ISSN 1833-4474 Does Demographc Change Affect the Current Account? A Reconsderaton # Mchael Graff, a,* Kam K Tang, b, Je Zhang c Ths paper re-examnes the mpact of demographc factors on the current account balance. To ths end, we develop an analytcal framework that s more general than the one commonly used n the lterature n three aspects. Frst, t accounts for the facts that the world current account balance must be equal to zero. Second, a bgger economy wll have greater mpacts on others, but be nfluenced less by them. Thrd, a more open economy wll have greater mpacts on others and at the same tme be more readly nfluenced by them. We then confront two alternatve emprcal specfcatons based on both the new and the conventonal framework wth a panel of data. In contrast to the fndngs based on the conventonal framework, our results wth the new framework ndcate that populaton ageng does not appear to have dscernble mpacts on the current account balance. Key words: Demography, ageng, openness, current account, general equlbrum JEL Classfcaton: F30, J10 a b c ETH Zürch, Swtzerland and Jacobs Unversty, Bremen, Germany Unversty of Queensland, Australa Natonal Unversty of Sngapore, Sngapore * Correspondng author. Prof. Dr. Mchael Graff ETH Zurch KOF, ETH Zentrum WEH CH-8092 Zurch Swtzerland graff@kof.ethz.ch # The authors would lke to acknowledge the support from the Australan Research Councl's Dscovery Projects fundng scheme (project number DP0878752).
Does Demographc Change Affect the Current Account? A Reconsderaton # Mchael Graff, a,* Kam K Tang, b, Je Zhang c Abstract Ths paper re-examnes the mpact of demographc factors on the current account balance. To ths end, we develop an analytcal framework that s more general than the one commonly used n the lterature n three aspects. Frst, t accounts for the facts that the world current account balance must be equal to zero. Second, a bgger economy wll have greater mpacts on others, but be nfluenced less by them. Thrd, a more open economy wll have greater mpacts on others and at the same tme be more readly nfluenced by them. We then confront two alternatve emprcal specfcatons based on both the new and the conventonal framework wth a panel of data. In contrast to the fndngs based on the conventonal framework, our results wth the new framework ndcate that populaton ageng does not appear to have dscernble mpacts on the current account balance. Key words: Demography, ageng, openness, current account, general equlbrum JEL Classfcaton: F30, J10 a b c ETH Zürch, Swtzerland and Jacobs Unversty, Bremen, Germany Unversty of Queensland, Australa Natonal Unversty of Sngapore, Sngapore * Correspondng author. Prof. Dr. Mchael Graff ETH Zurch KOF, ETH Zentrum WEH CH-8092 Zurch Swtzerland graff@kof.ethz.ch # The authors would lke to acknowledge the support from the Australan Research Councl's Dscovery Projects fundng scheme (project number DP0878752).
1. INTRODUCTION Motvated by the observaton that advanced economes are ageng rapdly and emergng economes are followng soon, there has been ncreasng nterest n the macroeconomc effects of ageng. 1 One recurrent topc of nterest n the lterature s the mpacts of ageng, and demography n general, on the current account balance. 2 Accordng to the lfe cycle hypothess of consumpton, young households borrow aganst ther future ncome, mddle-age households save for relnqushng debts and retrement, and old-aged households dssave. Accordngly, countres wth a relatve young or old populaton are more lkely to run current account defcts (more on ths later). The focus of ths paper s on how to model demographc mpacts on the current account balance n a world of heterogeneous economc openness and szes. The htherto standard practce of accountng for these heterogenetes s to nclude measures of openness n the regressons and to normalse the current account balance wth Gross Domestc Product (GDP), respectvely. We show n ths paper, frst, that n a general equlbrum settng ths practce s nsuffcent and second, that the problem can be addressed by usng a new framework. Where cross-country (ncludng panel) data are used as s typcally the case n the related lterature three factors need to be consdered n modellng the demographc mpacts on the current account balance. The frst factor s the general equlbrum condton. Snce the current account balances must sum to zero for the world as a whole, demographc changes that affect the home country s balance must affect the rest of the world s (ROW s) balance n the opposte drecton; conversely, demographc changes n the ROW must also affect the home country. Ths mples that t s not the home country s own pace of demographc changes that matters, but ts pace relatve to the ROW (Chnn and Prasad 2003). The second factor s country sze. Other thngs beng equal, the larger the home country s, the more nfluence t should have on others, whle at the same tme t should be less susceptble to external dsturbances. The last factor s openness. For a gven sze, a more open economy should have a more pronounced mpact on others and smultaneously be more readly nfluenced by the ROW. Wthout due consderaton of these three factors taken together, any emprcal estmaton of the effects of demography on the current account balance s lkely to be based n varous ways, de- 1 Some mportant earler works nclude Hggns and Wllamson (1997) and Hggns (1998). Also see Edwards (1996) and L, Zhang and Zhang (2007). 2 Recent examples nclude Chnn and Prasad (2003), Chnn and Ito (2007), Gruber and Kamn (2007) and Legg, Prasad and Robnson (2007).
pendng on the dstrbutons of demographc measures, openness and szes n the sample. To the best of our knowledge, ths s the frst study that pertans to ths structural ssue n the related emprcal lterature. To llustrate our argument, we estmate the mpacts of demographc change on the current account balance usng both the conventonal and our proposed new framework, whch allows us to compare and contrast the results. Our panel dataset comprses 84 countres and 47 years from 1960 to 2006. Based on our data and methods, we fnd that the effect of populaton ageng on the current account balance s much more subdued than what has been recorded n prevous studes based on the conventonal framework. The rest of the paper s organsed as follows. Secton 2 revews the related lterature. Secton 3 derves the new modellng framework. Secton 4 descrbes the data used for our emprcal analyses. Secton 5 presents and dscusses the results, and the fnal secton concludes. 2. LITERATURE REVIEW Natonal account denttes dctate that the current account balance s equal to the excess of natonal savngs ( T - G + S) over domestc nvestment ( I ) or the excess of output ( Y ) over domestc absorpton ( C + I + G) : CA = ( T - G) + S - I = Y - ( C + I + G) (1) where CA, T, G, S and C represent, respectvely, the current account balance, tax revenue, government expendture, prvate savng and prvate consumpton. It s on the bass of ths dentty that demography s hypothessed to nfluence the current account balance. Accordng to Modglan s (1970) lfe cycle hypothess, households at workng age are the prme net savers of socetes, whle young households are lkely to be borrowers and old-age ones are lkely to be dssavers. Therefore, countres wth a relatve young or old populaton are more lkely to consume more than what they produce, resultng n a current account defct. Whle the majorty of the lterature adopts ths hypothess and thus expects youth and old-age dependency 3 to have negatve mpacts on the current account balance, ths s not the full story. Often mssng n the dscusson s the nvestment aspect. Snce the current account balance s equal to excess savng over nvestment, demography can only nfluence the current account balance to the extent that ts respectve effects on savng and nvestment do not net out each other. In 3 Here, youth and old-age dependency are defned as populaton aged 0-14 and 65+ to populaton aged 15-64. 1
ths aspect, youth and old-age dependency could have opposte effects on the current account balance. Ths s because countres wth a young populaton wll see ther labour force growng n the future, makng long-term captal nvestment more attractve there than n countres wth an agng populaton (Cooper 2008). Another mssng pece s longevty. Populaton ageng s charactersed not only by rsng old-age dependency but also by rsng longevty, 4 yet the theoretcal mpacts of the two on savng are dfferent. As people expect to lve longer, they are nduced to save more, counterbalancng the mpacts of hgher old-age dependency (L, Zhang and Zhang 2007). Natonal account denttes also necesstate a country s current account balance to be equal to ts recepts from the ROW for goods, servces, nvestment ncome, and unlateral transfers mnus ts payment to the ROW. Thus, a current account surplus (defct) represents a net outflow (nflow) of captal. As such, the effects of demography on the current account balance must be casted n the lght of open economes. Clearly, these consderatons reflect the three factors dscussed n the ntroducton, namely general equlbrum, sze, and openness, and we shall consders these factors smultaneously n our modellng of the current account balances wthn the world economy. The paper s related to a strand of the lterature on the macroeconomc mpacts of populaton ageng. Wthn ths lterature, Hggns and Wllamson (1997) and Hggns (1998) fnd that countres wth relatvely young populatons are captal mporters, whereas those wth relatvely old populatons are captal exporters. Amongst these two studes, only Hggns (1998) controls for openness. He fnds that demography does not affect the balance of trade n economes classfed as closed based on the Sachs and Warner (1995) bnary measure of openness. Our study dffers from these two studes n that we take nto consderaton relatve rather than absolute demography shfts across countres, as well as the heterogenety of countres n terms of openness and relatve economc sze. In recent years, some studes examnng the macroeconomc effects of demography also ncorporate the condton that external balances must sum up to zero for the world economy as a whole. These nclude Ferol (2003), Domej and Flodén (2006), and Attanaso, Ktao, and Volante (2006). However, these studes use numercal smulatons as ther man nvestgaton tool, n contrast to the emprcal parameter estmaton by regresson analyses n the current paper. Another strand of related studes seeks to explan current account mbalances; e.g. Chnn and Prasad (2003), Chnn and Ito (2007), Gruber and Kamn (2007), and Legg, Prasad and Robn- 4 Another aspect of populaton ageng s a fallng fertlty rate. However, n the short to medum term, a fallng fertlty rate means a lower youth dependency rate, but not a rsng old-age dependency. 2
son (2007). The last two mentoned above emphasse the effect of the Asan fnancal crss as a catalyst of the Asan economes compulson to buld up large foregn reserves, known as the global savng glut hypothess. Km and Lee (2007), runnng a VAR, fnd that for ten East Asan countres from 1981-2003, the dependency rato has a sgnfcantly negatve effect on the current account balance, thus provdng evdence that the effects of the present demographc transton are no longer restrcted to the wealthest economes. Whle these studes dffer vastly n terms of sample coverage and econometrc approaches, n general, t s far to summarse that they fnd that demography matters and that a hgher dependency rato tends to result n a smaller current account balance. In what follows, we shall challenge ths fndng. Importantly, the exstng body of lterature ndcates that cross-country or panel data are more nstrumental than ndvdual country tme seres data n dentfyng demographc effects on savng, because the varaton of demographc varables s typcally more pronounced across countres than across tme (Masson, Bayoum and Same 1998). For ths reason, our analyses draw on a large panel data set rather than on a cross secton of countres or on longtudnal data for partcular countres. Another dfference between the current paper and the prevous lterature s that we estmate both statc and dynamcs models. Hggns (1998) and Gruber and Kamn (2007), for nstance, do not account for the strong persstence of the dependent varables n ther econometrc models and ther results are thus subject to severe omtted varable bas. 5 Hggns and Wllamson (1997) and Legg, Prasad and Robnson (2007) do take nto account the dynamc nature of the data, but use dfferent estmaton technques. In partcular, Hggns and Wllamson use a two-stage-least-square estmator, whle Legg, Prasad and Robnson use a varant of dynamc panel data estmaton from Anderson-Hsao (1981). For the dynamc modellng, our emprcal methods also nclude the generalsed methods of moments (GMM) estmators proposed by Arellano and Bover (1995) and Blundell and Bond (1998). 3. THE NEW MODELING FRAMEWORK Consder the followng general expresson of the current account balance: CA = l ( X, q ), (2) 5 Although Hggns (1998) uses 5-year averages to try to smooth out short run volatlty, ths does not necessary remove the autocorrelaton of the data. 3
where CA s defned as the current account balance as a share of GDP n country ; q s a scalar measure of economc openness (1 for fully open, 0 for completely closed); and X s a vector of domestc factors n the home country that are expected to affect ts current account balance. In what follows, we modfy ths general framework to take nto account of, one at a tme, general equlbrum, openness and sze as dscussed prevously. Frstly, the general equlbrum condton mples that, f a domestc factor affects the home country s current account balance, t must affect the balance for the ROW n the opposte drecton. As such, t s the change n country relatve to that n the ROW that matters n determnng ts current account balance. To accommodate ths, we modfy (2) nto CA = l ( X - RWX, q ), (3) where RWX measures the value of the same varables as n X, but for the rest of the world (ROW): 6 RWX º average { X, j ¹ }. (4) Assumng X and j RWX enter l (.,.) addtvely s reasonable, when the arguments are ex- pressed ether n percentage terms or as log term values. Secondly, the mpact of ( X - RWX ) on the home country s current account balance should be condtonal on ts openness. In partcular, the less open ts economy s, the less susceptble ts current account balance to both nternal and external dsturbances should be; for a completely closed economy, the current account balance must be equal to zero for any values of ( X - RWX ). Ths mples that (3) can be wrtten as where CA RWX s defned as n (4). = q f ( X - RWX ), (5) Ths specfcaton, however, stll neglects another aspect of openness. Not only has a country to be open to be nfluenced by others, but also that other countres have to be open to be able to exert ther nfluence. To account for ths, we replace the unweghted averagng functon n (4) wth an openness-weghted averagng functon: 6 Ths s the specfcaton n Chnn and Prasad (2003). However, we found that for our dataset X and ( X - RWX ) as specfed n (4) are correlated close to unty, meanng that just accountng for the general equlbrum condton alone would make lttle practcal dfferences to not consderng t at all. 4
RWX º / å q X q j j å. (6) j j¹ j¹ Assumng f (.) s a lnear functon of ts arguments, the correspondng regresson model of (5) s CA = a + q( X - RWX )' α + e, (7) 0 1 where e s the error term. Thrdly, for the same degree of openness, a larger economy wll have a more pronounced nfluence on others, but be less affected by others at the same tme. To account for ths sze factor, we need to modfy both (6) and (7) as follows: CA = a + qz ( X - RWX )' α + e (8) 0 1 RWX º / å q GDP X qgdp j j j å (9) j j Z º j¹ j¹ å j q GDP j j j j GDP The modfcaton to the dentfy functon for / å q. (10) RWX s straghtforward, as we merely re- place the openness-weghted averagng functon n (6) wth an openness-gdp weghted averagng functon n (9). The latter means that shocks from larger and more open foregn countres wll be weghted hgher than those from smaller and less open ones n mputng the external dsturbances confronted by the home country. The modfcaton to the equaton for nteracton term q ( X - RWX ) wth a new term, Z. Ths new term CA s somewhat more complcated, as we multply the Z s the nverse of the rela- tve sze of the domestc economy compared to the world average, and the sze of each foregn country s also weghted by ts openness. Z s a scalar, and t wll be larger than one for small economes and smaller than one for large economes. The reason for ths specfcaton s clear when we consder a world consstng of only two countres, n whch the home country s twce the sze of the foregn country j, and the vector of X has only one element. Then further suppose q = q = 1 and D( X - X ) = 1. In the counterfactual case that Z does not appear n equaton j j (8), D CA = a. Snce D( X - X ) = - 1, D CA = - a holds smultaneously. In dollar terms, 1 j j 1 the current account surplus n country has to match the current account defct n country j. How- 5
ever, snce CA s expressed as a rato to GDP a standard practce n the lterature to ensure that the world current account balance s zero, the effect of D( X - X ) on CA must be half as that of j D( X - X ) onca. That s, the margnal effect wll be smaller for the larger economy, and vce j j versa. In the absence of Z, the estmated margnal effect wll be somewhere between the actual margnal effects on CA andca. In a more general, mult-country settng, the error wll depend on the dstrbuton of economc szes. The ncluson of j Z n (8) provdes a soluton to ths problem and should hence lead to a more accurate estmaton of the effect of ( X - RWX ) on CA. 7 Fur- thermore, the openness measure enters others only f t s open. Z because, lke that n RWX, an economy can nfluence 4 EXPLANATORY VARIABLES AND DATA 4.1 Explanatory varables Our objects of nterest are the demographc effects on the current account balance. Ths s related to but dfferent from studes of the determnants of the current account balance. The man dfference s that we focus on demographc factors and nclude a lmted number of control varables. Gven that the lterature has elcted numerous determnants of these two macro varables, a legtmate concern s how many of those determnants have to be ncluded as controls to avod based estmatons for the demographc effects. In ths aspect, t s useful to notce that the demographc structure of a country s shaped by ts fertlty and mortalty rates over decades, makng t deeply predetermned and most lkely exogenous wth respect to many other macroeconomc varables. Accordngly, the excluson of those varables should not cause omsson bas to the estmates for demography. On the other hand, t can be argued that demography and many other macroeconomc varables are all drven by even deeper determnants such as culture, and therefore models that are too parsmonous would stll be subject to omtted varable bas. Our strategy, therefore, s frst to estmate models wth only demographc varables, and then to add a number of control varables to test f the results of the more parsmonous specfcatons are robust. 7 There s a reason why we measure the sze of the home economy relatve to the world average, but not the average of the ROW. Consder the two-country example n the text. If we use the average sze of the ROW, for the small country, ts relatve sze wll be equal to 2 and that of the bgger country wll be equal to ½. As a result, the margnal effect of the home country wll be a quarter of that of the foregn country, nstead of just half of t. 6
Followng common practce n the lterature, we measure demographc structure usng youth and old-age dependency ratos. For the control varables, snce the current account s equal to excess savng over nvestment and s assocated to nternatonal captal flows, factors pertanng to the lfecycle consumpton smoothng, nvestment return and rsk are partcularly relevant. Varables pertanng to consumpton smoothng nclude ncome growth, the rural populaton share, and labour partcpaton rate. Our controls pertanng to nvestment return nclude human captal (measured by average years of schoolng) and the busness cycle (measured by the output gap, lags of the dependent varables and year dummes). Factors pertanng to rsk nclude fnancal development (measured by a fnancal development ndex), nsttutonal qualty and poltcal stablty (measured by a composte rsk ndex), and a measure of economc openness. A number of these varables requre some elaboraton. Frst, we do not nclude ncome n levels because ncome s found to be non-statonary 8, and ths would weaken the nternal nstruments used n GMM estmatons; as a result, only ts frst-dfference,.e. the ncome growth rate, s used. 9 Second, the share of the rural populaton n the total populaton s ncluded as a control varable because rural households, especally those n developng countres, may not have as good access to fnancal ntermedaton as ther urban counterparts and thus have dfferent savng behavour. Lastly, a measure of economc openness s ncluded, manly to ease comparson wth the conventonal model, but we also nclude t n our alternatve model as a standalone control varable to avod the transformed varables from pckng up the effect of the openness measure. 4.2 Data The defntons, data sources and basc summary statstcs of the varables referred to are summarsed n Table 1. Most of the varables are measured accordng to standard practce n the related emprcal lterature that draws on cross-country panel data. Yet, a few of our varables deserve some dscusson. Snce the current account balance s assocated wth nternatonal captal flows, we focus on fnancal openness n constructng the openness varable q. Ths varable s based on the ndex constructed by Chnn and Ito (2007), whch s the frst prncple component of three bnary varables on captal controls recorded n the IMF's Annual Report on Exchange Arrangements and Exchange Restrctons. Usng ths ndex nstead of a trade openness measure also has the advantage that t s n- 8 Ths s based on the results of Fsher and Im-Pesaran-Shn panel unt root tests. 9 However, for non-gmm estmatons such as fxed effects models, we have expermented wth ncludng ncome as well. 7
sttuton-based and therefore less susceptble to endogenety problems and varaton due to busness cycles. Notwthstandng, we also refer to a standard trade openness measure for robustness check. The usual proxes for fnancal development or actvty rely on money and credt volumes. 10 However, they suffer from a number of shortcomngs that cast doubt on ther usefulness n crosscountry and nter-temporal comparsons. 11 We therefore refer to a new mult-ndcator measurement of fnancal actvty that captures not only the degree of monetzaton or fnancal ntermedaton, but n addton the share of resources a socety devotes to run ts fnancal system. In partcular, our measurement conssts of the followng four ndcators: the share of the labour force employed n the fnancal system; the number of banks and branches per capta; the share of the fnancal system n GDP; and the tradtonal measure M2/GDP. Whle each of these varables ndvdually s plagued wth a host of valdty problems, we trust that collectvely they can be transformed nto a reasonably relable measure for the ntended noton of fnancal actvty. To that end, the factor scores of the frst prncpal component of the four varables are taken as our numercal estmates for fnancal development. 12 Ths measure stands for a resource-based concept of fnancal development. Ths noton of fnancal development s thus dfferent from the common noton of fnancal depth, as t sgnfes a real rather than a monetary phenomenon. For nsttutonal qualty and poltcal stablty (combned), we refer to the poltcal rsk ndex from the Internatonal Country Rsk Gude. A hgher value of ths ndex mples better nsttutonal qualty and hgher poltcal stablty. For smplcty, we refer to the varable as nsttutonal qualty. The coverage of the panel results from a pragmatc approach. We started wth the broadest cross-secton allowed for by data avalablty; yet to curb samplng effects due to excessve unbalancedness of the panel, we excluded those countres for whch data are avalable for recent years 10 Both Gruber and Kamn (2007) and Chnn and Ito (2007) use the rato of prvate credt to GDP, expressed as a devaton from ts GDP-weghted sample means as a proxy for fnancal development. Legg et al. (2007) use the annual stock market turnover as a proporton of share market captalsaton as a proxy for fnancal depth. 11 The standard varable s M2/GDP (e.g. see Kng and Levne, 1993). However, due to defntonal nstablty and dffcultes n nternatonal comparablty, the selecton of a sutable monetary aggregate creates a serous problem (Sklos and Barton, 2001). Moreover, monetary aggregates may be hghly msleadng as they ndcate monetsaton rather than fnancal sophstcaton (Lynch, 1996). Another standard proxy s credt wthn the prvate sector/gdp, whch s used by Kng and Levne (1993). The measure, however suffers from the fact that hghly benefcal credts are lumped together wth non-performng loans. In addton, ths varable refers to bank-based fnance rather than total ntermedaton, and thereby may be a msleadng ndcator of fnancal development n countres where a substantal part of nvestment s rased through the prmary captal market. 12 Specfcally, the data of these four varables for all countres and nne ponts n tme (1960, 1965,..., 2000) were pooled nto a panel, and the frst prncpal component was extracted. The frst component already accounts for 75% of total varance, and all communaltes are 0.69 or hgher, ndcatng a one-dmensonal data space. For a comparable approach to measure fnancal development, see Graff (2005). 8
only. 13 Moreover, followng common practce n cross-country studes, we dsregarded very small countres, 14 as ther ntegraton nto the global economy can be assumed to dffer substantally from larger economes, as well as OPEC countres whose natonal ncome derves manly from exportng crude ol. The remanng 84 countres are lsted n the Appendx. The longtudnal dmenson of the panel conssts of yearly data that gven the coverage n most nternatonal data sets referred to n ths lterature start n 1960. We extended the data collecton to 2006, resultng n a maxmum of 47 yearly observatons. 15 As can be expected, there s a consderable number of mssng observatons for the maxmum of 3948 data ponts. The dataset s hence an unbalanced panel. A typcal sample sze for our regressons comprses slghtly more than 2700 observatons. 16 The most severe constrant on the sample sze comes from the nsttutonal qualty measure and then the labour partcpaton rate. Includng these two varables means a loss of more than 1000 observatons. Therefore, we shall frst estmate our models wthout these two varables to take advantage of the maxmum sze of the dataset, and then add them to the models to test for the senstvty of our estmates. 5. EMPIRICAL RESULTS Whle our objects of nterest are the demographc effects on the current account balance, a premse of the paper s that our suggested framework s theoretcally sounder than the conventonal one. Nonetheless, whether the new framework would lead to dfferent fndngs n practce s a dfferent matter. Thus, n what follows we focus on the comparson between the results regardng demography obtaned usng the new and conventonal frameworks respectvely. Although the new framework s more elaborated than the conventonal one, ther comparson s straghtforward once we notce that both can be represented n smlar ways: CA = a + Xˆ α + e (new) (11) t, 0 t, 1 t, CA = b + X β + e (conventonal), (12) 0 1 13 Ths s manly because these countres became ndependent enttes only recently, partcularly after the collapse of the Sovet Unon and Yugoslava. 14 Our cut-off s 1 mllon nhabtants durng any year from 1960-2004 as recorded n the Penn World Tables, Mark 6.1. 15 Notce that we delberately refran from pushng the sample perod to the latest avalable years (2009 for most countres), as ths would nclude the 2008/2009 recesson as well as the bubble economy of 2007-2008. Durng these years, current accounts were frst nflated and later plunged as never before snce the end of World War II, and ths spectacular varance was certanly not drven by the long-term determnants, n whch we are nterested. 16 For a strctly balanced panel, the sample coverage drops to 75 countres and 22 years from year 1985 to 2006. 9
where X ˆ º q Z ( X - RWX ) can be consdered as a transformed verson of X. Snce X ˆ and have dfferent means and standard devatons, to make the comparson of ther coeffcents meanngful, they are standardsed before enterng the regressons. X 5.1 Fxed effects estmatons wthn the conventonal framework We frst estmate the conventonal model usng two-way fxed effects (.e. country and year fxed effects). The country fxed effects are to control for tme nvarant country heterogenety lke geography or culture, and the year fxed effect are to control for country common but tme varyng factors lke the global busness cycle. The results are reported n Table 2. Fgures n parentheses are robust standard errors. [Table 2] When only the two demographc varables are ncluded n the model (2a), both coeffcents pont estmates are negatve, but nsgnfcant at standard levels. Recall that whle most of the lterature assumes an expected negatve sgn for the two demographc varables based on the lfe cycle theory, the lterature revew n secton 2 suggests that a more comprehensve consderaton wll lead to no conclusve presumpton regardng ther sgns. Accordngly, we assess statstcal sgnfcance for these varables wth two-sded tests. In model (2b), a number of control varables are added to the model, ncludng the rural populaton share, average years of schoolng, fnancal development, ncome growth, the output gap and fnancal openness. Now old-age dependency becomes sgnfcant at the 10% level, retanng the negatve sgn. In model (2c), two more control varables are added, namely the labour partcpaton rate and nsttutonal qualty. Ths cuts the sample sze by more than one-thrd. However, the fndngs regardng the two demographc varables reman the same n qualtatve terms, albet that the coeffcent of old-age dependency becomes somewhat larger. Thus, nterpretng these results mechancally, an ncrease n the old-age dependency rato by one standard devaton may reduce the current account balance by some 0.4 standard devatons. In what follows, we take ths as the baselne model for the senstvty analyses. We also estmated the baselne model usng random rather than fxed effects (2d). The results are smlar to those of the fxed effects model n qualtatve terms. However, the Hausman test strongly rejects the null hypothess that the random effects estmates are consstent. Thus, we mantan the fxed effects framework. 10
As a robustness check, n model (2e), we add yet another demographc varable, namely lfe expectancy, to the baselne model. The reason of ncludng ths varable s that populaton ageng s charactersed not only by hgher old-age dependency, but also by rsng lfe expectancy. 17 Snce these two aspects are hghly correlated, whle havng opposte expected effects on savng, omttng lfe expectancy may lead to based estmaton for old-age dependency. On the other hand, snce the nformaton content of the three demographc varables overlaps substantally, ncludng all of them mples multcollnearty problems. Despte ths concern, the results are almost dentcal to those of the baselne model (.e. 2c), and the lfe expectancy varable s not sgnfcant at any conventonal level. Another robustness check s to nclude ncome n the model (2f). Although ths varable s non-statonary, that s less an ssue for the fxed effects estmatons than for GMM. The results reman vrtually the same as n the baselne case. Overall, the fndngs based on the conventonal framework ndcate that after controllng for a reasonable number of confounders both youth and old-age dependency appear to have negatve effects on the current account balance, but only the effect of the old-age dependency s statstcally sgnfcant. Ths would therefore suggest that populaton ageng ndeed has dscernble negatve mpacts on the current account balance, whle a queston mark s n order regardng youth dependency. Prevous studes also tended to fnd negatve effects for both age dependency ratos, 18 even though not all record both of them as statstcally sgnfcant. So, n general, our results are consstent wth prevous fndngs based on a smlar framework, whch s reassurng as a startng pont. Let us now proceed to our new specfcaton. 5.2 Fxed effects estmatons wthn the new framework Wthn the new framework, we start wth re-estmatng the baselne model (3a). The results are reported n Table 3. Both dependency ratos are now nsgnfcant at even most moderate standard levels,.e. referrng to statstcal nference, the coeffcents are not dstngushable from zero. We also estmated the model usng random effects (3b) but, once agan, the Hausman test strongly rejects the null hypothess that the random effects estmates are consstent. We also repeat the two ro- 17 In L, Zhang and Zhang (2007), the fertlty rate nstead of youth dependency s used as a measure of demography, along wth lfe expectancy and old-age dependency. Yet, n low-ncome countres, nfant and chld mortalty rates are often regrettably hgh, so the youth dependency rate s more accurate to capture the demographc structure, as t reflects both fertlty and mortalty. 18 However, n some of ther estmatons, Chnn and Prasad (2003) record a postve effect of old-age dependency rato on the current account balance. 11
bustness tests of addng lfe expectancy (3c) and ncome (3d), respectvely, to the model and fnd the results remanng the same as n the baselne case. [Table 3] The key message we get from these results s that the new framework leads to very dfferent conclusons regardng the effects of demographc factors on the current account balance. Specfcally, compared to the results from the conventonal framework, the new results do not support the hypothess that populaton ageng would have unambguous and negatve mpacts on the current account balance. 5.3 Dynamc panel estmatons wthn the new framework Gven that the current account balance s charactersed by strong nerta, 19 omttng ts lagged values n the fxed effects models could result n nconsstent estmaton of the coeffcents for demography and other explanatory varables (Bond 2002). To address ths ssue, we nclude lagged values of the current account balance nto the model and estmate t usng GMM estmators. The Dfference (DIFF) GMM estmator followng Holtz-Eakn, Newey and Rosen (1988) and Arellano and Bond (1991) nvolves frst dfferencng the equaton to remove the cross sectonal fxed effects: å D CA = D Xˆ α + D CA + D v + D e (new) (13) g t, t, 1 k= 1 k t, - k t t, and then usng lagged level varables as nstruments. The conventonal framework s obtaned by substtutng X for X ˆ. t, t, Arellano and Bover (1995) and Blundell and Bond (1998) mprove the DIFF GMM estmators by smultaneously estmatng the level equaton usng the lagged frst-dfference of a varable as an nstrument. The modfed verson s commonly referred to as the System (SYS) GMM estmator. Post-estmaton Dfference-n-Hansen tests are performed to assess f the addtonal moment condtons wth the SYS GMM estmator as compared to the DIFF GMM estmator are vald. Both estmators requre the frst-dfferenced errors to be serally uncorrelated, whch can be tested postestmaton. The number of lags for the dependent varable s extended untl ths condton s fulflled. The estmatons are performed usng a two-step process, as ths does not requre homoskedastcty of the error term, and the standard errors are corrected for small sample bas followng Wndmejer (2005). Snce the two-step GMM estmaton presupposes that the errors are not corre- 19 For our dataset, the frst order autocorrelaton of the current account balance s 0.73. 12
lated across countres, the year fxed effects are retaned n the model to remove unversal tmerelated shocks from the errors (Roodman 2009a). The youth dependency rato, the old-age dependency rato, the rural populaton share, average years of schoolng, and fnancal openness are treated as exogenous, as they are slowly evolvng stock or level varables. Fnancal development s treated as a predetermned varable, meanng that ts frst and deeper lags are used as nstruments. The remanng varables are treated as endogenous, meanng that only ther second and deeper lags are used as nstruments. Usng too many nstruments can lead to downward bas of the standard errors. Therefore, we lmt the number of lags to two n generatng nstruments, unless a post-estmaton dagnostc test results n a demand for longer lags, as well as collapsng the moment condtons. 20 The results are reported n Table 4. All estmatons reported n the table pass the autocorrelaton and over-dentfcaton tests descrbed above. For the conventonal model (4a), the lagged value of the current account balance s sgnfcant and shows a szable coeffcent, vndcatng ts ncluson. 21 The ncorporaton of short run dynamcs n the model sees some substantal changes to the results compared to the statc model. Both dependency ratos are now sgnfcantly negatve at the 5% level. Ths fndng thus would delver support for the lfecycle hypothess descrbed n the ntroducton. [Table 4] For the new model (4b), once agan, the lagged value of the current account balance s sgnfcant and szable. Moreover, there are notceable changes to the demographc effects as compared to the statc models. Youth dependency s now sgnfcantly negatve. On the other hand, oldage dependency has a postve pont estmate, but remans nsgnfcant, as n the other specfcatons based on our suggested new framework. We furthermore conducted a number of robustness tests for the dynamc panel model. To keep the dscusson focused, we only go through the results regardng the new framework. 22 Frstly, when lfe expectancy s added to the model (4c), the youth dependency rato gets nsgnfcant. Ths s probably due to multcollnearty. Secondly, snce some szeable OECD countres tend to have comparatvely low savng ratos, and at the same tme attract a lot of nternatonal captal, one may also queston f ther current account balances respond to demographc changes dfferently from other countres. To examne ths, we splt the sample nto an OECD and a non-oecd subsample. However, as n the OECD subsam- 20 For a detaled dscusson, see Roodmand (2009b). 21 Results from dagnostc tests suggest that one lag term s suffcent. 22 The other results can be obtaned from the correspondng authors upon request. 13
ple, there s severe multcollnearty between regressors, ncludng the youth dependency rato, we can only report meanngful results for the non-oecd subsample (4d). Importantly, excludng the OECD countres does not change the man fndngs from the baselne case n that only the youthage dependency rato s sgnfcantly negatve. Thrdly, we dvde the dataset nto two 10-year perods, one from 1986 to 1995, and the other one from 1996 to 2005. 23 There are two consderatons leadng to ths robustness check. Frstly, t allows us to examne whether the demographc effects on the current account balance change over tme. Secondly, the SYS GMM estmator s desgned for datasets of large N, but small T. For the perod of 1986-1995 (4e), both dependency ratos are nsgnfcant, whereas for the perod of 1996-2005 (4f), the key results are agan the same as n the full sample n that only the youth dependency rato s sgnfcant and negatve. Lastly, we refer to a standard measure of trade openness n place of fnancal openness n constructng X %. The most commonly used measure of trade openness s total trade normalsed wth GDP. Ths measure, however, s senstve to the busness cycle. Therefore, we extract the long run trend usng the Hodrck-Prescott flter and treat t as a predetermned rather than exogenous varable. The youth dependency rato retans the negatve sgn, as before, but ts standard error has ncreased somewhat such that ts sgnfcance level falls out of the two sde 10% range (p=0.12) (4g). 5.4 Dscusson What can we take from the reported regressons? Although the detals vary from estmaton to estmaton, as expected, a number of general observatons regardng the demographc effects on the current account balance s n order. Frstly, accountng for the nerta of the current account balance makes substantal dfferences to the results wthn the new and conventonal frameworks. Accordngly, when dscussng the prncpal fndngs on the demographc effects on the current account balance, t s mportant to emphasse whether the results are based on statc or dynamc models. Secondly, for ether the statc or the dynamc models, our results based on the new framework are notceably dfferent from those based on the conventonal approach. Ths suggests that at the very least, accountng for general equlbrum, relatve sze and openness s mportant n modellng the current account balance; and that at the very most, all prevous estmatons based on the conventonal framework are based. 23 There are nsuffcent data to estmate the full model before 1985. 14
Thrdly, for ether the statc or the dynamc models, the results based on the new framework ndcate that the effect of populaton ageng on the current account balance s much more subdue than reported n prevous studes usng the conventonal framework. As dscussed n the lterature revew, ths s not contradctory to the theoretcal arguments, because populaton ageng could affect both savng and nvestment, leavng ts effect on excess savng over nvestment (.e. the current account balance) ambguous. Furthermore, people movng toward retrement may save more, and retrees may dssave more slowly due to longevty uncertanty. In ths regard, t s plausble for populaton ageng to have ether a postve or a negatve mpact on the current account balance, or no dscernble mpacts at all as found n ths study. 6. CONCLUSION Ths paper re-examnes the effects of demography on the current account balance. A key departure of the paper from, and hence ts contrbuton to, the current lterature s ts consderaton of the general equlbrum nature of the current account balance n a world of heterogeneous economc sze and openness. Our modellng framework dffers from the conventonal one n three aspects. Frstly, t recognses that n a general equlbrum settng, t s the relatve pace of demographc changes across countres rather than the absolute pace of changes n a partcular country that contrbutes to determnng ts current account balance. Secondly, t acknowledges the fact that a country can nfluence others and lkewse be nfluenced by others only up to the extent that t s economcally open. Thrdly, t accounts for the fact that the nfluence of one country on another depends on ther relatve szes. Ths new framework s put to test referrng to a large panel dataset. The emprcal strategy s to estmate the new and conventonal models and contrast ther dfferences. It s demonstrated that the new framework leads to notceable changes n some conclusons based on the conventonal one. The new evdence ndcates that populaton ageng as measured by old-age dependency does not have dscernble mpacts on the current account balance, and thus suggestng that dfferent fndngs n the prevous lterature may be due to the neglgence of the general equlbrum, sze and openness factors. In a nutshell, the effects of demographc factors on the current account balance may comprse more and sometmes conflctng drvng forces than prevously consdered, whch calls for more research along the lnes suggested here. Last but not the least, although the current study focuses on demography, the key ratonale behnd the new framework s qute general and hence applcable to emprcal studes of many other open economy ssues. 15
REFERENCES Arellano, Manuel and Stephen Bond (1991). Some tests of specfcaton for panel data: Monte Carlo evdence and an applcaton to employment equatons. Revew of Economc Studes, 58(2): 277-297. Arellano, Manuel. and Olympa Bover (1995). Another look at the nstrumental varables estmaton of error components models. Journal of Econometrcs, 68(1): 29-51. Attanaso, Orazo P., Sagr Ktao, Govann L. Volante (2006) Quantfyng the effects of the demographc transton n developng economes. Advances n Macroeconomcs 6(1): Artcle 2. Barro, J. Robert, and Jong-Wha Lee (2001) Internatonal data on educaton attanment: updates and mplcatons. Oxford Economc Papers 53(3): 541-563. Bond, Stephen (2002), Dynamc panel data models: A gude to mcro data methods and practce. CEMMAP Workng Paper CWP09/02. The Insttute for Fscal Studes, Department of Economcs, Unversty College of London. Blundell, Rchard. and Stephen Bond (1998). Intal condtons and moment restrctons n dynamc panel data models. Journal of Econometrcs, 87(1): 115-143. Chnn, Menze D., and Hro Ito (2007) Trade account balances, fnancal development and nsttutons: assayng the world savng glut. Journal of Internatonal Money and Fnance 26(4): 546-69. Chnn, Menze D., and Eswar S. Prasad (2003) Medum-term determnants of current accounts n ndustral and developng countres: an emprcal exploraton. Journal of Internatonal Economcs 59(1): 47-76. Cooper, Rchard N. (2008) Global mbalances: globalzaton, demography, and sustanablty. Journal of Economc Perspectve 22(3): 93-112. Domej, Davd, and Martn Flodén (2006) Populaton agng and nternatonal captal flows. Internatonal Economc Revew 47 (3): 1013-1032. Edwards, Sebastan (1996) Why are Latn Amerca's savng rates so low? An nternatonal comparatve analyss. Journal of Development Economcs 51: 5-44. Ferol, Mchael (2003) Captal flows among the G-7 natons: A demographc perspectve. Fnance and Economcs Dscusson Seres 2005-54, Board of Governors of the Federal Reserve System. Graff, Mchael (1999) Educaton mbalance, soco-economc nequalty, poltcal freedom, and economc development Journal of Economc Development, 24(2): 1-18 Graff, Mchael (2005) Soco-Economc Factors and the Fnance-Growth Nexus. European Journal of Fnance 11 (3): 183-205. Gruber, Joseph W., and Steven B. Kamn (2007) Explanng the global pattern of current account mbalances. Journal of Internatonal Money and Fnance, 26(4): 500-22. Hggns, Matthew (1998) Demography, natonal savngs and nternatonal captal flows. Internatonal Economc Revew, 39 (2): 343-69. Hggns, Matthew and Jeffrey G. Wllamson (1997) Age structure dynamcs n Asa and dependency on foregn captal. Populaton and Development Revew, 23 (2): 261-93. Holtz-Eakn, Douglas, Whtney Newey and Harvey S. Rosen (1988). Estmatng vector autoregressons wth panel data. Econometrca, 56(6): 1371-1395. 16
Km, Soyoung and Jong-Wha Lee (2007) Demographc changes, savng, and current account n East Asa. Asan Economc Papers 6 (2): 22-53. Kng, Robert. G. and Ross Levne (1993) Fnance and Growth. Schumpeter Mght Be Rght. Quarterly Journal of Economcs, 108 (3): 717-738. L, Hongbn, Je Zhang and Junsen Zhang (2007) Effects of longevty and dependency rates on savng and growth: Evdence from a panel of cross countres. Journal of Development Economcs, 84 (1): 138-154. Legg, Anthony, Naln Prasad and Tm Robnson (2007) Global mbalances and the global savng glut a panel data assessment. Research Dscusson Paper, Reserve Bank of Australa, 2007-11. Lynch, Davd (1996) Measurng Fnancal Sector Development: A Study of Selected Asa- Pacfc Countres. The Developng Economes, 34 (1): 3-33. Masson, Paul R., Tamn Bayoum, Hossen Same (1998) Internatonal evdence on the determnants of prvate savng. The World Bank Economc Revew, 12(3): 483-501. Modglan, Franco (1970) The lfe cycle hypothess of consumpton. n: W. A. Elts, M. F. G. Scott, and J. N. Wolfe, eds., Inducton, growth, and trade: Essays n Honour of Sr Roy Harrod, Oxford Unversty Press: 197 225.Obstfeld, Maurce, and Kenneth Rogoff (2000) The sx major puzzles n nternatonal macroeconomcs: s there s a common cause? NBER Macroeconomcs Annual. The MIT Press. Roodman, Davd (2009a) How to do xtabond2: an ntroducton to dfference and system GMM n Stata. The Stata Journal, 9 (1): 86-136. Roodman, Davd (2009b) A note on the theme of too many nstruments. Oxford Bulletn of Economcs and Statstcs, 71 (1): 135-158. Sachs, Jeffrey, and Andrew Warner (1995) Economc reform and the process of global ntegraton. Brookngs Papers on Economc Actvty 1, pp.1-118. Sklos, Perre L. and Andrew G. Barton (2001) Monetary Aggregates as Indcators of Economc Actvty n Canada: Emprcal Evdence. Canadan Journal of Economcs, 34(1): 1-17. Wndmejer, Frank (2005). A fnte sample correcton for the varance of lnear effcent two-step GMM estmators. Journal of Econometrcs, 126 (1):25-51. 17
Table 1 Data defntons, sources and summary statstcs Varable Defnton Source No. of Mean Std. dev. obs. The current account balance The current account balance to GDP rato (n %) Penn World Tables 6.2 (PWT) 2530-2.83 6.59 Youth dependency rato Populaton aged 0-14 to populaton aged 15-64 World Development Indcators (WDI) 2530 0.61 0.25 Old-age dependency rato Populaton aged 65+ to populaton aged 15-64 WDI 2530 0.11 0.07 Lfe expectancy Lfe expectancy at brth WDI 2530 64.55 11.62 Rural populaton share Share of rural populaton n total populaton (n %) WDI 2530 48.29 24.82 Years of schoolng Average years of schoolng Barro and Lee (2001) 2530 5.59 2.97 Fnancal development The frst prncple components of four fnancal development ndcators Graff (2005) 2528 0.32 1.11 Income growth rate Growth rate of ncome per capta (%) PWT 2516 1.77 4.51 Output gap Output gap (n % of potental GDP) Relatve devaton of annual GDP (PWT) from potental output, where the latter s derved by HP-flterng ( l = 100 ) the logged emprcal GDP seres. 2522 0.12 4.16 Labor partcpaton rate Labor partcpaton rate (n %) WDI 2167 69.11 9.26 Insttutonal qualty Composte poltcal rsk ndex (0-100, a larger number means better nsttutonal qualty and hgher poltcal stablty) Internatonal Country Rsk Gude 1688 64.63 15.98 18
Table 2 Fxed effect estmatons wthn the conventonal framework; dependent varable: current account balance/gdp 2a 2b 2c 2d 2e 2f Youth dependency -0.122-0.005-0.010-0.214-0.002-0.102 (0.15) (0.20) (0.29) (0.16) (0.28) (0.32) Old-age dependency -0.064-0.296* -0.405** -0.203* -0.410** -0.425** (0.14) (0.17) (0.20) (0.12) (0.20) (0.18) Rural populaton share 0.101 0.790 0.147 0.784 0.647 (0.43) (0.52) (0.14) (0.52) (0.48) Years of schoolng 0.683*** 0.547** 0.357*** 0.549** 0.551** (0.25) (0.25) (0.14) (0.25) (0.26) Fnancal development 0.297 0.054 0.182 0.052-0.037 (0.21) (0.18) (0.16) (0.18) (0.15) Income growth 0.009 0.034 0.046 0.035 0.037 (0.04) (0.05) (0.05) (0.05) (0.05) Output gap -0.066** -0.074* -0.081* -0.074* -0.086* (0.03) (0.04) (0.04) (0.04) (0.04) Fnancal openness -0.053-0.105-0.076-0.104-0.101 (0.08) (0.09) (0.07) (0.09) (0.08) Labor partcpaton rate 0.143 0.034 0.139 0.127 (0.13) (0.08) (0.13) (0.14) Insttutonal qualty -0.067-0.078-0.073-0.051 (0.08) (0.07) (0.08) (0.08) Lfe expectancy 0.061 (0.24) Income 0.349 (0.34) R-squared 0.071 0.103 0.084 0.084 0.088 N 2739 2551 1686 1686 1686 1686 *, **, *** denote sgnfcance at 10, 5 and 1% respectvely. Fgures n parentheses are robust standard errors. 19
Table 3 Fxed effect estmatons wthn the new framework; dependent varable: current account balance/gdp 3a 3b 3c 3d Youth dependency 0.045-0.231 0.088 0.047 (0.23) (0.21) (0.22) (0.23) Old-age dependency -0.193-0.022-0.149-0.209 (0.27) (0.21) (0.27) (0.33) Rural populaton share 0.141 0.150 0.197 0.139 (0.26) (0.18) (0.24) (0.26) Years of schoolng 0.628* 0.325 0.551* 0.603 (0.34) (0.24) (0.33) (0.41) Fnancal development -0.151-0.114-0.165-0.163 (0.30) (0.32) (0.31) (0.26) Income growth -0.031-0.025-0.033-0.031 (0.05) (0.05) (0.05) (0.05) Output gap -0.056-0.056-0.054-0.057 (0.05) (0.05) (0.05) (0.05) Fnancal openness -0.054 0.062-0.062-0.050 (0.10) (0.08) (0.10) (0.10) Labor partcpaton rate -0.054-0.005-0.037-0.062 (0.09) (0.08) (0.10) (0.13) Insttutonal qualty -0.149-0.164* -0.179-0.144 (0.12) (0.09) (0.12) (0.11) Lfe expectancy 0.220 (0.28) Income 0.047 (0.40) R-squared 0.068 0.070 0.069 N 1619 1619 1619 1619 *, **, *** denote sgnfcance at 10, 5 and 1% respectvely. Fgures n parentheses are robust standard errors. 20
Table 4 Dynamc panel estmatons; dependent varable: current account balance/gdp Youth dependency Old-age dependency Rural populaton share Years of schoolng Fnancal development 4a 4b 4c 4d 4e 4f 4g -0.240** -0.468** -0.298* -0.408** 0.217-0.422** -0.346 (0.12) (0.21) (0.17) (0.18) (0.48) (0.20) (0.22) -0.224** 0.178 0.182-0.097-0.015 0.060 0.010 (0.11) (0.19) (0.24) (0.25) (0.43) (0.24) (0.15) -0.063-0.059-0.042-0.265 0.190-0.003-0.258 (0.12) (0.18) (0.18) (0.25) (0.22) (0.28) (0.25) 0.017 0.073 0.009 0.754 0.286 0.021 0.034 (0.12) (0.13) (0.14) (0.54) (0.32) (0.20) (0.25) 0.149-0.506-0.482-1.087** 0.286-0.337-0.679 (0.11) (0.38) (0.50) (0.47) (0.61) (0.42) (0.55) Income growth -0.077 0.028 0.033* 0.036 0.014-0.136 0.047 (0.08) (0.02) (0.02) (0.02) (0.01) (0.18) (0.06) Output gap -0.032-0.026-0.027-0.036 0.059 - -0.019 0.147*** (0.05) (0.04) (0.04) (0.04) (0.04) (0.05) (0.04) Labor partcpaton rate 0.276** 0.165** 0.206 0.296** 0.123 0.151 0.249 Insttutonal qualty Fnancal openness Lagged current account balance (0.13) (0.08) (0.15) (0.13) (0.20) (0.19) (0.16) -0.050-0.200* -0.199-0.299* 0.052-0.082-0.192 (0.09) (0.11) (0.16) (0.16) (0.41) (0.17) (0.18) -0.050 0.058 0.048 0.041 0.026 0.099 (0.05) (0.04) (0.04) (0.07) (0.04) (0.06) 0.600*** 0.574*** 0.578*** 0.497*** 0.614*** 0.488*** 0.596* ** (0.06) (0.07) (0.07) (0.07) (0.07) (0.09) (0.08) Lfe expectancy 0.283 (0.40) Trade openness -0.051 (0.17) R-squared N 1686 1619 1619 1205 697 727 1674 *, **, *** denote sgnfcance at 10, 5 and 1% respectvely. Fgures n parentheses are robust standard errors. 21
Appendx: Country coverage Angola Ecuador Lesotho Poland Argentna Egypt Madagascar Portugal Australa El Salvador Malaysa Rwanda Austra Ethopa Mal Senegal Bangladesh Fnland Maurtana Serra Leone Belgum France Maurtus Sngapore Benn Germany Mexco Span Bolva Ghana Morocco Sr Lanka Brazl Guatemala Mozambque Sudan Bulgara Hat Nepal Sweden Burund Honduras Netherlands Swtzerland Cameroon Hungary New Zealand Syra Canada Inda Ncaragua Thaland Chle Indonesa Nger Togo Chna Israel Ngera Trndad &Tobago Colomba Italy Norway Tunsa Congo Jamaca Pakstan Turkey Costa Rca Japan Panama Uganda Cote d`ivore Jordan Papua New Gunea Unted Kngdom Denmark Kenya Peru Unted States Domncan Republc Korea Phlppnes Uruguay 22