THE INTRODUCTION OF EMERGING CURRENCIES INTO A PORTFOLIO: TOWARDS A MORE COMPLETE DIVERSIFICATION MODEL

Transcription

1 THE INTRODUCTION OF EMERGING CURRENCIES INTO A PORTFOLIO: TOWARDS A MORE COMPLETE DIVERSIFICATION MODEL Sophe Brana, Stéphane Prat To cte ths verson: Sophe Brana, Stéphane Prat. THE INTRODUCTION OF EMERGING CURRENCIES INTO A PORTFOLIO: TOWARDS A MORE COMPLETE DIVERSIFICATION MODEL. 9. <hal- 6658v> HAL Id: hal Submtted on 8 Nov HAL s a mult-dscplnary open access archve for the depost and dssemnaton of scentfc research documents, whether they are publshed or not. The documents may come from teachng and research nsttutons n France or abroad, or from publc or prvate research centers. L archve ouverte plurdscplnare HAL, est destnée au dépôt et à la dffuson de documents scentfques de nveau recherche, publés ou non, émanant des établssements d ensegnement et de recherche franças ou étrangers, des laboratores publcs ou prvés.

2 The ntroducton of emergng currences nto a portfolo: Towards a more complete dversfcaton model Sophe Brana, Stéphane Prat LAREFI Workng Paper CR9-EFI/ 9 LAREFI Unversté Montesqueu-Bordeaux IV Bâtment Recherche Econome Avenue Léon Dugut 68 Pessac

3 Abstract We draw on portfolo theory and nternatonal dversfcaton n order to analyse strateges allowng to reduce emergng economes exposure to exchange-rate rsk. We show n partcular that t may be effcent for an nvestor, n terms of maxmsng the return-to-rsk rato, to buld up a portfolo of emergng-country assets denomnated n local currency - unhedged aganst currency rsk - compared wth a strategy ncludng emergng-country securtes denomnated n foregn currences. Ths strategy would lead to a reducton n the orgnal sn (.e. the nablty of emergng economes to borrow n local currency), and de facto to a reducton n currency msmatches n the balance sheets of emergng economes. Résumé Nous nous appuyons sur la théore des portefeulles et de la dversfcaton nternatonale afn d analyser les stratéges permettant de rédure l exposton des économes émergentes au rsque de change. Nous montrons en partculer qu l peut être effcent pour un nvestsseur, en termes de maxmsaton du couple rendement/rsque, de consttuer un portefeulle d actfs émergents lbellés en monnae natonale non couvert contre le rsque de change par rapport à une stratége qu nclurat dans le portefeulle des ttres émergents lbellés en devses. Cette stratége condurat à une dmnuton du péché orgnel (.e. l ncapacté des économes émergentes à emprunter en monnae natonale), et de fat à une réducton des déséqulbres en devses dans les blans des économes émergentes. JEL Classfcaton : G ; E ; F. Mots-clés : Internatonal portfolo dversfcaton, Orgnal Sn, Emergng countres, Downsde rsk..

4 . Introducton Balance sheet msmatches n emergng economes, n partcular currency msmatches, have played a fundamental role n the repeated crses that have ht these economes for more than ten years. In the late 99s, Echengreen and Hausmann (999) hghlghted the fact that a major source of fnancal fraglty n emergng economes was related to the composton n foregn currences of ther external debt. Wth what they called the orgnal sn theory, the two authors showed that emergng economes were more vulnerable to fnancal crses than ndustralsed economes because of ther nablty to borrow n nternatonal captal markets n ther own currency. Indeed, the weght of outstandng external commtments denomnated n foregn currences ncrease fnancal vulnerablty because of the hgh exposure to foregn exchange and nterest rate rsk of these economes. It can trgger foregn exchange crses. For Echengreen et al (, 7), the orgnal sn prmarly reflects the characterstcs of nternatonal fnancal markets. In partcular, the shortfall n hedgng possbltes and the exstence of transacton costs result n nternatonal nvestors gvng ther preference to a small number of currences when buldng ther portfolo. The portfolo allocaton ntated by nternatonal nvestors s thus combned wth a transfer of currency rsk to emergng economes, whch are badly prepared to shoulder ths rsk. We show, however, wthn the framework of an approach based on portfolo dversfcaton, that emergng economes mght free themselves from such a rsk n order to mprove ther reslence to shocks. To do so, we show that a strategy consstng n ncludng emergng-country assets, denomnated n the local currency and not hedged aganst currency rsk n the portfolo of a foregn nvestor s not necessarly rsker that an emergng-country asset allocaton denomnated n foregn currences. Ths s because the potental reducton of market rsk (currency rsk), va the dversfcaton of portfolos composed of emergng securtes denomnated n the local currency, can be hgher than the potental n terms of reducton n credt rsk (or default rsk) related to an nternatonal portfolo composed of emergng securtes denomnated n foregn currences, thereby steerng the structure of ndebtedness of emergng countres towards a structure n the local currency that would be more stable and less rsky. The rest of the artcle unfolds as follows: secton presents the theoretcal model we draw upon, by justfyng notably our choce to use an asymmetrc measure of the portfolo s rsk. Secton presents the methodology used whle we emprcally assess the varous components of the portfolo s rsk n secton. We draw our conclusons n secton 5.

5 . Theoretcal model We operate wthn the framework of portfolo dversfcaton theory proposed by Markowtz (95, 959) so as to model the return/rsk relatonshp of an nternatonally dversfed portfolo. Intally, Markowtz s model, whch s based on the strong hypothess that economc agents have a quadratc utlty functon, use the standard devaton (or the varance) of returns on securtes to measure a portfolo s rsk. A frst lmtaton of ths measure s that t takes nto account, wthout makng any dstncton between them, both upward and downward devatons of returns n comparson wth the average. Ths s napproprate n terms of assessng the concept of rsk from an nvestor s vewpont because the atttude of nvestors vares accordng to the an of the utlty functon, as nvestors are n partcular more senstve to the losses they ncur than to the gans they make (Campbell and Kräussl, 7) (cf. below). Moreover, the utlsaton of the standard devaton as a measure of rsk supposes normal dstrbuton of returns. However, returns on emergng securtes, n partcular bonds, are charactersed by negatve skewness (Bekaert and Harvey, 997; Bekaert, Erb, Harvey and Vskanta, 998; Burger and Warnock, 7). The utlsaton of the frst- and second-order moments of the dstrbuton of returns n complance wth the portfolo model leads, by consequence n the case of emergng securtes, to a non-optmal asset allocaton (Bawa and Lndenberg, 977; Harlow and Rao, 989; Harlow, 99). Followng Roy (95), varous measures of rsk and, accordngly, varous models have been proposed to take nto consderaton the characterstcs of dstrbutons of returns on emergng securtes, as well as the behavour of nvestors wth respect to rsk, whle mantanng the ntal two-dmenson rsk-return relatonshp (Hwang and Pedersen, ). By defnton, these measures of rsk, so-called downsde rsk measures, take nto account only one part of the dstrbuton of returns rather than the entre dstrbuton. These measures solate dvergences n returns n comparson wth a target return only on the left-hand sde of the dstrbuton (Harlow, 99). Markowtz (959) notably defned the sem-varance as the most robust measure of rsk from a theoretcal vewpont. It evaluates the average squared devatons of returns below a benchmark. Formally, the author suggested evaluatng the sem-varance (SV) of returns n two ways: SV SV m = T = T t = T Max [, ( R Max [, ( TC R t )] TC R t T t = )] () () Where R t stands for the return on securty n perod t and T for all perods t. SV m evaluates the adverse devatons of returns from the average of returns R (sem-varance n comparson wth the average) whle SV TC determnes the adverse devatons of returns from a benchmark rate of return or an arbtrarly chosen target rate TC. Accordng to ths defnton, the sem-varance expresses the fact that nvestors are concerned only wth negatve devaton from a gven and arbtrarly chosen proftablty threshold,.e. nvestors care only about ther potental worst-case returns. The development of alternatve measures f rsk, such as sem-varance, made t possble to determne n a more general framework the lower partal moments of order n defned by (Bawa, 975): TC stands for target rate.

6 T n LPM ( n, TC) = Max[,( TC R t )] () T t= TC stands for the chosen target proftablty rate, T the number of observatons, n the degree of the lower partal moment and R the return on securty n perod t. t From a theoretcal vewpont, the order n of the measure of the Lower Partal Moment (LPM n ) defnes the type of the nvestor s utlty functon that s consstent wth hs degree of rsk averson. The partal moment of order s used for nvestors attracted by rsk (postve dervatve of the utlty functon), whle the partal moment of order suts all the utlty functons of rsk-averse nvestors (postve dervatve of the utlty functon and negatve second dervatve). Lastly, the partal moment of order concerns rsk-averse nvestors who also have a preference for a postve asymmetrcal dstrbuton of returns (postve thrd dervatve) (Harlow, 99; Nawrock, 999). When n = and the benchmark proftablty rate s equal to the average of returns, LPM (, R) represents the prevously presented tradtonal measure of the sem-varance. Generally speakng, the use of lower partal moments allows the restrctve hypotheses of Markowtz s ntal portfolo model to be eased, on the one hand wth respect to nvestors preferences and, on the other hand, wth respect to the propertes of the dstrbuton functons of the assets returns we are lookng at. Ultmately, the average-lower partal moment approach s not only consstent wth the atttude of nvestors wth regard to rsk but remans vald whatever the characterstcs of returns (Harlow, 99). Recently, Jarrow and Zhao (6) and Estrada (7) have shown, n ths respect, that the optmal meanvarance portfolo dffered sgnfcantly from the optmal mean sem-varance portfolo, notably n the case of a bond portfolo (Jarrow and Zhao, 6). Accordng to the authors, the mean-varance framework s effectvely napproprate wth regard to the management of the rsks nherent to ths type of asset. Lastly, we choose to model the portfolo s rsk by the lower partal moment of order by estmatng the negatve devatons of returns from the mean of the dstrbuton (downsde rsk measure va sem-standard devaton ). Formally, the model, ntally descrbed by Bawa and Lndenberg (977), can be defned as follows: Réf Mn LPM p = w Under constrant s N N = j = N = w w w E( R j CoLPM p ) = E( R ( R p ) t and + e, R N = w jt + e = et w j ) > (Non - authorsed short sales) Wth Investors take nto account the skewness of returns on securtes and are generally averse to assets dsplayng negatve skewness,.e. unlkely but hgher potental losses and probable but modest gans. Accordng to Harlow and Rao (989) the pertnent target proftablty s the mean of the dstrbuton of returns. The sem-standard devaton s defned as the square root of semvarance.

7 CoLPM + p ( R j t t w Max(; R Wth + e, R R jt t ) + e Max(; R Rt + Mn(; e e ) = + Mn(; e t j ) Max(; R 7 ) Max(; R e ) j j t R R + Max(; R jt jt t ) + Max(; R ) + Mn(; e R 5 t R ) Mn(; e 6 t ) Mn(; e e ) Mn(; e t j e j e e R / jt the return of asset /j n perod t expressed n the local currency (.e. the borrower s) and the return of the exchange rate 5 correspondng to asset /j over perod t. p e / jt CoLPM R + e, R + e ) represents the co-lower partal moment between the returns ( t jt j on emergng-country assets denomnated n the local currency and exchange rate fluctuatons. The rsk of an nternatonal portfolo unhedged aganst currency rsk s consequently composed of several components: - the sem-standard devaton (or downsde volatlty) of returns emergng-country assets, denomnated n the local currency or n foregn currences, whch compose the portfolo (term ); - the downsde volatlty of exchange rate returns (term ). If we consder the portfolo of securtes denomnated n foregn currences, ths term dsappears (no currency rsk); - (ntra-class) correlatons between downsde fluctuatons n returns on emergng securtes (term ); - (ntra-class) correlatons between downsde fluctuatons n returns on exchange rates (term ). Lkewse, for a portfolo of securtes denomnated n dollars, ths term dsappears; - lastly, (ntra-class) correlatons between downsde movements n returns on emergng securtes and returns on the correspondng exchange rates (term 5). Ths term concerns only the portfolo of emergng-country assets denomnated n local currences. Terms 6 and 7 of expresson (), whch correspond to the covarance between downsde movements n returns on emergng securtes and downsde movements n emergng currences are supposed to be neglgble: ths s because t s assumed that downsde movements n the fluctuatons of currency j (term 6) or (term 7) are hardly correlated to downsde movements n returns of a securty denomnated n currency (term 6) or j (term 7), for j. The prevously presented model enables us to analyse emprcally the advantages of nternatonal dversfcaton for a foregn nvestor. jt ) t ) jt ) () 5 Exchange rate e s defned va an uncertan quotaton from the borrower s pont of vew,.e. as unts of the estc currency of asset /j per unt of the reference currency.

8 . Methodology The purpose of our approach s to carry out an arbtrage between two types of strateges: a strategy of nvestments n emergng currences n comparson wth a strategy of nvestment n dollars. To do so, we use the EMBIG and ELMI+(LC) ndces publshed by J.P. Morgan for varous emergng countres. EMBI Global ndcators are ndces trackng the proftablty of soveregn assets of emergng countres ( December 99 = ) ssued n nternatonal markets and denomnated n dollars, whle the ELMI+(LC) ndcators are ndces trackng the proftablty of assets of the estc money market of emergng economes ( December 99 = ) denomnated n the local currency. The ELMI+(LC) ndcator corresponds more precsely to the total proftablty of the estc money market nstruments denomnated n the local currency (J.P. Morgan, 997). The EMBIG ndcator, for ts part, refers to the total proftablty of assets ssued by soveregn or quas-soveregn enttes n emergng economes, and concerns only nstruments denomnated n US dollars. As for nstruments ncluded n the ELMI+(LC) ndcator, the assets of the EMBIG ndcator have to meet mnmum crtera n terms of lqudty and accessblty for foregn nvestors (J.P. Morgan, 999). These two ndces, apart from ther currency of denomnaton, are therefore not strctly comparable, as the ELMI+(LC) ndex covers securtes wth a shorter duraton than nstruments elgble for the EMBIG ndex 6. Both, however, refer to ndces trackng the proftablty of emergng debt securtes and are regularly compared as alternatve nvestment strateges (Drjkonngen et al. 6). Moreover they are avalable for a sample of countres for a perod rangng from July 997 to December 7, allowng us to compare the performances of two dversfed portfolos, one denomnated n dollars, the other n emergng currences, over a relatvely long perod 7. The countres ncluded n the sample are South Afrca, Argentna, Brazl, Chna, South Korea, Mexco, the Phlppnes, Poland, Thaland, Turkey and Venezuela 8. An nvestor nterested n exposure to emergng-country markets wll therefore be able to compare two strateges: exposure to the local publc debt or exposure to the local currency. The market of local debt denomnated n foregn currences enables nvestors to gan access to credt rsk on emergng-country markets (as the return s determned by the rsk-free rate of US debt plus a credt spread reflectng default rsk), whle the nvestment n local currency n emergng-country markets exposes them to a greater extent to currency rsk (n addton to credt rsk on the local debt). Our approach breaks down nto three stages. Intally, we assess the downsde rsk of proftablty rates of emergng-country assets denomnated n the local currency and of emergng-country assets denomnated n foregn currences (term of equaton ). Second, we calculate the downsde volatlty, measured by the sem-standard devaton, of returns on emergng currences (term of equaton ). To do so, we use blateral nomnal exchange rates of quoted emergng currences aganst the US dollar. The data come from J.P. Morgan wth respect to the two proftablty ndcators, and from Reuters for the blateral exchange rates. Thrd, we look at the correlatons between downsde movements n proftablty rates of assets (term of equaton ), between downsde fluctuatons n exchange rates (term ) as well as between cross correlatons (term 5) whch correspond to the three other components of a portfolo s rsk. We carry out our study on daly data. 6 The other possble ndces for example, the GBI ndex are avalable only for the recent perod and for a smaller number of countres: Brazl, Mexco, Poland, South Afrca and Thaland snce. Ths means they cannot be drawn upon to buld a real portfolo strategy. 7 Medo et al. (9) estmate the optmal sze of a portfolo s ten assets gven the dversfcaton potental. 8 Avalablty of data has strctly determned choces of countres and the perod studed, although a noteworthy pont s that we wanted to nclude the Asan crss.

9 . Results.. Downsde volatlty of returns and of emergng currences We carry out a comparatve analyss of the downsde volatlty of the returns of the EMBIG and ELMI+(LC) ndcators over the perod July December 7 on daly data (.e.,7 tradng days). Chart presents changes n the asymmetrcal month-on-month (M/M) rsk for the two composte ndcators of returns for all emergng countres. These composte ndcators are calculated as the average of daly returns of each country weghted by ther daly market captalsaton. We can see that the downsde volatlty of returns on estc securtes denomnated n the local currency (ELMI+(LC)) s to a large extent lower than the downsde volatlty of returns on securtes traded n nternatonal markets (EMBIG), even durng a crss perod.,5,,5 Chart. Dow nsde volatlty of asset return EM BIG Composte ELM I+(LC) Composte,5,,5,,,5,5, Sources : JP M organ, Reuters, authors' calculaton, The low volatlty of the ELMI+(LC) ndex s not only due to ts short duraton, but, as we shall see below, also to the fact that the volatltes of the varous currences offset one another (low ntra-class correlaton). If we carry out a study on a country-by-country bass, we can see that the economes wth the hghest downsde volatlty over the entre perod studed here are the ones that have suffered from a fnancal crss (Charts A to F): Argentna, Brazl, Turkey, Thaland and Venezuela. However, the downsde rsk of estc securtes remans lower than the downsde rsk of nternatonal securtes (except for South Korea where the two knds of downsde volatlty are comparable):.687% on average versus.% for Argentna,.% versus.69% for Turkey,.7% versus.5% for Thaland over the entre perod. The downsde rsk for the ELMI+(LC) composte ndcator that covers all countres for ts part stood at.9% over the perod versus.5% for the EMBIG composte ndex (Table ).

10 Table. Downsde volatlty calculatons downsde volatlty(%) EMBIG downsde volatlty(%) ELMI+(LC) downsde volatlty(%) FX downsde volatlty(%) ELMI+(LC) + FX Index composte,5,9,69,58 Sources: JP Morgan, Reuters and Datastream, authors'calculatons

11 Charts A to F: Downsde volatlty of EMBIG and ELMI+(LC) returns (M/M as %) 5 Chart a. Argentna 5,6 Chart b. Chna,6,5,5 EM BI GLOBAL ELM I+(LC),, EM BI GLOBAL ELM I+(LC),,,,,, Sources : Datastream Sources : Datastream, ,, Chart c: Mexco,, Chart d. Phlppnes,,5 EM BI GLOBAL ELM I+(LC),5,, EM BI GLOBAL ELM I+(LC),,,8,8,,,6,6,5,5,,,, Sources : Datastream, , Sources : Datastream, , Chart e. Thaland Chart f. Turkey EM BI GLOBAL ELM I+(LC) EM BI GLOBAL ELM I+(LC) Sources : Datastream Sources : Datastream We subsequently assess the downsde volatlty of emergng currences, quoted aganst the US dollar. After surgng to all-tme hgh levels durng crss perods (notably n 997 durng the Asan crss or between and n Latn Amerca and n South Afrca), the downsde volatlty of exchange rates has notceably declned, n these countres, snce and s now below % n monthon-month terms (Charts A, B and C). The average downsde volatlty of the composte ndex, for ts part, stands at.69, up slghtly over the perod as a whole (Chart ).

12 ,7,6,5,,,, Chart. Downsde rsk FX Composte,7,6,5,,,, Sources : Reuters, authors' calculatons, , Chart a. Sem-volatlty FX 5 Chart b. Sem-volatlty FX 5 Argentna Brazl M exco Chna South Korea Phlppnes Thaland Sources : Datastream Sources : Datastream Chart c. Sem-volatlty FX 5 South Afrca Poland Turkey Sources : Datastream The comparson of [EMBIG] and [ELMI+(LC) + FX] downsde volatlty shows that the rsk related to holdng emergng securtes denomnated n the local currency (composed of the downsde volatlty of returns and exchange rates) s hgher than the one related to holdng soveregn securtes ssued n foregn currences, except n Chna, because of currency rsk. However, n several countres, the rsk dfferental n favour of EMBIG securtes remans low (Brazl, Mexco, Poland and Thaland). A noteworthy pont s that the downsde volatlty of the weghted composte ndcator [ELMI+(LC) + FX] s slghtly hgher than the downsde volatlty of the EMBIG weghted composte ndcator over the perod (cf. the frst and last columns, Table ). The analyss, however, remans ncomplete. As we have prevously emphassed, the assessment of a portfolo s rsk must take nto account correlatons between the varous downsde movements n returns,.e. n the possblty of reducng rsks va dversfcaton.

13 .. Correlatons between downsde movements n returns (EMBIG, ELMI+(LC)) and currences (FX). We have analysed correlaton coeffcents between downsde movements n returns on EMBIG securtes, n returns on ELMI+(LC) securtes and n the proftablty of exchange rates on daly data. We ntally assessed the matrces of correlaton coeffcents (assocated wth ther p-value n order to determne the sgnfcance 9 of the lnk) between downsde movements n proftablty rates of varous countres, for the two EMBIG and ELMI+(LC) ndcators. To do so, we use Spearman s correlaton coeffcent, whch s more approprate than the standard one (or Pearson s correlaton coeffcent) when the dstrbuton of sere s not normal. We are thus able to compare the average level of correlatons between downsde movements n EMBIG returns and n correlatons between downsde movements n ELMI+(LC) returns (Tables and ). We then determne the matrce of the correlaton coeffcents between downsde movements n the return of exchange rates of each emergng currency (Table ). Table. Corrélaton matrx* of downsde movements n EMBIG asset returns (over the perod) EMBIG South Afrca Argentna Brazl Chna South Korea Mexco Phlppnes Poland Thaland Turkey Venezuela Mean EMBIG South Afrca,5,69,85,6,5,8,6,6,56,9, Argentna,5,58,58,,,9,56,7,56,56,5 Brazl,69,58,9,8,67,8,98,,8,598,7 Chna,85,58,9,9,6,,9,68,8,6,8 South Korea,6,,8,9,,7,5,57,,86,65 Mexco,5,,67,6,,78,8,6,,58,89 Phlppnes,8,9,8,,7,78,96,5,5,9,6 Poland,6,56,98,9,5,8,96,75,6,,7 Thaland,6,7,,68,57,6,5,75,9,7, Turkey,56,56,8,8,,,5,6,9,,9 Venezuela,9,56,598,6,86,58,9,,7,, Mean EMBIG,,5,7,8,65,89,6,7,,9,,7 * : bold characters mean sgnfcant correlaton between countres at 99% confdence level Source: authors' calculatons Table. Corrélaton matrx* of downsde movements n ELMI+(LC) asset returns (over the perod) ELMI+(LC) South Afrca Argentna Brazl Chna South Korea Mexco Phlppnes Poland Thaland Turkey Venezuela Mean ELMI+(LC) South Afrca,9,,,9,6,6,,,,,7 Argentna,9,5,7,,5,56,9,56,5,,87 Brazl,,5,6,,6,76,9,6,8,6,7 Chna,,7,6,95,,5,7, -,,, South Korea,9,,,95,77,,6,66,7,5,7 Mexco,6,5,6,,77,,,5,,65,8 Phlppnes,6,56,76,5,,,5,5,6,,6 Poland,,9,9,7,6,,5,7,6,6,8 Thaland,,56,6,,66,5,5,7 -,7,5, Turkey,,5,8 -,,7,,6,6 -,7,7, Venezuela,,,6,,5,65,,6,5,7, Mean ELMI+(LC),7,87,7,,7,8,6,8,,,, * : bold characters mean sgnfcant correlaton between countres at 99% confdence level Source: authors' calculatons 9 We have set the sgnfcance threshold at %.

14 Table. Corrélaton matrx* of downsde movements n FX returns (over the perod) FX South Afrca Argentna Brazl Chna South Korea Mexco Phlppnes Poland Thaland Turkey Venezuela Mean FX South Afrca,6,,,8,,87,,,9,, Argentna,6,,6,,6 -,,6 -,,57,,6 Brazl,,,8,58,,6,65,5,,5, Chna,,6,8,5,,5,5,6, -,, South Korea,8,,58,5,89,,,79,5,9, Mexco,,6,,,89,75,,,6,7, Phlppnes,87 -,,6,5,,75,,9,56 -,,9 Poland,,6,65,5,,,,88,,, Thaland, -,,5,6,79,,9,88,79,,5 Turkey,9,57,,,5,6,56,,79,,8 Venezuela,,,5 -,,9,7 -,,,,,5 Mean FX,,6,,,,,9,,5,8,5,9 * : bold characters mean sgnfcant correlaton between countres at 99% confdence level Source: authors' calculatons The comparson of results between Tables and shows a sgnfcant dfference between the coeffcents of average ntra-class correlatons of the two types of return. The coeffcents are close to for. between downsde movements n returns on securtes denomnated n the local currency (average correlaton for the ELMI+(LC) ndcator of around.) and downsde movements n returns on securtes denomnated n foregn currences (average correlaton for the EMBIG ndcator of around.7). Another mportant dfference s that there s no upward trend n correlatons when the return decreases for ELMI ndces, whle unfavourable perods lead to contagon effects on EMBIG ndces. Accordng to J.P. Morgan (999), ths stuaton s explan by the fact that nvestors nvest n an asset class (debt securtes) when they nvest n the EMBIG, to a greater extent than they nvest n a local perspectve (nvestment n a country). They wll therefore be more senstve to nternatonal events and notably to movements n the US Treasury market. An n-depth study of the average correlatons of each country for the two ndcators throughout the perod shows, wthout any excepton, the low level of coeffcents assocated wth estc securtes n comparson wth nternatonal securtes. In partcular, the devaton between correlaton coeffcents s sgnfcant n the entre perod for Turkey (.9 versus.), Argentna (.5 versus.87), Venezuela (. versus.) or Brazl (.7 versus.7). Moreover, the downsde correlatons of EMBIG securtes are all sgnfcantly postve. An analyss of correlaton coeffcents for emergng currences (Table ) shows a low average correlaton between downsde movements n exchange rates throughout the perod (of around.9). In partcular, the downsde movements n currences n Argentna, Chna, the Phlppnes and Venezuela are close to zero over the perod. Conversely, currences wth the most hghly correlated exchange rates are those of South Afrca and Poland wth an average correlaton coeffcent of. over the perod. As a result, the average correlatons assocated wth emergng securtes denomnated n the local currency, for whch one also needs to take nto account correlatons between fluctuatons n exchange rates, are lower for the perod as a whole than average ntra-class correlatons of securtes denomnated n foregn currences (.7 versus.+.9 =.8). These results add credence to the argument callng for a greater dversfcaton of portfolos n favour of emergng securtes denomnated n the local currency nsofar as one part of rsk s mnmsed. However, before drawng a defntve concluson, we need to analyse the last component of the overall rsk of a portfolo unhedged aganst currency rsk: the cross correlatons between downsde movements n proftablty rates of emergng-country assets and downsde movements n correspondng currences (term 5 of equaton )... Correlatons between downsde movements n returns (ELMI+(LC)) and n correspondng currences (FX)

15 The last component of the rsk of an nternatonal portfolo s determned by the level of cross correlatons between downsde movements n returns on emergng securtes expressed n the local currency and downsde movements n the correspondng emergng currences. We have emprcally assessed the degree of correlaton between these two varables by calculatng the Spearman coeffcents for each country of the sample throughout the perod (July 997-December 7). The p-values calculated for the correlaton coeffcents enable us to test the null hypothess of a correlaton not sgnfcantly dfferent from zero. The results are presented n the followng Table : Table 5. Cross correlatons ELMI+(LC) - FX over the perod Cross correlatons ELMI+(LC) July December 7 ( countres) Rand - South Afrca -,5 Peso - Argentna,5 Real - Brazl -, Yuan - Chna,6 Won - South Korea -,5 Peso - Mexco -,76 Peso - Phlppnes -,67 Zloty - Poland -, Baht - Thaland -, Lre - Turkey,9 Bolvar - Venezuela -,9 Mean correlaton -, Cross correlatons are the last component of the rsk of a portfolo ncludng emergng soveregn securtes denomnated n local currences. Over the perod as a whole, only one country posts a sgnfcantly postve cross correlaton: Chna. Ths means that the asymmetrcal change n ths country s returns s qute sgnfcantly correlated to an asymmetrcal change n the local currency s exchange rate. Conversely, Mexco, the Phlppnes and Thaland post a sgnfcantly negatve cross correlaton that helps lower the portfolo s overall rsk. Over the perod as a whole, eght countres out of eleven post a negatve cross correlaton. The average cross correlaton s ultmately negatve and stands at -.. All n all, we can emprcally compare the varous levels of rsks nternatonal nvestors face by drawng on all the results of the fve components of portfolo rsk (Table 6). If the component of rsk defned by the downsde volatlty of returns and of exchange rates are approxmately smlar for the two types of securtes we have looked nto (downsde volatlty for the EMBIG composte ndcator s close to.5 whle the downsde volatlty for the ELMI+(LC) composte ndcator + the FX composte ndcator s approxmately.58), the comparson of the other components of overall rsk determned by the levels of correlatons enables us to draw a dstncton between these two types of assets. Over the perod as a whole, the correlaton assocated wth EMBIG securtes s close to.7 (Table ), whle for ELMI+(LC) securtes unhedged aganst currency rsk, the average correlaton stands at.75 (.8-.),.e. more than /rd lower than that of EMBIG securtes. Table 6. Summary of rsk portfolo components Downsde volatlty Intra-class correlatons Cross correlatons Global portfolo rsk EMBIG ELMI-FX EMBIG ELMI-FX EMBIG ELMI-FX EMBIG ELMI-FX Composte ndex,5,58,7,8 - -,,67,5 Source: authors' calculatons Coeffcents n bold type show a sgnfcant correlaton at the % threshold. Correlaton coeffcents are calculated accordng to Spearman s method, used when seres are not dstrbuted normally.

16 All n all, the overall rsk of a dversfed portfolo made up of emergng securtes denomnated n foregn currences s hgher for the sample of countres we have studed than the overall rsk of a portfolo made up of emergng-county soveregn bond securtes denomnated n the local currences unhedged aganst currency rsk over the perod July 997-December 7 (.67 versus.5). By consequence, t would have been n the nterest of an nvestor not to hedge aganst currency rsk over the perod under consderaton. 5. Concluson At ths pont, portfolo and nternatonal dversfcaton theory becomes fully meanngful: va a process of rsk rankng, we can show that the downsde potental for market rsk, acheved by an nternatonal dversfcaton of portfolos that ncludes emergng-country assets denomnated n local currences, s hgher than the downsde potental for the credt rsk shouldered by an nvestor who ncludes exclusvely n hs portfolo emergng securtes denomnated n foregn currences. Such a strategy nevertheless supposes buldng a dversfed enough portfolo, over a relatvely long perod. The advantages ganed from dversfcaton due to low correlatons between changes n proftablty rates of emergng securtes, but also wth other asset classes, should nduce nvestors to modfy structurally ther asset allocatons n favour of securtes denomnated n the local currency n order to mprove the effcency of ther portfolo.. Such a strategy could reduce the orgnal sn these economes face.

17 Bblography Bawa, V.S., 975. Optmal Rules for Orderng Uncertan Prospects, Journal of Fnancal Economcs, vol. (), pp Bawa V.S. et E.B. Lndenberg, 977. Captal Market Equlbrum n a Mean-Lower Partal Moment Framework, Journal of Fnancal Economcs, vol. 5[], pp Bekaert G. et C. R. Harvey, 997. Emergng Equty Market Volatlty, NBER Workng Papers 57, Natonal Bureau of Economc Research, Inc, revsed. Bekaert G., C.B. Erb, C.R. Harvey et T.E. Vskanta, 998. Dstrbutonal Characterstcs of Emergng Market Returns and Asset Allocaton, Journal of Portfolo Management, Wnter, pp Burger J.D. et F.E. Warnock, 7. Foregn Partcpaton n Local Currency Bond Markets, Revew of Fnancal Economcs, vol. 6[], pp. 9-. Campbell R.A. et R. Kräussl, 7. Revstng the Home Bas Puzzle - Downsde Equty Rsk, Journal of Internatonal Money and Fnance, vol. 6[7], November, pp R., Oosterwoud M., van der Made B. (6), Accessng local markets n emergng market debt, Focus on Emergng Asset Classes, PWM, jun, pp.-5. Echengreen B. et R. Hausmann, 999. Exchange Rate and Fnancal Fraglty, Proceedngs, Federal Reserve Bank of Kansas Cty, pp Echengreen B., R. Hausmann et U. Panzza,. The Pan of Orgnal Sn, n Echengreen et Hausmann [éd.], Other People s Money: Debt Denomnaton and Fnancal Instablty n Emergng Market Economes, Chcago, The unversty of Chcago Press. Echengreen B., R. Hausmann et U. Panzza, 7. Currency Msmatches, Debt Intolerance and Orgnal Sn: Why They Are Not the Same and Why t Matters, n Edwards [éd.], Captal Controls and Captal Flows n Emergng Economes: Polces, Practces and Consequences, Chcago, The unversty of Chcago Press. Estrada J., 7. Mean-Semvarance Behavour: Downsde Rsk and Captal Asset Prcng Internatonal Revew of Economcs & Fnance, vol. 6[], pp Harlow W. V. et R. K. S. Rao, 989. Asset Prcng n a Generalzed Mean-Lower Partal Moment Framework: Theory and Evdence, Journal of Fnancal and Quanttatve Analyss, vol., No., September, pp Harlow W.V., 99. Asset Allocaton n a Downsde Rsk Framework, Fnancal Analysts Journal, September-October, pp.8-. Hwang S. et C.S. Pedersen,. Asymmetrc Rsk Measures when Modellng Emergng Markets Equttes: Evdence For Regonal and Tmng Effects, Emergng Markets Revew, vol. 5[], pp Jarrow R. et F. Zhao, 6. Downsde Loss Averson and Portfolo Management, Management Scence, vol. 5, Issue, pp Medo M., Yeung C.H., Zhang Y-C, 9. How to quantfy the nfluence of correlatons on nvestment dversfcaton, Internatonal Revew of Fnancal Analyss, vol. 8, pp.-9. J.P. Morgan, 997. The Emergng Local Markets Index Plus, J.P.Morgan Securtes Inc, Emergng Market Research, November 9. J.P. Morgan, 999. Introducng the JP Morgan Emergng Markets Bond Index Global (EMBI Global), J.P.Morgan Securtes Inc, Emergng Market Research, August. Markowtz H., 95. Portfolo Selecton, Journal of Fnance, 7 [], pp Markowtz H., 959. Portfolo Selecton and Effcent Dversfcaton, J. Wlley & Sons. Nawrock, D., 999. A Bref Hstory of Downsde Rsk Measures, Journal of Investng, vol.8(), Fall, pp. 9-5.

18 9. Roy A.D., 95. Safety Frst and the Holdng of Assets, Econometrca, No., pp. -