Asymmetric exchange rate intervention and international reserve accumulation in India

Transcription

1 Asymmeric exchange rae inervenion and inernaional reserve accumulaion in India M Ramachandran Insiue for Social and Economic Change, Bangalore, India Naveen Srinivasan Indira Gandhi Insiue of Developmen Research, Mumbai, India Absrac The empirical evidence derived from he ARDL approach of Pesaran, Shin and Smih (1996) does no suppor he widely held view ha growing volailiy of exernal ransacions has significanly increased reserve demand. Insead, asymmeric exchange rae inervenion riggered, perhaps, by concerns abou expor compeiiveness seems o have conribued o large sockpile of reserves. Keywords: reserve demand; buffer sock model; asymmeric exchange rae inervenion. JEL Classificaions: E58; F31 Corresponding auhor: Professor and Head, Reserve Bank of India Uni, Insiue for Social and Economic Change; Nagarbhavi; Bangalore 56 7; India. Ph: ; Fax:

2 Asymmeric exchange rae inervenion and inernaional reserve accumulaion in India 1 Inroducion The recen surge in official inernaional reserves holding of emerging marke economies (EMEs) is largely aribued o increase in he volailiy of cross-border capial flows, subjec o sudden sops/reversal (Calvo, 1998; Edwards, 4; and Aizenman and Marion, 4). The available empirical evidence derived from panel daa suppor he view ha rising volailiy of exernal ransacions has significanly increased he precauionary demand for reserves (Flood and Marion, ; Aizenman and Lee, 5). However, his claim may no be valid in he presence of adminisraive conrols over capial flows. For insance, in India capial flows are highly resriced and ouflows are no as free as inflows (Nayyar, ; Miniane, 4). Such asymmeric conrol is an inegral par of exchange rae managemen, as i would minimize he probabiliy of sudden reversal of capial flows. Hence, i is hard o believe ha he surge in reserve holding reflecs precauionary demand. A recen empirical sudy by Ramachandran (4) shows ha volailiy has no played a major role in explaining growing demand for reserves in India. If so, wha explains he unprecedened accumulaion of reserves? I is ofen poined ou ha exchange rae inervenion policy is riggered by concerns abou expor compeiiveness (Dooley, Folkers-Landau and Garber, 3). If his is he case, he auhoriy s response o appreciaing pressure on domesic currency is likely o be more forceful han o depreciaing pressure of he same magniude. Such a policy response which aims a srenghening expor compeiiveness leads o accumulaion of reserves over ime. We empirically examine wheher evidence for such asymmeries exis which differeniaes our sudy from oher exising sudies. Secion presens he exended version of buffer sock reserve demand equaion; secion 3 deals wih empirical resuls; and secion 4 concludes. 1

3 The model We use he buffer sock model of Frenkel and Jovanovic (1981) o explain he growing demand for reserves. This model is found o be very popular in he empirical lieraure on reserve demand, as heir elasiciy esimaes were remarkably close o heir heoreical predicion. 1 The benchmark reserve demand equaion is: logr = β + β1 logσ + β logr + u β1 f, β p (1) where R is reserves, σ is volailiy of reserve incremen, r is opporuniy cos of holding reserves and u is whie noise error. The model assumes ha reserve movemens follow a random walk process (Wiener process in coninuous ime period). We incorporae asymmeric exchange rae inervenion by exending equaion (1) as follows: log( R ) = β + β log( σ ) + β log( r ) + αe log( R ) = β + β log( σ ) + β log( r ) + λ e 1 1 a 1 + u s + λ e d + u a () where e = ( log E ) x 1 [E is domesic price of one uni of foreign currency]; hence, e is percenage change in exchange rae. If α < hen auhoriies lean agains he wind and exchange rae variaion have a symmeric impac on reserve demand. Furhermore, e a and e d are measures of appreciaing and depreciaing pressure on domesic currency. Tha is, e a = d1e [d1 = 1 if e < and zero oherwise] and e d = de [d =1 if e > and zero oherwise]. The coefficiens λ 1 and λ measure he response of reserve demand o appreciaing and depreciaing pressure respecively. For example, if λ 1 < auhoriies 1 The heory predics ha β 1 =.5 and β = -.5. Ideally, we should use he deviaion of exchange rae from arge or from equilibrium real exchange rae. Insead, nominal exchange rae is chosen since he Reserve Bank of India does no follow any explici argeing framework. In fac, monioring he nominal exchange rae, as opposed o he real exchange rae, has been he official policy. For example, he former Governor of he RBI Jalan (1999) saes: From a compeiive poin of view and also in he medium erm perspecive, i is he REER, which should be moniored as i reflecs changes in he exernal value of a currency in relaion o is rading parners in real erms. However, i is no good for monioring shor-erm and day-o-day movemens as nominal raes are he ones which are mos sensiive of capial flows. Thus, in he shor run, here is no opion bu o monior he nominal rae.

4 buy foreign exchange in response o appreciaing pressure. Such a policy response in an era of coninuous ne capial inflows acceleraes he accumulaion of official reserves. 3 The empirical resuls The esimaes of reserve demand equaions are obained using weekly daa for he period from 5 January 1 o 1 Augus 5 3. We choose his sample as more han hree fourh of curren level of reserves (US$ billion as on 13 h January 6) has been accumulaed during his period. For esimaion purpose, reserve is measured as foreign currency asses while he implici yield on 91-day Treasury bill a cu-off price is used as a proxy for opporuniy cos. 4,5 The exchange rae is defined as rupee per US$. The daa are colleced from various issues of he Reserve Bank of India Bullein and The Handbook of Saisics on Indian Economy. The consrucion of volailiy measure (σ) is very crucial in esimaing reserve demand equaion. Flood and Marion () have demonsraed ha defining volailiy as rolling sandard deviaion of reserve incremen provides upwardly biased coefficien esimaes due o posiive skewness in he daa. However, Ramachandran (4) has shown ha he use of condiional sandard error of reserve incremen eliminaes such bias. Accordingly, we examine he presence of ARCH effec in reserve incremen using LM es. The es saisics consisenly rejec (no repored) he null hypohesis of no ARCH effec a 1% significance level for differen lag specificaions. This jusifies using condiional sandard errors of reserve incremen from an appropriae ARCH model (Engle, 198). Based on he Ljung Box es saisic, we consruced condiional volailiy using ARCH (1) process. The OLS esimaes of he reserve demand equaion (1) is: logr = log σ (.) (.) (.) 1.651ogr R =.71 F = 89.87(.) D W =.7 (3) 3 This is he minimum frequency a which daa on reserve is available for India. 4 The gold sock, SDRs and Reserve Tranche Posiion held wih he IMF are no included, as hey consiue a very negligible proporion of reserves and are no used as an inervenion asse. 5 Indeed, he difference beween he domesic ineres rae and shor erm ineres raes on he Unied Saes, European and Japanese governmen securiies could have been a beer proxy for opporuniy cos, because he bulk of he RBI s foreign currency asses are held in hese insrumens. Neverheless, we use only he domesic ineres rae; since he shor-erm raes in hese counries are low and he difference beween he domesic and foreign raes are largely influenced by domesic ineres raes. 3

5 where σ is condiional sandard deviaion of reserve incremen. The p-values in parenheses indicae ha he model and he esimaed coefficiens are significan a 1% level. Alhough he coefficiens have he expeced sign, he coefficien on volailiy is much lower han he heoreical predicion while he opporuniy cos elasiciy is larger. 6 However, he D-W saisic (.7) is lower han R value (.71) implying ha he esimaes migh be spurious. 7 Moreover, he sandard unis roo ess (no repored) confirms ha R follows I (1) process, consisen wih he assumpion of he buffer sock model. However, he auoregressive condiional sandard error- a measure of volailiy (σ) and change in exchange rae are I () whereas he opporuniy cos is I (1) process. Hence, he maximum likelihood approach of Johansen and Juselius (199) o es for coinegraion may no be appropriae, as i requires all he variables o follow he same order of inegraion. Neverheless, we can use he bounds es procedure proposed by Pesaran, Shin and Smih (1996) and Pesaran, and Shin (1998) as i does no involve pre-esing inegraion properies of he daa. The es yields asympoically efficien long run esimaes irrespecive of wheher he underlying regressors are I () or I (1) process. For insance esing for coinegraion among R, σ, and r involves he following seps. Firs, we need o esimae an unresriced error correcion model of reserves: log R = ax m + b log R i i = 1 + γ log R 1 1 i + γ log σ n + c log σ i i = 1 + γ log r 3 i 1 p + d log r 6 The mos difficul ask in esimaing he reserve demand equaion is obaining an appropriae measure of opporuniy cos of reserve holding. See Ben-Bassa and Golieb (199) for a debae on opporuniy cos measures. 7 The empirical sudies on reserve demand have widely used he buffer sock model of Frenkel and Jovanovic (1981) and esimae he parameers of he model using OLS mehod. However, he buffer sock model assumes ha reserves follow a random walk process in discree ime. If so, esimaion of reserve demand equaion using OLS mehod is meaningful if and only if one or all regressors of reserve demand specificaion follow random walk process and are coinegraed wih reserves. Hence, i is essenial o examine he inegraion and coinegraion properies of variables in reserve demand funcion. i = + ε i i (4) 4

6 where X is a vecor of deerminisic variables; b i, c i, d i are shor run dynamic coefficiens; γ s are long run muliplier; and ε is whie noise error. Rejecing he null hypohesis γ 1 = γ = γ 3 = indicaes ha here exiss long-run relaionship among R, σ, and r irrespecive of variables inegraion properies. However, we have o use he criical bounds available in Pesaran, Shin and Smih (1996) for esing he null, as he asympoic disribuion of Wald or F saisics is nonsandard. If variables have long-run relaionship, we can esimae he long run coefficiens and he corresponding error correcion model. This involves esimaing an auoregressive disribued lag model: log R q1 q q3 = a + a1 + i logr i + θi logσ i + ψ i logr i i= 1 i= i= δ + ν (5) The OLS esimaes of equaion (5) can be used o obain he long run coefficiens of reserve demand equaion. We esimae equaion (4) for specificaions: R σ, r; R σ, r, e; and R σ, r, e a, e d wih a linear rend and a consan as deerminisic variables. 8 The F saisics for esing he null hypohesis of no long-run relaionships are produced in Table 1. The es saisics for differen lag srucure across alernaive specificaions are consisenly above he criical values. This confirms ha here exiss long run relaionship among all he variables. Accordingly, we esimae equaion (5) o obain he long run coefficiens while he sandard errors are obained using he dela mehod. The resuls in Table indicae ha volailiy has posiive and saisically significan impac on reserve demand. However, he coefficien on volailiy is much lower han he heoreical predicion. The coefficien on opporuniy cos variable is negaive and saisically significan. Moreover, is magniude is closer o he heoreical predicion unlike OLS esimae of In he case of symmeric model, he negaive coefficien on percenage change in exchange rae 8 We consider shorer lags in he shor-run dynamic specificaion of error correcion model considering he high frequency of reserve adjusmen and found no significan difference in he qualiy of resuls for longer lags; hence, we presen resuls for symmeric lags of, 4, and 8. 5

7 implies ha he RBI is leaning agains he wind irrespecive of wheher rupee is under appreciaing or depreciaing pressure. Table 1: F saisics for esing coinegraion Symmeric lags R σ, r R σ, r, e R σ, r, e a, e d The criical bounds for 5 % significance level in he case of hree, four and five variable models wih consan and a linear rend are ; ; and respecively (Pesaran, Shin and Smih, 1996). If F > F U, one can rejec γ 1 = γ = γ 3 = ; hence, here is a long-erm relaionship among variables. If F < F L, one canno rejec γ 1 = γ = γ 3 = ; hence, here is no long-run relaionship. Finally, if F L < F < F U he inference is inconclusive. Table : Esimaes of long run coefficiens Variables Benchmark model Coefficiens of Symmeric model Asymmeric model σ.11 (.3).18 (.3).113 (.3) r (.) (.) (.) e (.) e a -.6 (.) e d -.99 (.14) c 1.98 (.) (.) (.) T.5 (.).4 (.).4 (.) ecm (.1) -.35 (.).38 (.) Figures in parenheses are p values. The SBC and AIC crieria suggesed ARDL order of (,, ) for benchmark buffer sock model; (,,, ) for symmeric inervenion model; and (,,,, ) for asymmeric inervenion model. The sriking feaure of he resuls in Table is ha reserve responds asymmerically o exchange rae variaions. There is a rise in reserve demand in response o appreciaing rupee whereas he reserves do no fall significanly in response o depreciaing rupee. This ype of asymmeric response o exchange rae changes over ime seems o have conribued o he huge accumulaion of official reserves in India. The las row of he Table presens he speed of adjusmen parameer. The coefficiens are significan wih expeced sign, bu he convergence rae is moderae. 6

8 Table 3: Predicive accuracy of reserve demand equaions Compeing models (mean square errors) T values (p values) Benchmark (.847) vs. Symmeric (.156) 4.74 (.) Benchmark vs. Asymmeric (.17) (.) Symmeric vs. Asymmeric (.) Since we find coinegraion among all he specificaions of reserve demand, we examine he superioriy of one specificaion over he oher using he predicive accuracy es suggesed by Diebold and Mariano (1995). The prediced values of reserves are consruced using he long run coefficiens of respecive models and he es resuls are produced in Table 3. The mean square error of benchmark model is he highes while ha of asymmeric model is he lowes. The las column provides he values o es he null hypohesis ha wo compeing models have equal forecas accuracy. The evidence indicaes ha he null is rejeced in hree possible compeing reserve demand specificaions. The lowes mean square error (and he values) observed indicae ha he forecas accuracy of asymmeric specificaion is superior o benchmark and symmeric specificaion. In sum, he resuls sugges ha asymmeric exchange rae inervenion riggered by concerns abou India s expor compeiiveness played a major role in explaining reserve accumulaion. 4. Conclusion The evidence derived from he ARDL approach of Pesaran, Shin and Smih (1996) does no suppor he view ha growing volailiy of inernaional ransacions has significanly increased reserve demand in India. Insead, asymmeric exchange rae inervenion i.e. aggressive purchase in response o appreciaing rupee and insignifican response o depreciaing rupee seems o have conribued o large sockpile of reserves. Acknowledgemen We hank N. S. S. Narayana and Vidya Mahambare for commens. 7

9 References Aizenman, J., Lee, J., 5. Inernaional reserves: precauionary versus mercanilis views, heory and evidence, NBER Working Paper, No Aizenman, J., and Marion, N. P., 4. Inernaional reserves holdings wih sovereign risk and cosly ax collecion, Economic Journal, 114, Ben-Bassa, A., Golieb, D., 199. On he effec of opporuniy cos on inernaional reserve holdings, The Review of Economics and Saisics 74, Calvo, G., Capial flows and capial-marke crises: he simple economics of sudden sops, Journal of Applied Economics, 1, Diebold, F., Mariano, R., Comparing predicive accuracy, Journal of Business and Economic Saisics, 13, Dooley, M., Folkers-Landau, D., Garber, P., 3. An essay on he revived Breon Woods Sysem, NBER Working papers, No Edwards, S., 4. Thiry years of curren accoun imbalances, curren accoun reversals, and sudden sops, IMF Saff Papers, 51, Special Issue. Engle, R. F., 198. Auoregressive condiional heroskedasiciy wih he esimaes of he variance of Unied Kingdom inflaion, Economerica 5, Flood, R., Marion, N.,. Holding inernaional reserves in an era of high capial mobiliy, IMF Working Paper, WP//6. Frenkel, J. A., Jovanovic, B., Opimal inernaional reserves: a sochasic framework, Economic Journal 91, Jalan, B. (1999), Inernaional financial archiecure: Developing counries perspecive, RBI Bullein. Johansen, S. & Juselius, K. (199). Maximum likelihood esimaion and inference on coinegraion wih applicaion o he demand for money. Oxford Bullein of Economics and Saisics,

10 Miniane, J., 4. A new se of measures on capial accoun resricions, IMF Saff papers 51, Nayyar, D.,. Capial Conrols and he World Financial Auhoriy: Wha Can We Learn from he Indian Experience?, CEPA Working Paper, No.14. Pesaran, M.H., Shin, Y., Smih, R.J., Tesing for he exisence of a long-run relaionship, DAE Working Paper No. 96, Deparmen of Applied Economics, Universiy of Cambridge. Pesaran, M.H., Shin, Y., An auoregressive disribued lag modelling approach o Coinegraion Analysis, in: Srom, S., Diamond, P. (eds.), Cenennial Volume of Ragnar Frisch, Cambridge Universiy Press. Ramachandran, M., 4. The opimal level of inernaional reserves: evidence for India, Economics Leers, 83,