree of economic specialization and the foreign exchange regime, are not significant determinants of the synchronization of a country with the global dynamic.

Taken together, our findings are well connected to several literatures. First, our findings naturally contribute to the empirical literature on international business cycle, in particular to Lumsdaine and Prasad (2003), Kose, Prasad and Terrones (2003), and Kose, Otrok and Whiteman (2008). Contrary to these contributions however, we do not find any empirical evidence that countries co-moved more during twenty years of intense trade and financial globalization than they did during the twenty years of Bretton Woods. In addition, the use of a quarterly dataset over a long sample allows us to investigate the determinants of changes in co-movement patterns over time. To our knowledge, existing contributions were limited to cross-section analysis, which did not control for important (time and country) fixed effects (in particular Kose, Prasad and Terrones (2003), and Kose, Otrok and Whiteman (2003)).

Second, our results illustrate the role of financial and trade linkages in affecting output co-movement across countries. In general, the positive impact of trade is in line with the vast literature documenting the positive impact of trade integration on output correlations.6 The strong and asymmetric impact of financial integration we identify echoes the recent empirical findings of Kalemli-Ozcan, Papaioannou and Perri (2013), Kalemli-Ozcan, Papaioannou and Peydro (2013) and Duval et al. (2015). Contrary to most empirical studies that reported a positive link between financial integration and output synchronization, these studies identified (i) a strong negative effect of banking integration on output synchronization once global shocks and country-pair heterogeneity are controlled for and (ii) a positive impact of financial integration on output co-movement during financial crisis. We show that this result affects not only country-pairs co-movement patterns but more generally the way countries co-move with the rest of the world, in particular in the long run.7 Using a long sample also shows that the asymmetric effect of financial integration exists only in the presence of global financial shock, but not real shocks.

From a theoretical perspective, our results are in line with Heathcote and Perri (2004) and Obstfeld (1994) who present a model in which financial integration reduces international correlations in GDP.8 We are also very closely connected to Kalemli-Ozcan, Papaioannou and Perri (2013) who develop a model in which changes in financial integration affect business cycle synchronization differently depending on the nature of the shocks faced by economies (real or financial). In this model, financial integration has significant effects on business cycle synchronization, both during tranquil times (where more integration leads to less co-movement) and during crises periods (where more integration leads to more co-movement).

A final contribution of this paper is to provide a detailed narrative analysis of the world economic cycle since 1950, based on the information published in IMF annual reports from 1950 to present. This analysis is reported in appendix at the end of the paper. Besides showing that the world cycle estimated from our dynamic latent factor model is strongly corroborated by contemporary publications and historical evidence, this narrative documents the potential drivers of the world business cycles. In particular we find that although the US (and US monetary policy) has clearly driven the global dynamic in several instances (e.g 1960-1961; 1979-1982; 1984-1986; 1999-2002), there is no systematic lead of the US over the global business cycle over the post-war period. A US recession does not necessarily affect the global cycle (as in 1953-1954); and a world recession is possible without a US slowdown (e.g. 1976-1977).

From a policy perspective, a corollary of our results is that a low level of integration does not imply, per se, a low level of co-movement in the economic system. Contrary to conventional wisdom (Williamson 1985), we find that the Bretton Woods period was also affected by common cycles, although capital controls were still the norm and financial systems were highly regulated. Conversely, we find that a high level of international financial integration does not always imply stronger co-movement. Whether the risk-sharing effect integration dominates the contagion effect of financial integration depends crucially on the type of shocks hitting the world economy (real vs. financial).

The rest of the paper is constructed as follows. Section II reviews the empirical literature and the dataset used in the paper. Section III presents estimation results for the full sample. Section IV contrasts full sample results with sub-sample results. Section V illustrates the role of finance and trade in rationalizing the results from Section IV. Section VI concludes. A number of robustness checks are also presented in appendix. In particular, we show that all key results (world cycle identification, variance decompositions and changes in synchronization) are robust to the use of GDP in lieu of IP indexes, when such a comparison is possible.

II- Literature and Dataset

Several empirical contributions have measured the importance of global co-movement in output over the post-war period. To date, the empirical support for the presence of a world business cycle is well established. Using a Bayesian factor model on a set of macroeconomic aggregates in a 60-country sample between 1960 and 1990, Kose, Otrok and Whiteman (2003) have shown that a common world factor was an important source of output volatility for most countries over that period. This result has been confirmed by a number of other contributions using different sample and/or methodology, and is also confirmed in the present paper.9

The rapid increase in globalization has also generated a literature interested in measuring the impact of trade and financial integration on the properties of global co-movement. However, robust empirical evidence on changes in the nature and strength of global co-movement is far less conclusive. With the exception of Stock and Watson (2005) and Doyle and Faust (2005), most empirical contributions have argued that macroeconomic fluctuations have become more closely linked since the collapse of Bretton Woods (BW). Relying on the same quantitative framework as in Kose, Otrok and Whiteman (2003), Kose, Otrok and Whiteman (2008) found that a common factor explains a larger fraction of output volatility in G7 countries during the Globalization period than in the BW period. Similar results are found in Lumsdaine and Prasad (2003) and Kose, Prasad and Terrones (2003). Using a longer perspective for 16 countries and a different method, Bordo and Helbling (2003) also showed that there was a secular trend towards increased synchronization for much of the twentieth century, and that output correlation was much lower during 1942-1973 than during 1973-2001. Overall, these studies have supported the (common) view that (i) intense integration generates greater business cycle synchronization and that (ii) until the oil shocks of the 1970s, capital controls and autonomous domestic policies implied little output synchronization (Williamson, 1985).

A common feature of all these studies, however, is that they suffer important data limitations. In the absence of quarterly GDP statistics over the whole post-war period for a large cross section of countries, some studies have relied upon annual GDP data (Kose et al. (2003), Mumtaz, Simonelli and Surico (2011)). Besides washing out small and short-lived cycles however, the use of annual GDP data prevents any time comparison because the amount of data is not sufficient to re-estimate the models on sub-samples. To circumvent this issue, several papers have turned to quarterly GDP data (e.g. Kose, Otrok and Whiteman (2008), Doyle and Faust (2005), Stock and Watson (2005)). But this move was made at a significant cost. First, because quarterly GDP statistics computed by the OECD - which have been extensively used in empirical contributions - are not available before 1960, these studies cannot identify a world cycle between 1950 and 1960. As a result, the sample used for estimation always excludes two important global cycles of the BW era (1952-1953 and 1957-1958). Second, even when they are reported, quarterly GDP data in 1960s and 1970s are, for most countries, the result of linear interpolations. As we discuss in Appendix A, this procedure washes out most of the cyclical dynamics at the country level, making it almost impossible to detect business cycle co-movement, in particular between 1960 and 1973. These data constraints imply that, at this stage, the common wisdom builds mostly on empirical studies, which consistently underestimated the amount of co-movement during BW.

To circumvent this data issue, this paper relies on a new dataset of quarterly Industrial Production (IP). Besides being a very accurate tracker of real activity in the present and in the past (see below), IP indexes have been produced and reported at a high frequency for a much longer period than GDP statistics and for a larger cross section of countries. More importantly, as opposed to quarterly GDP data, IP indexes are not based on interpolations and ex-post estimations. This implies that the quarterly profile derived from IP indexes are a much more accurate depiction of changes in real activity for the period 1960-1980 than the official quarterly GDP statistics (see Appendix A for a comparison).

To construct our sample of IP indexes we use the following procedure: from 1960 onwards, quarterly IP data are downloaded from International Financial Statistics for the largest sample of countries possible. Because IP data are not available in a digital format between 1950 and 1960, we then collect the missing data directly from IFS paper volumes collected from the IMF archives. After this procedure we find that only 21 countries (17 advanced and 4 emerging) have complete data over the whole sample period. As a result, the final panel of IP data we assemble and use for estimation covers 21 countries between 1950 Q1 and 2014 Q4.10 The composition of the final sample is reported in Appendix A.

How good are IP indexes at tracking real activity, in particular in the past? To illustrate their usefulness, we first compare the growth rates derived from the IP indexes collected from IFS paper volumes and digital records to the official GDP profiles of the United States and France; the only two countries for which official quarterly GDP are published and available from 1950 onwards. We find that the series are highly correlated. Over the whole sample period, we find correlations of 0.85 and 0.89 for France and the US respectively. Figure 1 and 2 below illustrate this finding over the period 1950-1972 (see Appendix A for the full sample). Overall we find that IP indexes are very good tracker of real activity (or GDP) in the past and in the present. In fact, as we shall see in Section 4, we find that all quantitative and qualitative results are unchanged in a shorter sample for which we are able to use GDP, rather than IP indexes.

Figure 1:
Figure 1:

France – IP vs GDP Growth (1950-1972)

Citation: IMF Working Papers 2016, 054; 10.5089/9781513564890.001.A001

Figure 2:
Figure 2:

USA – IP vs GDP Growth (1950-1972)

Citation: IMF Working Papers 2016, 054; 10.5089/9781513564890.001.A001

III- Empirical framework

To enhance the comparability of our results with the previous literature, in particular Kose, Otrok and Whiteman (2003, 2008), we rely on the same econometric methodology to assess the importance of a world business cycle. We estimate the following latent factor model:

yi,t=βiWftW+βiRegionft,jRegion+ɛi,t(1)

where yi,t is the y-o-y growth rate of IP in country i in quarter t, ftW is the (unobserved) factor affecting all countries in our sample at time t, ft,jRegion is the (unobserved) regional factor affecting all countries belonging to region j at time t, and βiW and βiRegion designate country-specific factor loadings measuring the responses of country i to the common world and regional factors respectively. Finally, εi,t is an unobserved country-specific residual factor.

Because we allow factors to follow AR processes, the model in (1) is in fact a dynamic latent factor model. More precisely, we assume that idiosyncratic factors follow an AR(p) process:

ɛi,t=ρi,1ɛi,t1+ρi,2ɛi,t2++ρi,pɛi,tp+ui,t(2)

where ui,tN(0,σi2) and E(ui,t, ui,ts) = 0 for s ≠ 0 and the world and regional factors follow the respective AR(q) processes:

ftW=ρ1ft1W+ρ2ft2W++ρqftqW+utW(3)
fj,tRegion=ρ1,jft1Region+ρ2,jft2Region++ρq,jftqRegion+uj,tRegion(4)

where utWN(0,σW2),uj,tRegionN(0,σj2), and E(utW,utsW)=E(uj,t,uj,ts)=0 for s ≠ 0.

Given that the factors are unobservable, standard regression methods do not allow for the estimation of the model. As a consequence, we rely on Bayesian techniques as in Kose, Otrok and Whiteman (2003, 2008) for estimation. As it is standard in the literature, as a first step, we normalize the sign of the factor/loadings by (i) restricting the loading on the world factor for the first country in our sample to be positive and (ii) restricting the loadings on the regional factor for one country in each region to be positive. Second, to normalize the scales, we assume that each of the factor variances is equal to 1. Note that these normalizations do not affect the qualitative results and simply allow the identification of the model. In addition, we use Bayesian techniques with data augmentation to estimate the parameters and factors in (1)-(4). This implies simulating draws from complete posterior distribution for the model parameters and factors and successively drawing from a series of conditional distributions using a Markov Chain Monte Carlo (MCMC) procedure. Posterior distribution properties for the model parameters and factors are based on 300,000 MCMC replications after 30,000 burn-in replications.

Following Kose, Otrok and Whiteman (2003), we use the following conjugate priors when estimating the model:

(βiW,βiRegion)N(0,I2)(5)
(ρi,1,,ρi,p)N(0,diag(1,0.5,,0.5p1)(6)
(ρ1,,ρp)N(0,diag(1,0.5,,0.5q1)(7)
(ρ1,j,,ρq,j)N(0,diag(1,0.5,,0.5q1)(8)
(σi2)IG(6,0.001)(9)

Where i=1,…,21 and IG denotes the Inverse Gamma distribution, implying a rather diffuse prior on the innovations variance. We also assume that the AR processes in (2)-(4) are stationary. In practice, in our implementation, we set the length of both the idiosyncratic and factor auto-regressive polynomials to 2. However, other (non-zero) values for p and q were tried with no substantial differences in the results. Similarly, reasonable deviations in priors did not generate any notable differences in the results presented below.

Besides estimating the factors, we are particularly interested in measuring the influence of the common world factor on the different countries in our sample. To do so, we compute variance decompositions, θiW, which denote the share of variance in the output of country i which is attributable to the world dynamic and is computed as follows:

θiW=(βiW)2var(ftW)/var(yi,t)(10)

Finally, models are estimated using two regions, namely: (i) Continental Europe (ii) Northern Europe (UK and Nordic countries), America and Asia. Note that although alternative regional decompositions could be used, the key results derived below are not sensitive to these decompositions as variance decompositions are invariant to the regional classification.

IV- Results

Section A first reports the results of the estimation for the full sample. Results for sub-samples are contrasted with the full sample results in section B.

A. World Factor and Variance Decompositions - Full Sample (1950 - 2014)

Figure 3 displays the mean posterior distribution of the world factor, along with 5 and 95% quantile bands. Figure 4 reports, for each country, the variance decompositions computed using (10) along with corresponding quantile bands. Several important findings emerge.

Figure 3:
Figure 3:

IP World Factor – Full Sample (1950 Q1-2014 Q4)

Citation: IMF Working Papers 2016, 054; 10.5089/9781513564890.001.A001

Figure 4:
Figure 4:

Variance Decompositions - Full Sample (1950 Q1-2014 Q4)

Citation: IMF Working Papers 2016, 054; 10.5089/9781513564890.001.A001

Note: Dots report the 5 and 95 percentiles of the posterior distribution.

First, besides being precisely estimated, the world factor captures the major economic events and cycles of the last 65 years, such as the oil shocks and the Global Financial Crisis (GFC). We postpone a complete discussion of the peaks and troughs identified by the global cycle to Appendix C, which is dedicated to a comparison of the peaks and troughs identified by the factor decomposition with a narrative analysis of the world cycle based on published IMF annual reports from 1950 to 2014. At this stage we only emphasize that world cycles described in IMF annual reports match almost exactly the turning points identified by our procedure.

Second, the variance decompositions point to a significant influence of a world business cycle over the postwar period. With only a few exceptions, the share of volatility in IP accounted for by the world dynamic is above one third in almost all advanced economies, with France, Germany, Italy, Belgium, Austria and Japan crossing the 50% threshold. In contrast, Ireland, Denmark, Norway, and emerging markets do not seem to be highly affected by the world cycle. On average we find that the world dynamic accounted for 37% of the volatility in Industrial Production of the 21 countries in our sample.

Several robustness checks are performed in Appendix B. First, we find that using quarterly GDP statistics (when available) does not change the shape or the strength of the world factor. Re-estimating the model on quarterly GDP statistics from 1973 onwards and comparing the results to those obtained using IP data, we find that the world factors and the variance decompositions are almost identical. In fact, our variance decompositions are also close to those estimated in Kose, Otrok and Whiteman (2003), which are computed using annual GDP data between 1960-1990 only, but for 60 countries. However, as the next section shows, such results conceal very important differences over time, even within countries.

B. World Factor and Variance Decompositions – Sub Sample Analysis

Has the strength of the world business cycle really changed over the last 65 years? To address this question, we re-estimate the factor model using four sub-periods: (i) the Bretton Woods period (1951 Q1-1971 Q4), the first common shock period (1972 Q1–1983 Q4), the Globalization period (1984 Q1-2006 Q4) and the second common shock period (2007 Q1-2014 Q4). We also refer to the first common shock period as the “oil shocks” period and the second common shock period as the Global Financial Crisis (GFC).

In line with Kose, Otrok and Whiteman (2008), we isolate periods of well-known global (or common) shocks because they inflate mechanically the amount of co-movement at the world level. The decade ranging from 1973 to 1983, in particular, features the demise of the BW system, two oils shocks and the widespread use of contractionary monetary policy in almost all advanced economies starting 1979. Similarly, the period from 2007 to 2014 was characterized by the GFC and the European debt crisis. Among other things, this procedure allows us to explore the ultimate object of interest of this paper, namely the strength of global co-movement under “normal” macroeconomic fluctuations. In that respect, both periods stand out as relatively tranquil periods of economic fluctuations. The Bretton Woods sample (1950-1971) is notable for its steady growth and stable business cycles dynamics, whereas the Globalization period (1984-2006) captures most of the Great Moderation. In addition, both periods are almost of equal length (21 vs. 22 years). On the other hand, these two periods also display very strong structural differences that should, at least intuitively, affect the nature of the world business cycle. In particular, the Bretton Woods period is marked by historically low financial and trade integration, a generalized fixed exchange rate regime and pro-cyclical monetary and fiscal policies, whereas the Globalization period is marked by a strong acceleration of financial and commercial integration and a wider use of flexible exchange rates and inflation targeting (Bordo 1993, Bordo et al. 1999).

Figure 5 plots the estimated world factor for the two sub-periods, along with the world factor for the full sample estimated in the previous section. Shaded areas refer to the periods of common shocks. Figures 6-8 report the variance decompositions for the respective sub-periods and contrast them with the results from the full sample. The key findings are as follows.

Figure 5:
Figure 5:

IP World Factor – Sub-Samples

Citation: IMF Working Papers 2016, 054; 10.5089/9781513564890.001.A001

Figure 6:
Figure 6:

Variance Decompositions by Country and Sub-Sample

Citation: IMF Working Papers 2016, 054; 10.5089/9781513564890.001.A001

Figure 7–
Figure 7–

Variance Decompositions Average – Full Sample vs. Sub sample

Citation: IMF Working Papers 2016, 054; 10.5089/9781513564890.001.A001

Note: averages are computed across all 21 countries for each sub-sample. Dots report the 5 and 95 percentiles of the posterior distribution.
Figure 8-
Figure 8-

Variance Decompositions - Bretton Woods vs. Globalization

Citation: IMF Working Papers 2016, 054; 10.5089/9781513564890.001.A001

First, as expected, we find that the estimates of the world factors are unchanged when we split the samples (Figure 5). Second, we find that the share of variance explained by the world dynamic varies significantly across sub-samples and countries. Over time, countries also experience significant changes in the way they co-move with the rest of the world.

In general, we find that the full sample result is significantly inflated by the two periods of common shocks (Figure 6 and 7). After averaging the variance decompositions across countries for each sub-sample, we find that the global dynamic was accounting for more than half of the variance of individual countries’ output during the first common shock period, and more than 65% during the second. More importantly, we find that once periods of global turmoil are excluded, the quantitative importance of the common factor is essentially the same between the Bretton Woods period and the Globalization period (Figure 7). Among other things, this result shows that when we use a longer sample with more countries, we do not find empirical support for the view that the recent period (starting in the mid-1980s), at least before the 2007 crisis, was associated with more co-movement at the global level (e.g Kose, Otrok and Whiteman (2008)). Interestingly, we find that this result is also corroborated by the narrative analysis of world economic cycles in Appendix C. Although average growth was higher during the 1950s and 1960s, our narrative analysis shows that Bretton Woods was marked by substantial business cycle fluctuations. As reminded by Swoboda (1983), the main motivation for a shift toward (partially) floating exchange rates in 1971 was to desynchronize individual countries from the rest of the world and to avoid the balance of payment constraints implied by the peg (despite capital controls).11

Another important finding is that although the average co-movement we observe is unchanged, the way countries co-move with the rest of the world has changed substantially between Bretton Woods and the Globalization period (Figure 8). Countries such as Austria, Germany, Finland and Sweden are found to co-move substantially with the rest of the world in both periods whereas a significant number of countries, such as Netherlands, Norway, Ireland, as well as most emerging markets, display a relatively low co-movement with the rest of the world in both periods. On the other hand, important changes between the two periods can be observed. The UK, Belgium, and Luxembourg have been significantly disconnected from the world dynamic during the Globalization period, whereas France, Italy and to a minor extent the US have been re-synchronizing with the world cycle. These dynamics (de-synchronization, re-synchronization) can be easily seen graphically and are found to be robust, once again, to the use of GDP in lieu of IP growth rates.12 Next section turns to a rationalization of this finding.

V- How does financial and trade integration affect co-movement with the rest of the world?

A. Baseline specification

Can trade and financial integration patterns explain why some countries change the way they co-move with the rest of the world? To answer this question, we run the following panel regression:

θi,tW=β1TradeInti,t+β2FinancialInti,t+β3Manufi,t+αi+dt+ɛi,t(11)t=1,2,3,4i=1,,20

Where θi,tW designates the share of variance accounted for by the global factor for country i in period t (plotted in Figure 6); Trade_Inti,t captures the level of trade integration of country i in period t; Financial_Inti,t captures the level of financial integration of country i in period t; Manufi,t measures the share of Manufacturing as a percentage of GDP; αi captures country-fixed effects and dt are time dummies capturing time fixed effects. Period 1 designates the Bretton Woods period; period 2 the first common shock period (or oil shocks); period 3 the Globalization period; and period 4, the GFC period. In practice, we measure trade integration by computing the average ratio of Exports plus Imports to GDP over each sub-sample for each country.13 Similarly, we measure financial integration using the average ratio of Foreign Assets plus Foreign Liabilities to GDP over each sub-sample for each country. We also include the share of manufacturing (as % of GDP). Although we show that this variable is not significant, it serves as a control, as one might expect countries with strong manufacturing sector to co-move more with the rest of the world by construction. Summary statistics and sources about variables used in the estimation are reported in Appendix A.

As expected, the inclusion of both time and country fixed effects helps us controlling for the presence of (i) obvious common shocks scaling up all countries at the same time, in particular in period 2 and 4 and (ii) country fixed effects, since some countries always co-move less (or more) than others (e.g Norway or Germany). Although this implies that we cannot say much about the between-variation in the panel, the within-variation is clearly the object of interest in our paper. More importantly, the inclusion of fixed effects allows us to assess, in a rigorous way, the importance of trade and financial integration in affecting co-movement patterns. To date, important contributions have limited their analysis to cross-section analysis, or to panel estimators without the inclusion of fixed effects, implying potentially a significant bias in the results (see in particular Kose, Prasad and Terrones (2003)).

Table 1 reports the result of the panel regressions, distinguishing between results for the full sample or for advanced countries only. Standard errors are clustered by country in all estimations. Columns 1 and 2 present results for the baseline regression in (11). Columns 3 to 6 present results when financial integration and trade integration measures are interacted with time dummies to investigate the presence of potential asymmetries during periods of common shocks. As we shall see, this allows us to isolate the effect of financial integration in normal times from its effect in crisis times.

Table 1–

Panel Regression Results

This table presents the results of the baseline panel estimations with fixed effects. In all specifications the dependent variable is the share of variance accounted for by the global factor for country i in period t (plotted in Figure 6). Columns (1) and (2) present results for the baseline regression in (11). Columns (3) to (6) present results when financial integration and trade integration measures are interacted with time dummies to investigate the presence of potential asymmetries during common shocks. Because of the scale of Luxembourg’s financial integration (25000% of GDP at the end of our sample), we leave the country out of estimations, so that the number of observations is 80 (and not 84).

article image
Standard errors in parentheses *p<0.10 ** p<0.05 *** p<0.01

Although the number of observation used for estimation is small, the panel regression yields striking results, some of which are well connected to recent empirical contributions. First, we find that trade integration tends to increase co-movement with the rest of the world. However, this effect depends crucially on the presence of emerging markets in our sample. This suggests that increasing trade integration has a non-linear effect, affecting strongly the strength of co-movement at a lower stage, but less so when countries have already reached a high degree of trade openness. Second, we find that financial integration has, on average, a negative impact on the synchronization with the business cycle (Columns (1) and (2)), and that this effect is true even when emerging markets are dropped. However this average effect conceals an asymmetric effect: when using the interaction term, the negative impact of financial integration more than triples and the precision of the estimate improves (Columns (5) and (6)). Quantitatively, we find that increasing financial integration by 10% decreases co-movement with the rest of the world by 1%. As a result, a 200% increase - the average change over the sample of advanced countries - is associated with a decrease in co-movement of 20 to 25%, depending on the sample considered. On the other hand, a strong positive impact appears in financial crisis periods. In that case, a 200% increase in financial integration would imply an increase in co-movement of 14% with the rest of the world. However, the total net effect remains negative. This result is also robust to the exclusion of emerging markets from the sample. Third and finally, we do not find evidence of other asymmetries. In particular, financial integration does seem to matter when real (rather than financial) shocks hit the world economy. Similarly, trade integration does not have a separate effect during periods of global shocks (either real or financial).

In normal times, the negative impact of financial integration can be seen graphically in Figure 9, which plots the change in financial integration against the change in co-movement with the world business cycle between Bretton Woods and the Globalization period for advanced economies. This figure shows clearly that countries which have accelerated their financial integration (UK, Belgium, Sweden) have de-synchronized themselves from the global dynamic, whereas countries with a slower pace of financial integration have been moving in the opposite direction (France, Italy and the US).

Figure 9–
Figure 9–

Change in Financial Integration vs. Change in Co-movement

This figure plots the change in financial integration against the change in co-movement with the world business cycle between Bretton Woods and the Globalization period, for the sample of advanced economies. The negative slope illustrates the negative impact of financial integration on co-movement in “normal” times. Note that Ireland, which stands out given the size of its financial integration, is not driving any the key results. In fact, dropping Ireland increases the precision and significance of the coefficients of interest (financial integration and its interaction) without altering any of the other results.

Citation: IMF Working Papers 2016, 054; 10.5089/9781513564890.001.A001

These results are well connected to other empirical contributions, in particular Kalemli-Ozcan, Papaioannou and Perri (2013), Kalemli-Ozcan, Papaioannou and Peydro (2013).14 Contrary to most empirical studies that reported a positive link between financial integration and country-pairs output synchronization, these studies identified (i) a strong negative effect of banking integration on output synchronization once global shocks and country-pair heterogeneity are controlled for and (ii) a positive impact of financial integration on output co-movement during financial crisis. Our results bring further support to these results by in that we find a strong and asymmetric impact of financial integration on co-movement in the long run. Using a long sample also shows that the asymmetric effect of financial integration exists only in the presence of global financial shock, but not global real shocks. Overall, this suggests that financial integration (i) disconnects countries from the global dynamic in tranquil times (ii) reconnects them during global financial shocks and (iii) does not play any (marginal) role during global real shocks.

B. Extensions and Robustness

This section performs a number of extensions and robustness checks. We report and discuss each of them below, connecting the results to the existing literature. Results are reported in Appendix B. Overall, we find that the significance and magnitude of the key results are almost unchanged to the addition of other controls.

(i) Trade integration measurement

A possible explanation for the insignificance of the trade integration variable in the full sample is that gross trade openness, measured by the ratio of gross exports plus imports to GDP, is a poor proxy of actual trade integration. Ultimately, valued-added trade, rather than gross exports, matter for growth, and therefore for growth co-movement (Duval et al, 2015). We find, however, that using value-added trade data does not change any of the results.15 In particular, the significance of trade integration relies on the presence of EMs in the sample. This is not surprising since working with value-added trade data change levels of trade integration for each country, but does very little to the changes in trade integration at the country level over time. Since we are mainly interested in explaining the within-variation in the panel, switching to value-added trade data has, therefore, very little influence on the key results.

(ii) Specialization

A number of empirical contributions have suggested that specialization might also affect the pattern of co-movement; both through direct and indirect effects.16 In particular, both trade and financial openness might, at least in theory, induce greater specialization, which in turn might affect co-movement, although in a non trivial way (see Imbs, 2004 for a thorough discussion). To control for the potential impact of specialization we construct, for each country, a standard Herfindahl index of concentration defined as follows: Hi,t=Σ110si,t, where si,t designates the share of sector i (out of 10 sectors) in the GDP of a given country at timet.17 These indexes are then normalized to range between 0 and 1. Data from both Groningen Growth and Development Center (GGDC) and EU-KLEMS are used to compute the index. We find that including the specialization variable does not affect any of the results, as (i) trade integration remains insignificant and (ii) the magnitude and significance of the coefficients attached to the financial integration variables (level and interaction) are unchanged.

(iii) Exchange rate regime

Finally, we control for the exchange rate regime using the (updated) exchange rate classification compiled by Ilzetzki, Reinhart and Rogoff (2014). In practice, for each country in our sample, we average the fine classification (ranging from 1 to 14) over each sub-period. By construction, a higher value of the index indicates greater exchange rate flexibility over that particular period of time. Because Euro area countries are classified as fixed (a value of 1) in the original database, we also perform estimations when recoding Euro countries as freely floating (a value of 13). In both cases, we find that the exchange rate regime has not been a major determinant of the synchronization of a country with world output, or that, if exchange rate regimes did matter, their effects are captured by the fixed effects included in the estimations.

VI- Conclusion

The global financial and economic crisis reignited the debate on the importance of globalization in shaping output co-movement which has accompanied the long rise of financial and trade integration since the early 1980s (Swodoba 1983, Rodrik 1997, Kose, Prasad, Rogoff Wei, 2006). This paper contributes to this debate by measuring the strength of global co-movement in a new quarterly sample of 21 countries since 1950. Among other things we find that using a longer time series and extending the sample changes considerably the empirically results, as well as the policy messages. After isolating four periods (1950 – 1971; 1972 – 1983; 1984 – 2006; 2007 – 2014), we find that the Bretton Woods era was not a period of low co-movement compared to the period between 1984 and 2006, which saw a massive rise in both trade and financial of globalization. At the country level however, we find a robust negative link between financial openness and the correlation with the world business cycle, although this effect was dampened after 2007. Inter alia, our results suggest that a low level of integration does not imply a low level of co-movement in the economic system. Contrary to what Rodrik (2002) and others have suggested, we find no evidence that reduced financial and trade integration disconnected countries from foreign shocks or left more room to national economic management in the 1950s and 1960s than during the period of 1984-2006. Our results suggest that reducing financial integration would only decrease co-movement in the face of global financial shocks, but would probably increase co-movement in normal times, when the world is dominated by idiosyncratic shocks.

Has Globalization Really Increased Business Cycle Synchronization?
Author: Eric Monnet and Mr. Damien Puy
  • View in gallery

    France – IP vs GDP Growth (1950-1972)

  • View in gallery

    USA – IP vs GDP Growth (1950-1972)

  • View in gallery

    IP World Factor – Full Sample (1950 Q1-2014 Q4)

  • View in gallery

    Variance Decompositions - Full Sample (1950 Q1-2014 Q4)

  • View in gallery

    IP World Factor – Sub-Samples

  • View in gallery

    Variance Decompositions by Country and Sub-Sample

  • View in gallery

    Variance Decompositions Average – Full Sample vs. Sub sample

  • View in gallery

    Variance Decompositions - Bretton Woods vs. Globalization

  • View in gallery

    Change in Financial Integration vs. Change in Co-movement

    This figure plots the change in financial integration against the change in co-movement with the world business cycle between Bretton Woods and the Globalization period, for the sample of advanced economies. The negative slope illustrates the negative impact of financial integration on co-movement in “normal” times. Note that Ireland, which stands out given the size of its financial integration, is not driving any the key results. In fact, dropping Ireland increases the precision and significance of the coefficients of interest (financial integration and its interaction) without altering any of the other results.