CENTRE FOR EMEA BANKING, FINANCE & ECONOMICS. Business Cycles, International Trade and Capital Flows: Evidence from Latin America

Transcription

1 CENTRE FOR EMEA BANKING, FINANCE & ECONOMICS Business Cycles, International Trade and Capital Flows: Evidence from Latin America Guglielmo Maria Caporale Alessandro Girardi Working Paper Series No 12/38

2 Business Cycles, International Trade and Capital Flows: Evidence from Latin America Guglielmo Maria Caporale* Brunel University, London, CESifo and DIW Berlin Alessandro Girardi Italian National Institute of Statistics November 2012 Abstract Tis paper adopts a flexible framework to assess bot sort- and long-run business cycle linkages between six Latin American (LA) countries and te four largest economies in te world (namely te US, te Euro area, Japan and Cina) over te period 1980:I-2011:IV. Te result indicate tat witin te LA region tere are considerable differences between countries, success stories coexisting wit extremely vulnerable economies. Tey also sow tat te LA region as a wole is largely dependent on external developments, especially in te years after te great recession of 2008 and Te trade cannel appears to be te most important source of business cycle co-movement, wilst capital flows are found to ave a limited role, especially in te very sort run. Keywords: International Business Cycle, Latin America, VAR models, Trade and Financial Linkages. JEL Classification: C32, E32, F31, F41. * Corresponding autor. Centre for Empirical Finance, Brunel University, West London, UB8 3PH, UK. Tel.: +44 (0) Fax: +44 (0) Guglielmo-Maria.Caporale@brunel.ac.uk 1

3 1. Introduction It is well known tat macroeconomic volatility generates bot economic and political uncertainty wit detrimental effects on investment and consumption plans and, ultimately, future economic growt (Acemoglu et al., 2003) and aggregate welfare (Atanasoulis and van Wincop, 2000). Tere is terefore considerable interest, among academics as well as policy-makers, in sedding ligt on te sources of output fluctuations, especially in te new economic environment caracterised by a muc greater role played by emerging market economies. Stronger international financial and trade linkages ave affected te relative importance of external, regional and country-specific factors in driving national business cycles, wit implications for te design of effective stabilisation policies (Fatàs and Miov, 2006). Economic teory does not provide unequivocal predictions: stronger linkages could result eiter in a iger or a lower degree of business cycle co-movement depending on weter or not demand- and supply-side (as well as wealt) effects dominate over increased specialisation of production troug te reallocation of capital (Baxter and Kouparitsas, 2005; Imbs, 2006; Kose et al. 2003, 2012). Tis cannot be establised ex-ante: it is essentially an empirical question. A knowledge of cross-country spillover effects is especially relevant for emerging countries because of teir iger degree of volatility compared to more mature economies. According to Loayaza et al. (2007), bot internal and external factors explain wy emerging economies are so volatile: i) te instrinsic instability induced by te development process itself; ii) te lack of effective mecanisms (suc as well functioning financial markets and proper stabilisation macroeconomic policies) to absorb external fluctuations; iii) te exposure to exogenous socks in te form of sudden capital inflows/outflows and/or large canges in te international terms of trade. Te Latin American (LA) economies in particular ave experienced a remarkable sequence of booms and busts in te last tree decades. After te debt crisis of te 1980s, most countries in te region benefited from uge capital inflows (wit a resulting ig growt rate) until te Russian crisis in te late nineties led to teir sudden drying up; ten in te early years of te following decade iger liquidity, a dramatic rise in commodity prices and low risk premia created a particularly favourable macroeconomic and financial environment in te region and generated again robust growt (Österolm and Zettelmeyer, 2

4 2007; Izquierdo et al., 2008); terefore te question as been asked weter tere as been a decoupling of te business cycle in te industrialised countries and te LA region respectively, te latter aving become an increasingly autonoumos source of growt for te world economy. Most of te existing literature on international business cycles focuses on te industrial countries, specifically te Group of Seven (Bagliano and Morana, 2010), Europe (Artis et al., 2004), East Asia and Nort America (Helbling et. al, 2007), Western Europe and Nort America (Mody et al., 2007). Tere are, owever, a few studies on LA differing in te set of countries examined and te adopted econometric metodology, and providing mixed evidence. Focusing on te average beaviour of te aggregate LA region, Izquierdo et al. (2008) find tat external socks account for a significant sare of te variance of regional GDP growt. Similar results are reported by Österolm and Zettelmeyer (2007) for bot te LA region as a wole and its individual countries, wilst Aiolfi et al. (2010) identify a sizeable common component in te LA countries business cycles, suggesting te existence of a regional cycle. By contrast, Hoffmaister and Roldos (1997) conclude tat domestic country-specific aggregate supply socks are by far te most important source of output fluctuations in te LA countries. Kose et al. (2003) find tat countryspecific factors explain te largest sare of te variance of output in tese countries, wit te exception of Bolivia, were te world component is more important tan te regional and country-specific ones. Finally, Bosci and Girardi (2011) report tat domestic factors account for by far te largest sare of domestic output variability in six major LA countries, and tat regional factors are more important tan international ones. None of te aforementioned papers includes te great recession of 2008 and By contrast, te present study examines te last tree decades to assess te relative role of country-specific, regional and external socks in explaining business cycle fluctuations in six major LA economies (namely, Argentina, Brazil, Cile, Mexico, Peru and Venezuela) and te LA region as a wole, wit te aim of sedding ligt on te role of bilateral trade flows and financial linkages in business cycle co-movements between te LA region and its main economic partners. Building on te work of Diebold and Yilmaz (2012), we specify a very flexible empirical model enabling us to analyse te propagation of international business cycles witout any restrictions on te 3

5 directions of sort- and long-run spillovers or te nature of te propagation mecanism itself. Using quarterly data from 1980:I to 2011:IV, we document a ig degree of eterogeneity among te LA countries: wile Argentina, Mexico and Peru appear to be increasingly dependent on external developments as a result of te great recession, Venezuela seems to be influenced mainly by te LA regional business cycle, wit only Brazil sowing a decreasing role of te external factors. As for te LA region as a wole, our results indicate tat it can be caracterised as a small open economy largely dependent on external developments. Tis applies especially to te te years following te great recession of 2008 and 2009, contradicting te so-called decoupling ypotesis. In particular, our findings imply tat te goods trade cannel is te most important source of tese linkages. Capital flows also affect business cycle co-movements, but teir role is limited, especially in te very sort run. Te disaggregate analysis focusing on teir components (debt, portfolio equity and foreign direct investment flows) reveals a negative effect of portfolio equity flows on te degree of business cycle syncronisation, as predicted by standard international real business cycle models wit complete markets. By contrast, sort-term capital and foreign direct investment flows reinforce in te sort run te role of te trade cannel and make te LA region more vulnerable to socks from abroad, consistently wit recent empirical evidence (e.g., Imbs, 2006, 2010). Te layout of te paper is as follows. Section 2 describes te metodology used to assess te propagation mecanism of international business cycles. Section 3 describes te data and presents te empirical results based on te forecast error variance decompositions for te individual LA countries and te LA region as a wole. Section 4 provides evidence on te role of financial and trade linkages. Section 5 offers some concluding remarks. 2. Te Metodology 2.1. Te Empirical Framework We focus on output growt in order to analyse te dynamic relationsips between te LA region and te rest of te world as well as te intra-area linkages among countries belonging to tat region. Given te increasing degree of integration of te global economy, it is essential to consider possible linkages wit a 4

6 number of foreign economies. It is equally important to allow for time variation, since a fixed parameter model is not likely to capture possibly important canges in te business cycle propagation mecanisms resulting from globalisation. 1 Consequently, te modelling approac cosen ere differs from previous ones in two ways. First, it is flexible enoug to accommodate possible nonlinear sifts in te propagation of international business cycles; second, it is based on analysing linkages wit te output growt rate of various economies outside te LA region rater tan a number of macroeconomic variables for a single foreign country (typically te US). Terefore we include te US as te main driving force beind business cycles co-movements in te LA region (see te literature on te US backyard, e.g. Amed, 2003; Canova, 2005; Caporale et al., 2011), but also te Euro area because of its istorical trade linkages wit te LA region, as well as Japan (given te financial linkages documented by Bosci, 2012, and Bosci and Girardi, 2011) and Cina, wose trade linkages wit te LA region ave become muc stronger in recent years (Cesa-Bianci et al., 2011). As in Diebold and Yilmaz (2012), te econometric framework is based on te following covariancestationary Vector AutoRegression (VAR) model y ( k ) p ( k ) ( k ) ( k ) t = Γ y u j= 1 j t j + t (1) in its moving average representation y (2) ( k ) ( k ) ( k ) t = Ψ j 0 j u = t j were ( k ) p ( k ) ( k ) i j= 1 i i j Ψ = Γ Ψ, p is te order of autoregression, te vector y includes te n endogenous variables of te system, t = 1,..., T indexes time, and e is te vector of residuals, wit E u =0, ( ) [ k t ] E[ u, u ] = Σ if s = t and ( k ) ( k ) ( k ) t s u E u u =0 oterwise. All elements refer to te generic k -t ( k ) ( k ) [ t, s ] 1 Including additional variables (suc as interest rates, excange rates, consumption or investment) for a wide range of countries would result in a system wose dimensions would not be manageable in te standard Vector AutoRegression (VAR) approac followed ere. Even advanced econometric approaces, suc as te Global VAR (see Cesa-Bianci et al., 2011, and Bosci and Girardi, 2011) or dynamic factor models (as in Kose et al. 2012, among oters), would not be a fully satisfactory modelling strategy since tey belong to te class of (linear) time-invariant models. 5

7 estimation sample wit window size of θ T observations, so tat if θ = T ten k = 1, wilst if θ < T, model (1) involves k = T θ + 1 different rolling estimates, were te sample initially spans te period from te first available observation to θ, and ten bot its starting and ending period are sifted forward by one datapoint at a time. As pointed out by Granger (2008), linear models wit time-varying parameters are actually very general nonlinear models, and terefore te cosen framework is ideally suited to analysing te issues of interest Innovation Accounting Examining all te effects of te lagged variables in a VAR model is often bot difficult and unnecessary for te purposes of te analysis (Sims, 1980). Rater, it is more convenient to resort to some transformations of te estimated model (1) in order to summarise te dynamic linkages among te n variables under investigation. In te business cycle literature, a useful metric often used to measure te extent of business cycle syncronisation is te sum of te variance sares of different classes of socks suc as countryspecific, regional or global sources of economic fluctuations (see Kose et al, 2012, among oters). 2 Since te reduced-form residuals u s are generally correlated, a common practice to obtain uncorrelated socks is to use te Coleski decomposition of Σ ( k ) u. Despite its straigtforward implementation, tis metod as te drawback tat it is sensitive to te ordering of te variables in te system, and terefore all possible permutations sould be considered wen carrying out te dynamic simulations for a toroug assessment. A popular alternative is provided by te Generalised Forecast Error Variance (GFEV) decomposition (Pesaran and Sin, 1998). Tis approac estimates te percentage of te variance of te -step aead forecast error of te variable of interest wic is explained by conditioning 2 Note tat te metodology used in Kose et al. (2012) differs from ours in tat tey compute te variance decomposition of te raw series of interest, wile in te present paper te forecast error variance decomposition is carried out. Terefore, wile we analyse te innovation (or unsystematic) part of te series represented by te residual of te estimated model, tey decompose its systematic part. Te limitation of teir approac, namely a Bayesian dynamic latent factor model, is tat it does not allow te identification of te geograpical origin of te factors affecting domestic business cycles, but rater of te world, region and country-specific components of a series. 6

8 on te non-ortogonalised socks wile explicitly allowing for contemporaneous correlations between tese socks and tose to te oter equations in te system 1 GFEV e e (3) σ ( k ) ( k ) ( k ) 2 l, m, = ( ) ( ) 0 lψ k q q Σ = u m m, m were te selection vectors e l and e m ave teir l-t and m-t entries equal to one, respectively, wit all oter elements being zero, and σ stands for te standard deviation associated to te m-t equation of ( k ) m, m te system. Altoug te GFEV metod does not allow a structural interpretation of te impulses, it overcomes te identification problem by providing a meaningful caracterisation of te dynamic responses of te variables of interest to observable socks. A furter useful feature of tis approac is its invariance to te ordering of te variables. Note, owever, tat owing to te possible non-diagonal form of matrix Σ ( k ) u, te sum over m of te elements (3) need not be unity in te original formulation provided by Pesaran and Sin (1998). In order to be able to interpret te results, we follow Wang (2002) and rescale (3) using te total variance in te generalised rater tan in te ortogonal case ξ = ( k ) l, m, 1 σ ( k ) ( k ) 2 ( e ) ( ) 0 l Ψ q Σ k q u e = m m, m 1 n ( k ) ( k ) 2 ( e ) 1 ( ) 0 l Ψ q Σ m k q u e = = m σm, m (4) so tat 1 (for all k and ), tat is te sum of te variance decompositions in (4) are n ξ ( k ) = m= 1 l, m, normalised to unity Measuring Spillover Effects at Country and Regional Level By computing te decomposition (4) for all variables in system (1) for a given recursion k and for a given simulation orizon, we obtain te following n n matrix (te spillover table according to te terminology of Diebold and Yilmaz 2012), 3 wic makes it possible to measure to extent to wic two or more variables of te system are connected to eac oter: 3 By construction, te elements in eac row of (5) sum up to unity, so tat te total variance of te system is equal to n. 7

9 ξ ξ... ξ ξ ξ... ξ ξ ξ ξ ξ ξ ξ M M O M M M O M ξ ξ... ξ ξ ξ... ξ ( k ) ( k ) ( k ) ( k ) ( k ) ( k ) ξ 1+ ξ ξ 1+ ξ ξ ξ 1+ ( k ) ( k ) ( k ) ( k ) ( k ) ( k ) ξ ξ 1+ ξ 1+ ξ ξ ξ 1+ M M O M M M O M ( k k k k k k ξn,1, ξ... ξ ξ ξ 1+ ξ ( k ) ( k ) ( k ) ( k ) ( k ) ( k ) 1,1, 1,2, 1, n1, 1, n1 + 1, 1, n1 + 2, 1, n, ( k ) ( k ) ( k ) ( k ) ( k ) ( k ) 2,1, 2,2,... 2, n1, 2, n1 + 1, 2, n1 + 2,... 2, n, ( k ) ( k ) ( k ) ( k ) ( k ) ( k ) n1,1, n1,2, n1, n1, n1, n1 + 1, n1, n1 + 2, n1, n, n 1,1, n 1,2, n 1, n, n 1, n 1, n 1, n 2, n 1, n, n 2,1, n 2,2, n 2, n, n 2, n 1, n 2, n 2, n 2, n, ) ( ) ( ) ( ) ( ) ( ) n,2, n, n, n, n 1, n, n 2, n, n, (5) Diebold and Yilmaz (2012) sow tat a syntetic measure of te spillovers received by (transmitted to) country l from (to) all oter countries can be obtained by summing by columns (rows) all te offdiagonal elements in te l-t row (column). We adapt teir framework by dividing te variable of system (1) into two distinct subsets (wic we label regional and external groups) wit dimension of n 1 = 6 and n n 1 = 4, respectively. We compute country-specific, δ ( k ) cs,, regional, δ ( k ) rs,, and external socks, δ ( k ) es,, as δ = ξ (6) ( k ) ( k ) cs, l, l, ( k ) n1 ( k ) ( k ) rs, m= 1 l, m, l, l, (7) δ = ξ ξ δ = ξ (8) ( k ) n ( k ) es, m= n + 1 l, m, 1 for all l = 1,..., n1. Moving from a single country to a regional perspective, we define te region-specific socks, γ, ( k ) reg, as ( k ) 1 n1 n1 ( k ) reg, l= 1 m= 1 l, m, n 1 (9) γ = ξ wilst te aggregate external socks (tat is, te innovations originating outside te LA region) are computed as ( k ) 1 n1 n ( k ) ext, l= 1 m= n l, m, 1 n 1 γ = ξ (10) 8

10 i.e., bot (6) and (7) are normalised so as to lie in te [0, 1] interval. In order to identify te direction of te linkages between te two (aggregate) blocs of countries we define te regional net spillover index, γ ( k ) net,, as te difference between te variability transmitted to and received by te elements belonging to te external bloc of te system k n n k n n k net, l= n + 1 m= 1 l, m, l= 1 m= n + 1 l, m, γ = ξ ξ ( ) 1 ( ) 1 ( ) 1 1 (11) so tat positive (negative) values for (11) indicate tat te region is a net transmitter (receiver) of variability to (from) outside. Using condition (11), it is straigtforward to obtain a breakdown for te individual countries forming te external bloc, so tat we can define n n1 pairwise regional net spillover indexes, γ ( k ) cty,, as ( k ) n1 ( k ) n1 ( k ) cty, m= 1 cty, m, l= 1 cty, m, γ = ξ ξ (12) were cty = n1 + 1,..., n. Before discussing te empirical findings it is wort noting tat bot te γ s and δ s indices depend on te simulation span (troug te index ) and on te estimation sample (troug te index k ). Tis is motivated by te need for a sufficiently flexible model specification to analyse te sources of business cycles in a period suc as te recent one caracterised by exceptionally large fluctuations. 4 Furter, considering a wide range of simulation orizons enables us to obtain a dynamic picture of ow crosscountry business cycle linkages evolve wen moving from te sort to te long run troug a sequence of 4 Note tat since te GFEVs are transformations of model parameters, allowing for time variation in te parameters of te underlying empirical model translates into time-varying nonlinear dynamic interactions among te elements in y in (1). 9

11 GFEV decompositions for wic te conditioning information is becoming progressively less important as te simulation orizon widens Assessing Business Cycle Co-movements in te LA Bloc: Country-specific and Regional Evidence 3.1. Data and Preliminary Analysis We use quarterly real GDP series for six major LA countries (namely, Argentina, Brazil, Cile, Mexico, Peru, and Venezuela) and for te four largest economies in te world (te US, te Euro area, Japan and Cina) over te period 1980:I-2011:IV. 6 Te ten cosen economies represent about 75 percent of real world GDP, wit te six LA countries included representing approximately 85 percent of real GDP in te LA region over te period according to te World Development Indicator data. As a preliminary step, we test for te presence of unit roots in te GDP series in logaritms. ADF tests are performed, bot on te levels and te first differences of te series. In eac case, we are unable to reject te null ypotesis of a unit root in te levels at conventional significance levels. On te oter and, differencing te series appears to induce stationarity. Standard stationarity tests corroborate tis conclusion. 7 Given te nonstationarity of te time series and te lack of an economic teory suggesting te number of long-run relationsips and/or ow tey sould be interpreted, it is reasonable not to impose te restriction of cointegration on a VAR model (Ramaswamy and Sloek, 1998). Tus, we ave opted for a 5 Note tat Diebold and Yilmaz (2012) focus on a selected forecast orizon rater tan a continuum of simulation steps. Obtaining full information from te entire simulation orizon is terefore novel in tis context. 6 Te GDP series are taken from Datastream and seasonally adjusted by using te X-12 metod, as suggested by Cesa- Bianci et al. (2011). Teir codes are: AGXGDPR.C (Argentina), BRXGDPR.C (Brazil), CLI99BVPH (Cile), MXI99BVRG (Mexico), PEI99BVPH (Peru), VEXGDPR.C (Venezuela), USXGDPR.D (US), EKXGDPR.D (euro area), JPXGDPR.D (Japan), CHXGDPR.C (Cina). Oter economies of te LA region (suc as Colombia or Bolivia) are not included in te analysis because of te lack of data on GDP for te eigties. Te same coice was made by Bosci and Girardi (2011) and Caporale et al. (2011), among oters. 7 Tese results are not reported to save space. 10

12 specification in first differences since te focus of our analysis is on (time-varying) sort-run linkages rater tan secular trends (as, for instance, in Bernard and Durlauf, 1995). More specifically, we coose size 80 for te rolling windows (i.e., 20 years of quarterly observations, 80 observations in all). Tis can be regarded as a compromise between stability and flexibility, as it turned out tat a smaller window size makes te VAR models more unstable. 8 Suc a coice implies tat te complete set of recursions produces 48 different sets of VAR estimates. Te GFEV decomposition analysis is ten conducted over a simulation orizon of 20 quarters (5 years) over te period 2000:I-2011:IV Te Role of Country-Specific, Regional and External Factors Te γ s and δ s defined in Section 2 above are tri-dimensional structures (surfaces) wose dimensions are given by te simulation orizon, te estimation window and te strengt (or direction) of te linkage among te elements of te system. 9 Figure 1 sows te decomposition of output variability of te LA countries into country-specific, regional and external sources. Figure 1 Te results reveal significant differences between countries wit respect to te relative importance of te variuos factors explaining output growt volatility. Regarding te country-specific components (grap I in Panels A-F), our model captures te deep crisis itting Argentina at te beginning of te current decade, as sown by te sarp increase in te contribution of te country-specific component in explaining output growt variability between te second alf of 2011 and te first semester of Furtermore, te global downturn led to a sarp drop in te contribution of internal factors in te last part of By contrast, te relative contribution of country-specific factors was more stable over time for te remaining LA economies (even in te troug te surface plot does not exibit dramatic canges in its sape), wit teir importance diminising in te long run. Over te period 2000:I-2011:IV, country-specific socks tend 8 For eac rolling estimate, te VAR models are specified wit two lags. Experimenting wit sorter and longer lag lengts (1 lag and 3 lags, respectively) did not cange muc te estimation results. 9 In particular, eac of tem summarises information obtained from combining several undred (48 times 20 = 960 to be precise) elements of matrix (5). 11

13 to dominate and account on average for between 75 (in Brazil, Cile and Venezuela) and 60 percent (in Mexico) of total variability in te sort run. At te end of te simulation orizon, instead, teir contribution drops markedly, ranging between 40 (in Argentina, Brazil and Cile) and 25 percent (in Venezuela). Concerning regional factors (grap II in Panels A-F), our results provide evidence of a sizeable regional business cycle component in te LA countries, as also found by Aiolfi et al. (2010) and Bosci and Girardi (2011), among oters. Averaging over all simulation steps and rolling estimates, we find tat regional factors account from about 20 (in te case of Cile) to 40 percent (for Venezuela) of output growt variability. For te largest economies of te LA region (Argentina and Brazil) te role of regional factors is relatively stable. Te same olds in Cile (and Peru) until te onset of te global downturn, after wic teir contribution decreases from 27 (40) to 17 (20) percent after 20 quarters. Finally, a slowly declining pattern is observed in te case of Mexico, in contrast to Venezuela, for wic a igly erratic evolution over time is found, wit te Argentine crisis of te early nougties translating into a sizeable increase (from about 30 to 60 percent after 20 quarters) in te contribution of regional factors. Finally, te average effect of external factors (grap III in Panels A-F) is witin a similar range to te one for te regional components (as also in Aiolfi et al., 2010), its minimum and maximum values being tose for Venezuela and Peru respectively. As for te individual countries, te observed pattern for Argentina mirrors tat of te country-specific component: te lowest value corresponds to te Argentine crisis, wilst te igest coincides wit te first symptoms of te global crisis. In all LA countries te role of external factors increases in te most recent years, te single exception being Brazil, were business cycles ave become less syncronised wit tose in te industrialised economies during te years of te great recession, te evidence suggesting terefore some partial decoupling Evidence from te LA Region Te variance decomposition for te LA region is computed as an (equally-weigted) average of individual country-specific figures. According to equations (9) and (10), it is based on a syntetic economy wic is an average LA country, as in Izquierdo et al. (2008). Figure 2 sows te decomposition of regional output variability between region-specific (Panel A) and external sources (Panel B). Figure 2 12

14 Regarding te regional sources of fluctuations, teir relative importance vis-à-vis te external ones appears to diminis over te simulation orizon. Te dominant role of external factors in te long run is found for all quarters. In particular, external factors account for about 30 percent of te long-run (20- quarter orizon) variance of LA GDP growt, consistently wit te evidence in Österolm and Zettelmeyer (2007) and in Aiolfi et al. (2010). As for te evolution over time of te estimated effects, te relative contribution of te two types of factors is remarkably stable up to te first alf of Wit te onset of te global crisis external factors appear to acquire an increasing role, especially at te very bottom of te global downturn (between mid and mid-2009), accounting for more tan 50 percent of total variability in 2008:IV. Subsequently, following a partial recovery, idiosyncratic factors ave regained some (but not all) of teir former importance. Our findings terefore give support to previuos evidence according to wic te LA region is still caracterised by eavy dependence on external factors and does not carry sufficient weigt to affect te international business cycle wit its own growt dynamics (Calvo et al., 1993; Izquierdo, 2008; Cesa- Bianci et al., 2011), tus contradicting te so-called decoupling ypotesis (Helbling et al., 2007). Furter evidence is presented in Figure 3, wic sows te difference between te variability transmitted to and received from te external bloc of te system as defined by (11). Tere is a predominance of negative values for te rolling estimates (especially wen considering long-run effects), suggesting tat te LA region can be caracterised as a net receiver of variability from te outside world. Tis applies even more strongly to te recovery period after te peak of te global crisis: te long-run net effect, after reacing a minimum of -17 percent in 2008:IV, is about -8 percent at te end of te sample. Figure 3 However, net spillover effects vis-à-vis an aggregate rest of te world could ide underlying eterogeneity, wic can only be detected by a more disaggregate analysis. Figure 4 presents te net pairwise spillover effects between te LA region and te US (Panel A), te Euro area (Panel B), Japan (Panel C) and Cina (Panel D). Figure 4 13

15 Bot sort- and long-run effects appear to be very stable, especially in te case of Cina. In particular, te balance between volatility transmitted to and received from te outside world is negative for te LA region in most cases. Tis is largely true for te years of te great recession ( ), during wic te LA region suffered from te recessionary impulses coming from te most advanced economies (but not from Cina). In te most recent years, owever, te overall picture seems to ave canged significantly, namely te impact of business cycle conditions in te US, te Euro area and Japan as diminised, wilst te influence of te Cinese economy as increased. Overall, te disaggregate results provide no evidence of de-coupling; tey also indicate tat bilateral linkages wit Cina ave become stronger, making te LA region vulnerable not only to economic ardsip in te industrialised economies but also to future developments in Cina, as already pointed out by Cesa-Bianci et al. (2011). 4. Determinants of te Linkages between te LA Region and te World Economy 4.1. Te Role of Trade and Capital Flows Since te study of Frankel and Rose (1998) a considerable body of empirical researc (Imbs, 2006, 2010; Kalemli-Ozcan et al., 2009) as sown tat bilateral trade flows (tra) and financial linkages ( fin) can affect business cycles correlations (ρ) across countries and/or regions. Following tis literature, a canonical regression model can be specified as ρ = ψ + ψ tra + ψ fin + ε (13) Te positive effect of bilateral trade flows on te degree of international business cycle syncronisation as been widely establised in te literature (see Imbs, 2006 among oters) and is consistent wit te teoretical predictions of te model developed by Kose and Yi (2006). As for capital flows, several studies give support to te view tat financial integration increases te degree of business cycle correlations in cross-sections and over time (Imbs, 2006, 2010), wilst oter papers (e.g., Kalemli- Ozcan et al., 2009) find tat financially integrated economies ave negative co-movements, as posited by 14

16 standard international real business cycle models wit complete markets (Backus et al., 1994). Te sign for ψ 3 in (13) is tus te object of empirical scrutiny. 10 intensities as We follow Frankel and Rose (1998) and compute (a time variant version of) bilateral trade tra t = X + X l, e, t e, l, t Y + Y l, t e, t were l, e, t X denote total mercandise exports from te LA region (l) to te external bloc ( e), X e, l, t are exports from te aggregate foreign economy to te LA region, Y l, t and Y e, t are te GDP nominal levels in te two economies, and t is a time index. As for capital flows between te two blocs, tey are computed as fin t NFA NFA = Y Y l, t e, t l, t e, t were NFA l, t and NFA e, t stand for te net foreign asset position in te two aggregate economies. Te rationale beind suc a proxy for capital movements is tat capital sould flow between countries or regions wit different or even opposite external positions (Imbs, 2006). In particular, we compute te net foreign asset position as te sum of net positions in debt ( dbt ), equities ( eqt) and foreign direct investment ( fdi) as in Caballero (2012), among oters In te literature additional explanatory variables for ρ, suc as excange rate arrangements and te structure of production and trade, ave been suggested. However, we are interested in explaining ow te degree of business cycle syncronisation as canged over time rater tan across countries, and terefore time-invariant regressors or explanatory variables tat only cange slowly over time are ruled out from te present study. A time series analysis of te determinants of business cycle correlations is quite novel, only a limited number of studies on tis topic being available at present (see, among oters, Kalemli-Ozcan et al., 2009; Imbs, 2010). 11 Bilateral trade data and statistics for capital flows are from te IMF s DoTS and IFS BoP databases, respectively. Te analysis focuses on net capital flows. Since IFS BoP records outflows as negative numbers, to obtain net flows assets and liabilities are added. FDI data for Cina are not available for te entire sample span considered in te analysis, and 15

17 4.2. Estimation Results Using spillover indexes rater tan standard correlation coefficients makes it possible to analyse (timevarying) business cycle correlations in a muc more flexible framework by distinguising between comovements at different forecast orizons. To see tis, we start by mapping our spillover index to te (timevarying) correlation coefficients. Following Forbes and Rigobon (2002), we consider te following least square regression between output growt rates of countries a and b, ya = a + b yb + u, so tat b σ ρ = 2 2 b 2 σa 0.5 or 2 ρ σ = b 2 b σa 0.5 (14) Te term in square brackets on te RHS of te second expression in (14) is te sare of output growt variability of country a explained by b. In terms of our framework, it is expressed by γ ext in (10). Condition (14) implies tat * ρ = ρ b = γ ext (for a given forecast orizon). Accordingly, equation (13) can be rewritten as follows ρ = α + α tra + α fin + ε * * * * * (15) were * α i = α i b, i = 1,...,3, b ε = ε, provided tat b 0. As γ ext is computed over a number of different simulation steps,, condition (15) can be tested at several forecast orizons in order to assess weter and ow te role of trade and financial linkages varies according to. In wat follows we consider selected simulation orizons (namely, = 1, 2, 4,8,12, 20). Te first step of te empirical analysis is based on standard correlation measures between * ρ and its main macroeconomic determinants. Table 1 sows tat te unconditional correlation coefficients for te * ρ s and tra variables are positive and statistically significant for all forecast orizons considered, confirming te well establised finding tat iger business cycle syncronisation is associated wit terefore fdi is computed using data only for te US, te Euro area and Japan. Te series ave been seasonally adjusted using te X-12 metod. 16

18 stronger trade intensity. One migt argue tat te positive correlation is spurious owing to te existence of factors correlated to bot variables. Wen conditioning on fin, te magnitude (and te statistical significance) of te partial correlation coefficients remains virtually uncanged. By contrast, te unconditional correlation between * ρ s and fin turn out to be statistically insignificant. Te same conclusion olds wen considering te partial measure of association (conditioned on tra), even toug te sign of te relationsip in general becomes negative. Table 1 Correlations are only partially informative as tey cannot gauge causality between te regressand and te explanatory covariates. In order to delve deeper into te effects of bilateral trade and capital flows on business cycle syncronisation, we estimate equation (15) by 2SLS for te cosen simulation orizons. 12 In order to control for te collapse (and te subsequent abrupt recovery) in trade flows wic occurred during te crisis, we augment te set of regressors by a crisis dummy ( dum) taking te value of -1 in 2008:III and 2008:IV and +1 in 2009:I and 2009:II. 13 Table 2 presents te estimation results of te baseline specification. 14 Single, double or triple asterisks denote statistically significant coefficients at te 1, 5 or 10 percent level, respectively. We also report robust standard errors (in parenteses) as well as some basic diagnostics for te cosen instruments ( J statistics), te serial correlation of te residuals ( DW ) and te goodness of fit of te regression ( R ). Table 2 2 adj 12 Typical external instruments for trade intensity are spatial caracteristics (e.g. geograpic proximity or te presence of common borders), and for financial integration institutional variables related to legal arrangements. As most of tese instruments are constant over time, tey cannot be used in a time series framework. In order to address endogeneity concerns, bilateral trade intensity and capital flows are measured at te beginning of te period and are treated as predetermined variables. 13 Te estimation results in Sections 4.2. and 4.3. are not affected by te inclusion of dum : re-estimating model (15) witout it produces qualitatively similar results to tose reported in te main text. 14 After considerable experimentation, our preferred specification is based on variables expressed in year-on-year canges. For te purpose of readibility variables ave been standardised. 17

19 Te estimation results indicate a clear dominance of trade flows over financial linkages as te main determinant of business cycle co-movements between te LA region and te foreign bloc, even controlling for te trade collapse in : te coefficient of bilateral trade intensity is positive and statistically significant at all simulation steps; moreover it increases almost monotonically wit. Capital flows reduce te degree of co-movement between cycles, but te coefficients of fin are generally small in magnitude and imprecisely estimated. Overall, tese findings reinforce te disaggregate results discussed in Section 3.3 and sow tat, in te presence of relatively weak financial linkages, propagation of te impulses from outside to te LA bloc as appened mainly troug trade flows. Te apparent de-coupling of te LA bloc from te most advanced economies tus arises not only from trade being increasingly oriented towards Cina rater tan its istorical trading partners (namely te US and te Euro area see Cesa-Bianci et al. (2011), but also from a low degree of financial integration wit te rest of te world economy Extensions In tis Section we present te results from a disaggregate analysis based on a breakdown of capital flows into debt, equity and FDI flows wit te aim of sedding ligt on wat type of flows are beind stronger business cycle co-movements. We first assess te role of tese components by replacing fin wit disaggregated capital flows (entering te model individually). Te results in Table 3 indicate tat te trade cannel, albeit dominant, is not te only one: capital flows can also affect te degree of international business cycle syncronisation in te sort run (namely, up to te fourt simulation step). Moreover, portfolio equity flows ave a negative effect on te degree of business cycle syncronisation wilst tat of debt and foreign direct investment is positive. Table 3 As a furter step, we consider a specification were te tree types of flows enter te model simultaneously (Table 4). Over te first year of te simulation orizon its explanatory power is iger wit 18

20 respect to its (nested) counterparts in Table 2 and Table 3, suggesting tat debt, portfolio equity and foreign direct investment act as additional cannels of transmission of socks from abroad. Our findings complement previous evidence for emerging markets according to wic bot trade and financial variables mattered prior to te global crisis (Blancard et al., 2010), since we document tat tese factors largely explain business cycle co-movements over te last decade. However, our framework makes it possible to go furter and to igligt te relative strengt of te different transmission cannels in te sort and long run: te increasing explanatory power of trade flows over te entire simulation span is largely corroborated, wereas capital flows affect business cycle co-movement in te sort term, as te 2 R adj statistics sow. 15 Moreover, in te very sort term debt and foreign direct investment ave an opposite effect compared to equity portfolio flows. Wile te result for eqt can be rationalised witin te standard international real business cycle framework wit complete markets, our findings for dbt and fdi suggest tat sort-term capital flows and internationalisation of production troug foreign direct investment may strengten te role of trade cannel making te LA region more prone to suffer from te propagation of socks from abroad. Table 4 5. Conclusions Tis paper presents a flexible framework to assess bot sort- and long-run linkages between business cycles. Specifically, we extend te econometric approac of Diebold and Yilmaz (2012) to analyse te extent to wic business cycle developments in six LA countries (namely, Argentina, Brazil, Cile, Mexico, Peru and Venezuela) and te four largest economies in te world (te US, te Euro area, Japan and Cina) are connected over te period 1980:I-2011:IV. 15 Tis also implies tat te statistically significance of te coefficient on equity portfolio flows after te first four quarters of te simulation span makes only a marginal contribution to explaining ow te LA bloc and te rest of te world co-move. 19

21 For tat purpose, we decompose macroeconomic fluctuations in domestic output growt rates into te following components: i) country-specific (idiosyncratic) factors; ii) regional factors, capturing fluctuations tat are common to all countries belonging to te LA region; iii) external factors, wic are related to business cycle development outside te LA region. Most importantly, we are able to determine te direction and te intensity of te propagation mecanisms and terefore establis to wat extent te LA economies and LA region as a wole ave been dependent on (or influenced by) external developments over time. Overall, te business cycle of te individual LA countries appears to be influenced by countryspecific, regional and external socks in a very eterogenous way. Also, te LA region as a wole is strongly dependent on external developments. Tis conclusion olds especially for te years after te great recession of 2008 and 2009, ruling out any decoupling of te LA region from te rest of te world. More specifically, we find a clear dominance of trade flows over financial linkages as a determinant of business cycle co-movements between te te LA region and te foreign bloc. Te apparent de-coupling of te LA area wit respect to te most advanced economies in recent years tus seems to ave been determined not only by increasing trade flows towards Cina but also by a low degree of financial integration wit its main economic partners. Te decomposition of capital flows into teir components (debt, portfolio equity and foreign direct investment flows) sows a negative effect of portfolio equity flows on te degree of business cycle syncronisation, consistently wit te predictions of standard international real business cycle models wit complete markets. In contrast, sort-term capital and foreign direct investment flows tend to reinforce in te sort run te role of te trade cannel and te responsiveness of te LA region to external developments. Te proposed econometric approac is of more general interest, since it does not include any variables wic are igly country- or region-specific and tus can also be used to investigate te relationsip between comovement across countries/regions or financial markets and teir macroeconomic determinants. For instance, it could be applied to analyse te factors tat ave influenced integration of te real economies in te European Monetary Union after te adoption of te euro, or to assess te 20

22 istorical determinants of te regional convergence/divergence dynamics witin countries over time, or to conduct a macro analysis of market integration in a given financial segment. Tese are all interesting topics for future researc. 21

23 References Acemoglu D., S. Jonson, J.A. Robinson and Y. Taicaroen, Institutional Causes, Macroeconomic Symptoms: Volatility, Crisis, and Growt, Journal of Monetary Economics, 50, pp Amed S Sources of Economic Fluctuations in Latin America and Implications for Coice of Excange Rate Regime, Journal of Development Economics, 72, pp Aiolfi M., L. Catão and A. Timmermann, Common Factors in Latin America s Business Cycles, CEPR Discussion Paper Artis M., J.H.-M. Krolzig and J. Toro, Te European Business Cycle, Oxford Economic Papers, 56, pp Atanasoulis S. and E. van Wincop, Growt Uncertainty and Risk-Saring, Journal of Monetary Economics, 45, Backus D., P. Keoe and F. Kydland, Dynamics of te Trade Balance and te Terms of Trade: Te J- Curve?, American Economic Review, 84, pp Bagliano F. and C. Morana, Business Cycle Comovement in te G-7: Common Socks or Common Transmission Mecanisms?, Applied Economics, 42, pp Baxter M. and M. Kouparitsas, Determinants of Business Cycle Comovement: A Robust Analysis, Journal of Monetary Economics, 52, pp Bernard A.B. and S.N. Durlauf, Convergence in International Output, Journal of Applied Econometrics, 10, pp Blancard O., H. Faruqee and M. Das, Te Initial Impact of te Crisis on Emerging Market Countries, Brookings Papers on Economic Activity, 41, pp Bosci M., Long- and Sort-run Determinants of Capital Flows to Latin America: A Long-run Structural GVAR Model, Empirical Economics, fortcoming. Bosci M. and A. Girardi, Te Contribution of Domestic, Regional and International Factors to Latin America s Business Cycle, Economic Modelling, 28, pp Caballero J., Do Surges in International Capital Flows Influence te Likeliood of Banking crises?, IDB Working Paper

24 Calvo G., L. Leiderman and C. Reinart, Capital Flows and Real Excange Rate Appreciations in Latin America, IMF Staff Papers, 40, pp Canova F., Te Transmission of US Socks to Latin America, Journal of Applied Econometrics, 20, pp Caporale G.M., D. Ciferri and A. Girardi, Fiscal Socks and Real Excange Rate Dynamics: Some Evidence for Latin America, Journal of International Money and Finance, 30, pp Diebold F.X. and K. Yilmaz (2012). Better to Give tan to Receive: Predictive Measurement of Volatility Spillovers, International Journal of Forecasting, 28, pp Fatàs A. and I. Miov, Policy Volatility, Institutions and Growt, mimeo, INSEAD. Forbes K. and R. Rigobon No Contagion, Only Interdependence: Measuring Stock Market Comovements, Journal of Finance, 57, pp Frankel J. and A. Rose, Te Endogeneity of te Optimum Currency Area Criteria, Economic Journal, 108, pp Granger C.W.J., Non-Linear Models: Were Do We Go Next - Time Varying Parameter Models?, Studies in Nonlinear Dynamics and Econometrics, 12, pp Helbling T., P. Berezin, M.A. Kose, M. Kumof, D. Laxton and N. Spatafora, Decoupling te Train? Spillovers and Cycles in te Global Economy, World Economic Outlook, pp Hoffmaister A.W. and J. Roldos, Are Business Cycles Different in Asia and Latin America?, IMF Working Papers 97/9. Imbs J Te Real Effects of Financial Integration, Journal of International Economics, 68, pp Imbs J., Te First Global Recession in Decades, IMF Economic Review, 58, pp Izquierdo A., R. Romero and E. Talvi, Booms and Busts in Latin America: Te Role of External Factors, IDB RES Working Papers Kalemli-Ozcan S., E. Papaioannou and J.L. Peydro, Financial Integration and Business Cycle Syncronization, NBER Working Papers Kose M.A., C. Otrok and E.S. Prasad, Global Business Cycles: Convergence or Decoupling?, International Economic Review, 53, pp

25 Kose M.A., E.S. Prasad and M. Terrones, How Does Globalization Affect te Syncronization of Business Cycles?, American Economic Review-Papers and Proceedings, 93, pp Kose M.A. and K.-M. Yi, Can te Standard International Business Cycle Model Explain te Relation between Trade and Comovement?, Journal of International Economics, 68, pp Loayaza N.V., R. Rancière, L. Servén and J. Ventura, Macroeconomic Volatility and Welfare in Developing Countries: An Introduction, World Bank Economic Review, 21, pp Mody A., L. Sarno and M.P. Taylor, A Cross-country Financial Accelerator: Evidence from Nort America and Europe, Journal of International Money and Finance, 26, pp Österolm P. and J. Zettelmeyer, Te Effect of External Conditions on Growt in Latin America, IMF Working Papers 07/176. Pesaran M.H. and Y. Sin, Generalized impulse response analysis in linear multivariate models, Economics Letters, 58, pp Ramaswamy R. and T. Sloek, Te Real Effects of Monetary Policy in te European Union: Wat are te Differences?, IMF Staff Papers, 45, pp Sims C.A., Macroeconomics and reality, Econometrica, 48, pp Wang P.J. (2002). Financial Econometrics: Metods and Models, Routledge, London. 24

26 Figure 1 LA countries: country-specific, regional and external factors (continued) Panel A Argentina I. Country-specific component II. Regional component III. External component Panel B Brazil I. Country-specific component II. Regional component III. External component