Measuring the Connectedness of the Global Economy☆
Introduction
Globalisation is the process of increasing interdependence among entities in the global economy. In layman’s terms, the world is becoming ‘smaller’ and the distinction between national, regional, and global issues less well-defined. Established views of the benefits of globalisation in relation to openness, liberalisation, and development have been challenged in light of the Global Financial Crisis (GFC), which has drawn renewed attention to the risks posed by aspects of financial globalisation (Mishkin, 2011). Recent research into financial connectedness has reshaped our understanding of systemic linkages and has shed new light on the identification and supervision of institutions that are ‘too big’ or ‘too connected’ to fail (IMF, 2009). However, our understanding of international macroeconomic linkages has not advanced to the same degree, and the network structure of the global economy remains poorly understood. We address this lacuna by developing an innovative and highly adaptable graph-theoretic framework to evaluate macroeconomic connectedness in a wide class of multi-country and global macroeconomic models.
International linkages may arise through diverse channels, including financial linkages, trade linkages, and relative price changes (Dees et al., 2007). We therefore consider macroeconomic connectedness to be an intrinsically multi-dimensional concept. However, the existing literature has largely focused on a single aspect of macroeconomic connectedness, namely the apparent convergence of business cycles across countries. A degree of consensus has emerged around the notion of a global business cycle that induces some common behaviour in national business cycles (e.g. Hirata et al., 2013, Kose et al., 2003, Kose et al., 2008 among others). Much of this research has modelled the global business cycle as a latent factor, an approach that is attractive by virtue of its parsimony. This is an important consideration in light of the relatively short samples and low sampling frequencies of much macroeconomic data. Indeed, the dimensional constraint was a key motivation underlying Croux et al.’s (2001) development of a synthetic measure of synchronisation across countries/regions that is defined in the frequency domain as opposed to the time domain. Their ‘cohesion’ measure can be used to trace the comovement of multiple time series, and its application to European sovereign and US state-level business cycles further supports the synchronisation hypothesis.
Although Croux et al. stress that estimating vector autoregressive (VAR) models may be ‘problematic when the number of time series is large’ (p. 232), subsequent innovations in the estimation of large multi-country VAR models—including panel/global VAR (Pesaran et al., 2004), factor augmented VAR (Bernanke et al., 2005), and large Bayesian VAR (De Mol et al., 2008)—have relaxed this constraint. Consequently, it is now possible to estimate large macroeconometric models in the time domain. Canova et al. (2007) were among the first to apply these techniques to the analysis of business cycle convergence, estimating a Bayesian panel VAR model that again highlights the importance of a global cycle relative to idiosyncratic effects.
In principle, a sufficiently detailed multi-country system provides a route to model the global business cycle as an observable process defined by the interaction of the countries comprising the model, without recourse to latent factors. Consequently, sophisticated multi-country models may provide a new perspective on the issues of globalisation and regionalisation that have emerged prominently in the existing literature (e.g. Hirata et al., 2013). A further advantage of these sophisticated models is that, by easing the dimensional constraint, they are able to accommodate a far greater wealth of interactions among countries and regions than was previously possible. This opens a new avenue to study macroeconomic connectedness in a truly multi-dimensional sense, as opposed to focusing simply on business cycle convergence.
Unfortunately, the development of techniques for global macroeconometric modelling has yet to be met by concomitant advances in techniques for the analysis of the linkages embedded in such models. Even as progress in the estimation of large VAR models has mitigated the curse of dimensionality associated with the limits imposed by the range and frequency of existing macroeconomic data sets, so it has introduced a new curse of dimensionality associated with one’s ability to process the model output. Consequently, the analysis of such models tends to be highly selective and does not properly illuminate the intricate network of linkages among variables in the system.
Significantly, unlike much of the recent literature on financial connectedness, the existing business cycle literature is mostly not grounded in network theory (Diebold & Yilmaz, 2015). Yet network models provide a natural vehicle for the analysis of complex systems—such as the global economy—that are composed of many interconnected entities. Our key contribution is therefore to unite modern techniques for global macroeconomic modelling with state-of-the-art developments in financial network analysis. Leading examples of financial network models include Billio et al. (2012) and Diebold and Yilmaz, 2009, Diebold and Yilmaz, 2012, Diebold and Yilmaz, 2014.1 In this literature, financial institutions are characterised as nodes within a network. Analysing the network topology provides a means to identify systemically important institutions and to study the propagation of shocks. Billio et al. consider a Granger causal network, while Diebold and Yilmaz develop a weighted directed network (or ‘connectedness table’) based on the decomposition of the forecast error variance of a VAR. The forecast error variance decomposition (FEVD) measures the proportion of the variance of the -steps-ahead forecast error of variable that is due to innovations to variable . Consequently, it has a natural bilateral interpretation that lends itself to network analysis. Relative to a Granger causal network, the FEVD approach has the advantage that it fully accounts for contemporaneous effects, and it directly measures not only the direction but also the strength of linkages among nodes in the network.
As originally conceptualised, the framework of Diebold and Yilmaz, 2009, Diebold and Yilmaz, 2012, Diebold and Yilmaz, 2014 is not well suited to the analysis of large and sophisticated multi-country models. Their approach operates at two extremes: (i) complete aggregation, where the bilateral linkages in an variable model are aggregated into a single spillover index; and (ii) no aggregation, where the bilateral linkages are either studied individually or on a variable-by-variable basis. However, modern multi-country macroeconomic models can contain many variables (often 100 or more). In such cases, as becomes large, it will rapidly become infeasible to study the network topology on a disaggregated basis, and the amount of detail obscured by relying on a single spillover index to summarise the connectedness of the system will increase. Moreover, in a multi-country model with multiple variables per country—a setting that is typical of sophisticated global models and that is central to our notion of multi-dimensional macroeconomic connectedness—one may wish to analyse linkages among countries or regions in the aggregate rather than among individual variables. The Diebold and Yilmaz approach does not provide any straightforward way to achieve this. We therefore require a flexible generalisation of the Diebold and Yilmaz connectedness measures.
Our solution is to introduce intermediate levels of aggregation, yielding a framework for the construction of generalised connectedness measures (GCMs). Our approach is to gather the variables in the model into a set of groups and then to aggregate the connectedness matrix according to this group structure. Significantly, aggregation occurs after estimation, so the underlying model is unaffected by the choice of aggregation scheme, facilitating comparisons across different levels of aggregation. The only restrictions on the allocation of variables to groups are that each group must contain one or more variables, no variable may enter more than one group, and every variable must be included in the analysis. For example, in a simple model with five variables for each of 10 countries (i.e. ), one may define 10 groups of five variables, with one group corresponding to each country. By aggregating the connectedness matrix accordingly, one may evaluate the connectedness of the system at the country-level. Similarly, suppose that the set of five variables for each country contains two financial variables and three real variables. In this setting, one could define a group containing the 20 financial variables in the system and another containing the 30 real variables in order to evaluate real–financial linkages in the global economy.
Our approach is sufficiently flexible that it can accommodate a very wide variety of aggregation schemes. Consequently, it can shed light on the interactions among a multitude of diverse entities in the global economy. As such, our approach provides a more comprehensive framework for the study of macroeconomic connectedness than the existing literature on business cycle synchronisation. Moreover, our use of aggregation mitigates the processing constraints encountered when analysing large and detailed macroeconomic models. Consequently, our framework unlocks the potential of such models to study macroeconomic connectedness in considerable breadth and detail.
We apply our technique to an updated version of the macro-financial global VAR model developed by Greenwood-Nimmo, Nguyen, and Shin (2012, hereafter GNS), which is initially estimated using data prior to the GFC to provide a benchmark. The model contains 169 endogenous variables covering 25 countries/regions that account for the large majority of global trade and output. In the absence of a precedent, we begin by carefully analysing the sensitivity of the estimated network to the choice of forecast horizon. Our GCMs typically converge to their long-run values after 3–5 quarters. Consequently, we focus on the four-quarters-ahead horizon, which aligns our analysis neatly with the medium-term forecast horizon of many central banks.
We exploit the conceptual links between a country’s macroeconomic connectedness, its dependence on (or openness to) overseas conditions, and the extent of its economic influence to draw out several key results. We identify the US, China, Brazil, and the Eurozone as the world’s most influential economies. Although the US acts as the principal driver of global conditions, as in Diebold and Yilmaz (2015), the presence of additional centres of economic activity is consistent with the rise of China as an economic power and with the regionalisation documented by Hirata et al. (2013). This phenomenon is discernible in the baseline specification of the GNS model, but it is particularly evident in robustness tests where time-varying trade weights are used to construct the global VAR system. The high degree of US influence relative to that of other economies that have experienced crises in our sample period provides an intuitive explanation of the global impact of the GFC compared to the local and regional effects of Black Wednesday in the UK, the 1997 Asian financial crisis, and the collapse of the Japanese bubble earlier in the same decade. Our results also reveal that the countries that are most dependent on external conditions include Canada, Singapore, and Switzerland, all of which are strongly affected by conditions within their respective free trade areas. We show that analysing a country’s relative dependence and influence provides an elegant summary of its role in the global economy, ranging from small open economies at one extreme (high dependence, low influence) to large dominant economies at the other (high influence, low dependence).
Having established a pre-GFC benchmark, we recursively update our estimation sample over the GFC period. The results reveal a marked increase in spillovers originating from the US financial system that coincides with the collapse of Lehman Brothers. Further analysis shows that the US financial shock was rapidly and strongly passed through to both nominal variables and to real economic magnitudes, with real exports and imports being particularly strongly affected.
Aside from the connectedness literature, our paper is most closely related to the panel VAR approach of Canova et al. (2007) and the dynamic factor model developed by Hirata et al. (2013). Both of these papers distinguish between global, regional, and local effects. Canova et al. emphasise the role of a global factor influencing G7 business cycles, while Hirata et al. stress that regional factors have come to play a prominent role since the mid-1980s, while the role of global factors has diminished. These observations furnish an a priori case for the development of new techniques such as ours, that offer a new perspective on the nature of international macroeconomic linkages.
This paper is organised as follows. Section 2 introduces the concept of connectedness in VAR systems and provides a detailed derivation of our GCMs. Section 3 introduces an updated version of the GNS global VAR model, which forms the basis of our empirical analysis. The results of GCM analysis of the linkages embedded in this model are presented in Section 4, while Section 5 concludes. Further details regarding the data set, the model setup, and several robustness exercises are contained in a separate Technical Annex.
Section snippets
Measuring economic connectedness
Following Diebold and Yilmaz, 2009, Diebold and Yilmaz, 2012, Diebold and Yilmaz, 2014, the connectedness measures that we develop are based on the FEVD of a th order VAR for the vector of endogenous variables . This approach is founded on the notion that the share of the forecast error variance (FEV) of variable explained by shocks to variable provides a directional measure of the association between these variables. An appealing feature of this framework is that the FEVD is
The GNS global model
We apply our GCM framework to an updated version of the global cointegrating VAR model developed by GNS, which, in turn, owes a significant intellectual debt to Dees et al. (2007). This model provides an ideal basis for the evaluation of macroeconomic connectedness, as it is a large system composed of multiple countries that collectively account for the majority of global economic activity. Furthermore, the model includes a range of key macroeconomic and financial indicators relating to real
Economic connectedness prior to the GFC
The first step in our analysis is to select an appropriate forecast horizon. In the absence of an uncontroversial precedent, we start by studying the variation in country-level connectedness over horizons quarters, as recorded in Fig. 1. In the subfigure for the th country, the upper panel plots the to contribution () as a red line and the from contribution () as a blue line. The net connectedness () is shown by the shaded region: red shading indicates a net
Concluding remarks
We developed a technique to measure macroeconomic connectedness in the global economy. Our GCM framework is an innovative generalisation of the FEVD-based connectedness methodology advanced by Diebold and Yilmaz (2009, 2012, 2014) for the study of financial networks. Our principal innovation is to introduce a new stratum between the level of individual variables and the level of system-wide aggregates. This allows us to measure connectedness between countries, regions, or any arbitrary group of
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (38)
- et al.
The great recession: US dynamics and spillovers to the world economy
Journal of Banking & Finance
(2012) - et al.
Econometric measures of connectedness and systemic risk in the finance and insurance sectors
Journal of Financial Economic
(2012) - et al.
Similarities and convergence in G-7 cycles
Journal of Monetary Economics
(2007) - et al.
Off the cliff and back? Credit conditions and international trade during the global financial crisis
Journal of International Economics
(2012) - et al.
From the financial crisis to the real economy: Using firm-level data to identify transmission channels
Journal of International Economics
(2012) - et al.
Forecasting using a large number of predictors: Is Bayesian shrinkage a valid alternative to principal components
Journal of Econometrics
(2008) - et al.
Better to give than to receive: Predictive directional measurement of volatility spillovers
International Journal of Forecasting
(2012) - et al.
On the network topology of variance decompositions: Measuring the connectedness of financial firms
Journal of Econometrics
(2014) Business cycle transmission from the US to Germany – a structural factor approach
European Economic Review
(2007)- et al.
Measuring monetary policy: A factor augmented autoregressive (FAVAR) approach
Quarterly Journal of Economics
(2005)
The initial impact of the crisis on emerging market countries
Brookings Papers on Economic Activity
The dynamic effects of aggregate demand and supply disturbances
American Economic Review
Are all business cycles alike?
Techniques for testing the constancy of regression relationships over time
Journal of the Royal Statistical Society. Series B.
Collective risk management in a flight to quality episode
The Journal of Finance
China’s emergence in the world economy and business cycles in Latin America
A measure of comovement for economic variables: Theory and empirics
The Review of Economics and Statistics
Exploring the international linkages of the euro area: A global VAR analysis
Journal of Applied Econometrics
Measuring financial asset return and volatility spillovers, with application to global equity markets
The Economic Journal
Cited by (41)
Imported financial risk in global stock markets: Evidence from the interconnected network
2024, Research in International Business and FinanceThe topological structure of panel variance decomposition networks
2024, Journal of Financial StabilityInterconnectedness between stock and credit markets: The role of European G-SIBs in a multilayer perspective
2024, Journal of International Financial Markets, Institutions and MoneySpillover among Sovereign Credit Risk and the Role of Climate Uncertainty
2024, Finance Research LettersMeasuring the G20 stock market return transmission mechanism: Evidence from the R<sup>2</sup> connectedness approach
2024, International Review of Financial AnalysisInterconnected networks: Measuring extreme risk connectedness between China's financial sector and real estate sector
2023, International Review of Financial Analysis
- ☆
We are grateful for the insightful comments of Heather Anderson, Sang-Don Bu, Efrem Castelnuovo, Woonkyu Choi, John Hunter, Yong-Min Kim, Vance Martin, Faek Menla Ali, Kostas Mouratidis, Adrian Pagan, Hail Park, Hashem Pesaran, Kalvinder Shields, Peter Smith, Ron Smith, Benjamin Wong, and Tomasz Woźniak. This work has benefited greatly from the stimulating discussion of delegates at the 21 Annual Symposium of the Society for Nonlinear Dynamics and Econometrics (Milan, March 2013), the Annual Conference of the Scottish Economic Society (Perth, April 2013), the BMRC-QASS Conference on Macro and Financial Economics (London, May 2013), the Econometrics of Social Interaction Symposium (York, May 2013), the 9th International Symposium on Econometric Theory and Applications (Seoul, July 2013), the Econometric Society Australasian Meeting (Sydney, July 2013), the Asian Meeting of the Econometric Society (Singapore, August 2013), the 45th Annual Conference of the Money, Macro and Finance Research Group (London, September 2013), the Econometrics Workshop at Victoria University of Wellington (November 2013), the 11th World Congress of the Econometric Society (Montreal, August 2015), and the 69th European Meeting of the Econometric Society (Geneva, August 2016), as well as the many detailed comments raised by seminar participants at Deakin and Monash Universities, the Universities of Lecce, Melbourne, Seoul, Sheffield, Sogang, Utah, Yonsei and York, the Melbourne Institute: Applied Economic and Social Research, the Sheffield Methods Institute, the Bank of Korea, and the Reserve Bank of New Zealand. Greenwood-Nimmo gratefully acknowledges financial support from the University of Leeds Seedcorn Fund, United Kingdom and the kind hospitality of the Universities of Sheffield and York during visits in 2013–5 when a substantial part of this work was conducted. Nguyen acknowledges funding from the Faculty of Business and Economics at the University of Melbourne, Australia . Shin acknowledges the hospitality of the Melbourne Institute: Applied Economic and Social Research and the Bank of Korea during research visits in the period April–July 2014. The usual disclaimer applies.