School of Economcs Workng Paper 0505 Sterlsaton, Captal Moblty and Interest Rate Determnaton for East Asa Tony Cavol School of Economcs Unversty of Adelade, 5005 Australa ISSN 1444 8866
STERILISATION, CAPITAL MOBILITY AND INTEREST RATE DETERMINATION FOR EAST ASIA Tony Cavol * February 05 * School of Economcs, Unversty of Adelade. The author thanks Kshen Rajan for helpful comments. All remanng errors are my own. 1
Abstract Ths paper uses a smple open economy nterest rate determnaton model to emprcally examne an mportant aspect of precrss monetary and exchange rate polcy. It nvestgates whether sterlsaton of the reserve effects of captal nflows helped keep nterest rates suffcently hgh that they may have prolonged the nflow of captal. Despte the use of a precrss sample n ths study, the ssues n ths paper have relevance today gven the recent trend n foregn captal nflow for Asa. The emprcal secton s concerned wth the effect of reserve flows on the nterest rate and s dvded nto two parts. The frst tests for a contemporaneous effect of the basc model usng OLS and IV methods. The second generalses the model to assess for lagged effects by way of VAR analyss. The results show that there are some contemporaneous effects of sterlsaton on the domestc nterest rate though the effects are stronger when estmatng the lagged model. JEL Classfcaton: F30, F32, F41 Key words: Sterlsaton, Captal flows, Interest Rates 2
1. Introducton Ths paper uses a smple open economy monetary style model of nterest rate determnaton to emprcally examne the effect of sterlsed nterventon on the domestc nterest rate n the East Asan regon n the perod pror to the Asan crss. The model s based on Edwards and Khan (1985) (henceforth EK). Ths paper alters the condtons for money market equlbrum to nclude the effects of sterlsaton of captal nflows. Ths allows the model to assess whether sterlsaton was effectve n placng upward pressure on domestc nterest rates, thus keepng nterest rates at the level that exsted before the captal nflows. Before the crss, t s wdely acknowledged that several East Asan economes experenced substantal and prolonged perods of foregn captal nflow. 1 In the presence of largely managed exchange rates, monetary control of these economes was acheved by the sterlsaton of the reserve effects of captal nflows. In ths paper, the focus s on the mpact of the reserve effects of captal nflows on the domestc nterest rate. In the absence of sterlsaton of these nflows, under a fxed exchange rate regme, a captal nflow should place downward pressure on nterest rates. The presence of sterlsaton should, f successful, reverse ths effect. Full (or complete) sterlsaton of the reserve nflow would mantan the nterest rates at levels that exsted prenflow. Indeed, f foregn captal s attracted to hgh domestc returns, successful sterlsaton may actually propagate further captal nflow by keepng nterest rates from decreasng. 2 1 See, for example, Rajan and Sregar, 01 and Rajan et al, 01. 2 The smultanety that results from ths twoway relatonshp s addressed n the emprcal secton (secton 3) of ths paper. 3
Even though the paper studes the precrss tme perod, the ssues examned are relevant today. 3 Consder Table 1 and Table 2. Table 1 examnes country level captal flows for the precrss perod. To put ths perod nto context, Table 2 provdes some recent trends for Asa n general. It s evdent that captal nflows are ncreasng n the regon after the captal flght brought about by the crss. A key feature of the model s the relatonshp between captal moblty and sterlsaton. 4 The more moble s captal, the more the domestc nterest rate wll be nfluenced by external factors such as foregn nterest rates and current and expected future exchange rate changes. As captal moblty decreases, the nterest rate s drven more by domestc varables. It s ths scenaro nvolvng mperfect captal moblty where sterlsaton polces mght have an mpact on nterest rates. The paper s set out as follows; Secton 2 formalses the relatonshp between sterlsed nterventon and captal moblty and assesses the effectveness of sterlsaton n addressng the ssue of sustaned captal nflow. It s shown that the presence of hgh captal moblty and/or the presence of full sterlsaton of captal nflows neutralses the effect of the reserve nflow on the domestc nterest rate. In other words, hgh captal moblty renders sterlsaton neffectve but, where there s low to moderate levels of captal moblty, full sterlsaton reverses the downward pressure on nterest rates that would arse from a captal nflow epsode n a fxed exchange rate regme. Usng the model as an organsng framework, Secton 3 wll present some emprcal estmates to nvestgate the extent and effectveness of sterlsaton for the East Asan regon usng monthly observatons from 199097. The selecton of 1990 as 3 The precrss scenaro s used ALSO because there s a suffcent amount of data to assess, ex post, the exstence of the extent and magntude of the captal nflow epsode and the extent of the sterlsaton of the reserve nflow. 4 The nteracton between sterlsaton and captal moblty s captured by the rrelevance hypothess (see Frankel and Okwongu, 1996 and Kumhof 00). 4
the startng pont s drven manly by the fact that many of the countres n ths sample had made substantal efforts to deregulate ther fnancal systems durng the 19s thereby elmnatng the necessty of modellng a structural break. The paper observes sterlsaton and captal moblty separately before evaluatng the effect of reserve nflow on the domestc nterest rate. The results ndcate that, due manly to the presence of hgh captal moblty or a hgh degree of sterlsaton, the contemporaneous effect of the reserves nflow on nterest rates s small n magntude. Some emprcal results based on VAR analyss are also derved to examne whether there are any lagged effects of sterlsaton on nterest rates. Secton 4 provdes some concludng comments. 2 The Interest Rate Model and Some Implcatons a) Model Ths secton derves a model based on Edwards and Khan (1985) (EK) where the effects of sterlsaton are ncorporated. Consder the followng: 5 t = ψ * t + (1ψ)τ t, 0<ψ<1 (1) Equaton (1) s the structural nterest rate equaton from EK. It states that the domestc nterest rate s a weghted average of nternatonal monetary condtons, * t and domestc monetary condtons, τ t. The parameter ψ refers to a country s level of captal moblty. As captal moblty ncreases, the domestc nterest rate s determned 5 Varables n lower case denote logs (except, ρ and π) 5
ncreasngly by external factors and as captal moblty decreases, the domestc nterest rate s determned more by domestc monetary condtons. The external factors, * t, are measured by uncovered nterest party (UIP). Ths s expressed as follows: t * = t f + (e e t+1 e t ) (2) where f t s a foregncurrency nterest rate wth whch to base UIP and e t s the log of the current exchange rate expressed as the domestc prce of foregn currency. e e t+1 s the expected deprecaton of the (log) exchange rate n the next tme perod. As n many studes of captal moblty based on UIP, the rsk premum s not explctly captured n the model here, but s very much mpled. Ths s partcularly the case for developng economes where sgnfcant uncovered nterest dfferentals (UIDs) are found to exst (de Brouwer, 1999 and Aras, 01). In the context of the model, τ t s the domestc nomnal nterest rate that would exst f t were manpulated entrely by domestc monetary condtons. 6 Hence, τ t s a (shadow) nterest rate that captures condtons of dsequlbrum arsng from excess demand or supply of money. As n EK, ths shadow rate can be calculated n the followng way: τ t = ρ + π e t+1 + γ(m d t m t ) (3) 6 The nterpretaton of varable lke τ n Edwards and Khan, 1985, Haque and Montel, 1991, Resen and Yeches, 1993 and Dooley and Matheson, 1994 and such others s that t s a closed economy nterest rate or shadow rate. Ths nterpretaton has to be refned n ths context. Under ths model, τ depends on (amongst others) the sterlsaton coeffcent, λ. Ths mples that the captal account must not be completely closed for τ to be determned. An alternatve defnton of τ would be that s the domestc rate that would be determned entrely by domestc consderatons where these consderatons nclude the monetary polcy stance taken by the central bank. 6
Equaton (3) derves the domestcally determned nterest rate, τ, as comprsng of ρ, a full equlbrum real nterest rate reflectng the long run margnal product of captal, the expected future nflaton rate, π e t+1, and a term capturng monetary dsequlbrum. Thus, any excess (shortfall) of log money demand (m d ) relatve to ts supply (m) wll result n an ncrease (decrease) n the domestcally determned nterest rate. The demand for money s smlar to the one gven n EK wth the addton of a stock adjustment term. It depends on the full equlbrum nterest rate, expected future nflaton and ncome (y). m d t p t = α 1 (ρ + π e t+1) + α 2 y t + α 3 m t1 (4) The effect of reserve nflow sterlsaton enters through the expressons for the money stock. The ncorporaton of sterlsaton yelds the followng: 7 M t = (1+λ) F t, λ 0 (5) where λ s the sterlsaton coeffcent. Equaton (5), when expressed n log dfferences, becomes: 8 m t = (1+λ)κ f t (6) 7 Sterlsaton s captured by consderng the relatonshp between domestc credt (D) and foregn reserves (F) as follow: D = λ F. Substtutng ths nto the famlar expresson for the money base (n frst dfferences), M = D + F, one obtans equaton (5). 8 From equaton (7), dvde both sdes by M t1 and then multply and dvde both sdes by F t1. Usng the result, (X t X t1 )/X t1 ln(x t ) ln(x t1 ),we obtan equaton (8). 7
where: κ = F t1 /M t1. The money stock can also be expressed as follows: m t = m t + m t1 (7) Substtuton of (6) nto (7) yelds: m t = (1+λ)κ f t + m t1 (8) The domestc nterest rate, t can now be calculated by substtutng (8) and (4) nto (3) to fnd τ t and ths can be substtuted nto (1): 9 t = θ 0 + θ 1 * t θ 2 f t θ 3 m t1 + θ 4 π e t+1 + θ 5 p t + θ 6 y t (9) where θ 0 = (1ψ)(1γα 1 )ρ θ 1 = ψ θ 2 = [(1ψ)γ(1+λ)κ] θ 3 = (1ψ)(γ γα 3 ) θ 4 = (1ψ)(1γα 1 ) θ 5 = (1ψ)γ θ 6 = (1ψ)γα 2 9 Cavol and Rajan (05) derve a smlar model to analyse the effect of sterlsaton on the uncovered nterest dfferental (UID). 8
b) Implcatons What does the model mply about the behavour of sterlsaton? The parameter of most nterest s that for the reserve nflow, f, θ 2 = [(1ψ)γ(1+λ)κ]. The degree to whch the sterlsaton of reserve nflows s successful n mantanng an upward pressure on the domestc nterest rate s drven by three factors. These are consdered brefly n turn. The frst factor nfluencng the effect of f on t s sterlsaton. (1+λ) = 0 mples that λ = 1 and that there s complete sterlsaton of reserve flows as a change n reserves s fully offset by a(n opposte) change n domestc credt. If ths s the case, the model asserts that there s no effect on the nterest rate as full sterlsaton negates any monetary mpact of a reserve flow and, as a result, nterest rates also reman constant through the lqudty effect. The downward pressure on the nterest rate due to a possble captal nflow has been reversed by sterlsaton and thus remans at the same level as before the nflow epsode. The second factor affectng the relatonshp between f t and t s the adjustment parameter, γ. It s the extent to whch a monetary dsequlbrum affects the domestc nterest rate. Ths effectvely represents a measure of the senstvty of the effects of sterlsaton on the nterest rate. Clearly, a hgher γ ndcates that the domestc nterest rate s more senstve to the authortes sterlsaton actvtes. Its nteracton wth the other parameters s mportant. The thrd factor s the level of captal moblty. For a gven and constant (1+λ), f the level of captal moblty, ψ ncreases (or (1ψ) decreases), the effect of a reserve nflow on t dmnshes. For levels of captal moblty that are extremely hgh (nearng 1), the effect of captal nflows on the nterest rate tends to zero rrespectve of the 9
extent of sterlsaton actvty. Ths s because the nterest rate s determned prmarly by foregn nterest rates and/or other external factors. 10 3. Estmatng the Interest Rate Model 3.1. Data and Estmaton The data s taken from the IMF IFS CD database takng monthly observatons from 1990:1 to 1997:5 as the sample perod. The sample excludes any effects of the crss. Exchange rates are taken from lne RF, nterest rates, prces and output are taken from lnes B, 64 and 66 respectvely. f s measured as [ FA/FA(1)]* to measure the senstvty to a percentage change n reserves and also because there were some observatons of negatve changes n reserves. Expected nflaton s measured as [log(cpi(12))log(cpi)]*, m t1 s log(mb(1)), and output and CPI are measured n logs. The equaton s based on Equaton (9) and s estmated usng OLS and TSLS. It ncludes a lagged dependant varable term to soak up the substantal amounts of seral correlaton n the data and are estmated usng NeweyWest heteroskedastcrobust standard errors. OLS and TwoStage Least Squares (TSLS) results for ths are presented n Table 4 and are dscussed n the next secton. TSLS s the preferred method for ths exercse to attempt to remove the possble bas resultng from the endogenety of f arsng from ts relatonshp wth domestc assets and the nomnal nterest rate through the monetary offset coeffcent (see Kour and Porter, 1974, 10 In the context of ths (monetary style) model that has no portfolo balance type features, the captal moblty parameter may also capture characterstcs more typcally assocated wth mperfect asset substtuton. These mght nclude sluggsh adjustment of returns followng a shock (or a sterlsed nterventon) and rsk factors. Hence, the captal moblty parameter may have a more general nterpretaton to nclude any factors that, n some way, hnder the convergence of the nterest dfferental In a recent paper, Hutchson (02) beleves that captal moblty (more strctly nterpreted) s a stronger factor n determnng the effectveness of sterlsaton than mperfect asset substtuton. For mperfect asset substtuton to have any meanngful effect, the relatve asset quanttes beng moved must be enormous. 10
Obstfeld, 1982, Frankel and Okwongu, 1995, Moreno, 1996, Bond, 1999, Emr et al, 00 and Mark, 01 for a descrpton). The nstruments used n the TSLS regressons are the regressors from the OLS model, d whch s measured as (DADA(1))/DA( 1) and the trade balance as per Bond (1999) and Mark (01). 11 The subsequent analyss s based on the TSLS results the OLS estmates are provded merely for completon. All estmates were subject to the BreuschGodfrey LM test for hgher order seral correlaton (Godfrey, 1988), Whte s test for heteroskedastcty and the Engle ARCH LM test for the exstence of ARCH processes (Engle, 1982) and the Ramsey RESET test for general specfcaton. For the most part, there were no sgnfcant ARCH coeffcents except for Malaysa, but the model estmated remans Least Squares due to the fact that estmaton n an ARCH model does not materally alter the value of the parameters for * and f. The nterest rate s estmated n levels as they are I(0) processes (except for Malaysa and weakly I(0) for Korea) over the sample. ADF tests on the resduals for Malaysa and Korea ndcate that the resduals are I(0) mplyng that a contegratng relatonshp exsted n the models as estmated (see Appendx 1). 3.2. Results a) Contemporaneous Relatonshp In the context of the model n Secton 2, the overall effect on the domestc nterest rate centres around the nteracton between the degree of captal moblty and the extent of reserve sterlsaton. Recall from Secton 2 that f captal moblty s very hgh (around 1) then nternal factors such as sterlsaton do not have an mpact on the 11 The trade balance s used to proxy for the current account for whch monthly data s dffcult to obtan for the sample used here. 11
domestc nterest rate. Sterlsaton can only affect nterest rates under the model f captal moblty s less than perfect. Whle the extent of sterlsaton s not estmated here, there have been several recent studes that present broadly smlar results. 12 Table 3 presents some sterlsaton coeffcents from Cavol and Rajan (05). The results ndcate that the sterlsaton s hgh and very close to complete for most countres studed. The estmates for Korea suggest possble oversterlsaton as the coeffcent exceeds 1 and those for Indonesa are lower than the others at 0.76 an ndcaton of partal sterlsaton. The level of captal moblty s estmated hare and s gven by the parameter θ 1. The results gven n Table 4 are qute mxed. For the Phlppnes and Malaysa, the coeffcent value s not sgnfcant. Korea s captal moblty coeffcent s 0.39, suggestng that, when takng nto account any rsk factors, around 1/3 rd of a foregn nterest rate move s transmtted nto domestc rates. Thaland records hgher coeffcent values at 0.83 and Indonesa s s 0.56. In contrast to much of the lterature on fnancal ntegraton of late (see Cavol et al, 03 for a survey), the above results, wth the excepton of Thaland, suggest qute low levels of captal moblty. Ths mght be due to the exstence of a rsk premum, captal controls and endogenous monetary polcy (see McCallum, 1994 and Anker, 1999). Under a scenaro where captal moblty s mperfect and sterlsaton s close to complete, the model predcts that the domestc nterest rate ought to be explaned by domestc factors. Take, for nstance, Thaland. The captal moblty parameter for Korea s 0.83. Ths means the remander (10.83) of the domestc rate that can be tentatvely explaned by domestc factors such as sterlsaton. From Table 3, we also know that the degree of sterlsaton for Thaland s 0.91. Ths would suggest that the 12 See Kwack (01), Fane (02), Cavol and Rajan (05). 12
sterlsed nterventon beng undertaken should negate much, but not all, of the downward pressure placed on the nterest rate by the captal flows. As a result, the value of θ 2 should be close to zero and slghtly negatve. The exstence of possble oversterlsaton for Korea would suggest that θ 2 s slghtly postve. Consder the value of θ 2 n Table 4. The sgn and sze of the effect of (%) changes n f on the domestc nterest rate s consstent wth the dscusson above n that they are close to zero. For Korea, where the sterlsaton parameter s less than 1, the effect of f on t s expected to be slghtly postve and s so but t s statstcally nsgnfcant. For the other countres n the sample where sterlsaton s less than complete, the sgn of θ 2 should be slghtly negatve. Ths s ndeed the case n Thaland. Only Indonesa s sterlsaton coeffcent s statstcally sgnfcant to any reasonable degree (but the ncorrect sgn). The lack of decsve results n ths secton motvate the analyss n the next secton, whch examnes whether there s any lagged nfluence on the domestc nterest rates. b) Usng a VAR to test for a lagged relatonshp The model n Secton 2 and subsequent analyss examnes the contemporaneous relatonshps between sterlsaton, captal moblty and the domestc nterest rate. In ths secton, we nvestgate whether there s a lagged relatonshp by presentng a Vector Autoregresson (VAR) model on the varables of nterest,, f and *. The VAR s of a standard form and s gven by the followng: 13 13 The exogenous varables n equaton (10) are left out of the VAR manly to help wth model dentfcaton. In addton, wth the excepton of Korea, they are largely nsgnfcant. In order to dentfy the model, the Cholesk decomposton s mposed, a 22 =a 31 =a 32 =0. The restrcton of a 22 =0 corresponds to the nstrument equaton n the TSLS regresson and, as such, addresses the endogenety ssues as n the statc model. 13
t = a m m m * * 10 + a11t + a12 f t + γ 1 jt j + β1 j f t j + α 1 jt j + δ 1 X t + µ t j= 1 j= 1 j= 1 (10a) f t = a m m m * * + a21t + a22t + γ 2 jt j + β 2 j f t j + α 2 jt j + δ 2 X t + µ t j= 1 j= 1 j= 1 (10b) * t = a 30 + a 31 f t + a 32 t + m m m α 3 jt j + δ 3 X t + j= 1 j= 1 j= 1 * γ 3 jt j + β 3 j f t j + µ t (10c) where X t s a vector of exogenous varables n the OLS and TSLS equatons as outlned above. The lagged effects of foregn nterest rates, reserve changes and the domestc nterest rate on the current domestc nterest rate can be analysed by way of coeffcent restrctons tests. 14 Snce the prmary nterest s on what determnes the domestc nterest rate, the focus s on the coeffcents affectng t, Equaton (10a). Equaton (10b) and (10c) are essental n the analyss of a VAR because they determne the current and future values of f and * whch, n turn, mpact on subsequent values of. Table 5 reports the χ 2 statstcs of the Wald Test for coeffcent restrctons for the VAR model. Three lag structures of the model are examned; m = 12, 6, 3 months. For each structure, the Wald Test s performed for each set of m parameters for γ 1, β 1 and α 1. The am of the test s to ascertan whether the lagged effects of *, f and the domestc nterest rate, are sgnfcant n explanng the current value of. Let s look frstly at the results regardng the sgnfcance of restrctng the coeffcents of the lagged * terms (the γ 1 terms). Generally, the omsson of the γ 1 terms has lttle effect regardless of the lag structure of the model. The excepton s Korea, where the results are statstcally sgnfcant for m=6 and m=3. Hence, for 14
Korea, there s strong evdence suggestng an external nfluence on the domestc nterest rate. For Thaland and Indonesa, recall from Table 4 that there s a sgnfcant contemporaneous relatonshp between * and. From Table 5, t can be seen that the effect appears to be lmted to the same tme perod and not be carred over as a lagged effect, except for Indonesa for a threemonth lag. The weak lagged results for the Phlppnes and Malaysa seem to back up the weak contemporaneous results from Table 4. The second set of results relate to the effect of restrctng the β coeffcents. In other words, the test examnes whether a change n reserves, one that may arse from sterlsaton, nfluences the domestc nterest rate n the future. The results of the Wald Tests are qute mxed. For the most part, sgnfcant relatonshps are detected for Malaysa (3 and 6 months), Indonesa (6 months) and Thaland (3 months). The Phlppnes has a strong result for a 12month lag. Ths s not a strange result f one thnks about the tme t may take for a reserve flow to affect the money market rate. In fact, a delayed nterest rate response to a reserve change s consstent wth the weak contemporaneous results from Table 4. The strongest Wald Test results are by far those testng for the effect of the lagged domestc nterest rate. The excepton here s the Phlppnes. Ths s expected as there s always gong to be persstence n monetary varables. It s mportant to nclude the effect of the lagged dependent varable as t s an ndcaton of the effect of domestc vs external varables. The more sgnfcant the effect of lagged, the less lkely t s that external determnants and, thus, captal moblty, play a materal part n determnng the domestc nterest rate. 14 A Wald test for restrctons on γ and β are smlar tests to the Granger Causalty test for a selecton of lag lengths. 15
The VAR model can also be used to produce mpulse response functons. As wth the Wald tests, the focus s on the effect on the domestc nterest rate,. As such, Fgure 1 presents the mpulse responses of to shocks to the nnovatons n, f and * for a 24 month tme horzon. Model selecton crtera (AIC) are employed to determne the approprate lag length of the model for each country. 15 The orderng used s ( *, f, ). 16 The mportant relatonshp n the context of ths study s the effect of an nnovaton n f on. For Indonesa and the Phlppnes, the relatonshp appears to fluctuate over tme and, especally for the Phlppnes, the effect does not dsspate. In contrast, for Korea, Thaland and Malaysa, there s a negatve ntal effect on the domestc nterest rate. For Thaland and Malaysa, ths s consstent wth a scenaro of ncomplete (albet hgh) sterlsaton (see Table 3). The effect for Malaysa s especally strong as confrmed by the Wald test results. A reserve flow that s not fully sterlsed wll place downward pressure on the nterest rate. The result for Korea s anomalous. Table 3 suggests that Korea oversterlses. Ths would suggest that the effect of a shock to reserve changes should have a postve effect on the nterest rate. To further assess the relatve mportance of lagged, * and f n determnng, the standard VAR s used to analyse some varance decompostons. The lag length chosen for the decompostons and the orderng of the nnovatons s as above for the 15 Lag lengths are 2 months for Korea and Thaland, 12 months for Indonesa and the Phlppnes and 1 month for Malaysa. Appendx 2 presents the mpulse response functons for the model wth 12, 6 and 3 month lags respectvely. Whle not dscussed n the man text, they are provded for the reader s convenence and for comparson to the Wald results. Also, t s worth pontng out that AIC s chosen because t favours longer lag lengths than BIC and the objectve of ths exercse s to keep as much explanatory power n the model as possble. 16 The orderng s based on a Cholesk decomposton on the VAR model as specfed above. The decomposton s such that the contemporaneous effect of * and f s mantaned as t s n equaton (9). The value of f s determned contemporaneously by * but not, and the value of * not determned contemporaneously by ether f or. 16
mpulse responses. 17 Fgure 2 presents the varance decompostons for each country up to a 24 month forecast horzon. Each graph shows how a random nnovaton to, * and f n relatve terms affects the varaton of the domestc nterest rate n the model. The advantage of reportng the effect of an nnovaton n s that t addresses the relatve amount of persstence n the model. The domestc nterest s sgnfcantly drven by ts own nnovatons. In fact, the lne correspondng to never crosses the * or the f lne. In other words, the effect of the domestc nterest rate nnovatons on the domestc nterest rate s generally stronger then the effect of the foregn nterest rate and reserve changes for most countres tested. In Malaysa, from about the fvemonth pont, the effects of nnovatons n reserves and the domestc nterest rate seem to nfluence the domestc nterest rate n almost equal proportons. For the other countres, the relatve effect s not as hgh, wth the nnovatons to * and f each ultmately explanng a smaller proporton of the varaton n. Ths s an ndcaton, frst, that there s some dynamc relatonshp between reserves and the nterest dfferental but not an overly strong one and, second, t ndcates the strength of the persstence of the domestc nterest rate. 4. Concludng Remarks Motvated prncpally by the large and persstent captal nflows n East Asa before the crss, ths paper has presented a smple theoretcal model that examnes the lnk between the central banks desre to sterlse captal nflows and the domestc nterest rate n a fxed exchange rate regme. Under the model, the success of sterlsaton depends substantally on the level of captal moblty. If captal moblty s 17 Appendx 3 contans varance decompostons for the model wth 12, 6 and 3month lag respectvely. 17
perfectly hgh, domestc rates are determned entrely by foregn nfluences and, therefore, sterlsaton s neffectve (rrelevance hypothess). As such, the paper has nvestgated the possble connecton between a central bank s desre to sterlse and the domestc nterest rate. The nteracton of sterlsaton and captal moblty s also nvestgated emprcally. The paper estmates the sterlsaton and captal moblty coeffcents ndvdually from the structural equatons n the model. These are then used to assess the overall effect on the domestc nterest rate. The results presented show that, sterlsaton s mostly very hgh, wth possble oversterlsaton for Korea. The model s generalsed n order to show some lagged relatonshps between sterlsaton, captal moblty and the nterest rates. Wald coeffcent restrcton tests show that lagged relatonshps exst strongly for Malaysa and, to a lesser extent, Thaland and Indonesa but not so for other countres. The mpulse response functons show that a dynamc relatonshp does exst between reserve changes and the domestc nterest rate and they provde some support for the Wald tests. Varance decomposton tests ndcate that there s some relatonshp between reserve changes and nterest rates but that nterest rate shocks are hghly persstent. What do the results above mply about the effectveness of sterlsaton and about the determnants of the nterest rate nonconvergence? There s evdence of mperfect captal moblty n most countres. Ths suggests that sterlsaton polces desgned to mantan the nterest rate at prenflow levels should have some effect. For the contemporaneous (TSLS) results usng Equaton (9), the prncpal determnant of the domestc nterest rate s the foregn channel the foregn nterest rate, expected devaluaton of the currency and a possble rsk premum and these drve the level of 18
captal moblty. Gven that the perod examned was a tme of qute rgd exchange rates, t follows that the foregn (US) nterest rate can be solated as a man determnant. The VAR results (Equaton 10) ndcate that the nterest rate s sgnfcantly determned by ts own lag. That asde, the foregn channel s weaker n the VAR than the TSLS results and the effect of reserve flows are stronger n the VAR than n the least squares results. Ths s possbly suggestve that, for sterlsaton to be effectve n nfluencng the nterest rate, t may have to be admnstered n a smoothed fashon over a perod of tme. 19
References Anker, P., (1999), Uncovered nterest party, monetary polcy and tmevaryng rsk prema, Journal of Internatonal Money and Fnance, vol.18 (6) December: 83551. Aras, G., (01), Devatons from Uncovered Interest Party, Lessons to be Drawn from Currency Crss Models mmeo, CEFI, March Bond T.J., (1999), Captal Inflows to Asa: The Role of Monetary Polcy, n Gacs, Holzman and Wyzan (eds), The Mxed Blessng of Fnancal Inflows Transton Countres n Comparatve Perspectve, Edward Elgar Cavol, T., R. Rajan and R. Sregar, (03), A Survey of Fnancal Integraton n East Asa. How Far? How Much Further to Go?, mmeo, Aprl Cavol, T. and R. Rajan, (05), What Explans the Interest Premum Puzzle n Asa n the 1990s: Is Monetary Sterlsaton the Answer? mmeo, January de Brouwer. G., (1999), Fnancal Integraton n East Asa, Cambrdge: Cambrdge Unversty Press. Dooley, M. and D.J. Matheson, (1994), Exchange Rate Polcy, Internatonal Captal Moblty, and Monetary Polcy Instruments, n Glck, R. and M. Hutchson (eds), Exchange Rate Polcy and Interdependence: Perspectves from the Pacfc Basn. Cambrdge Unv Press Edwards, S. and M.Khan, (1985), Interest Rate Determnaton n Developng Countres. IMF Staff Papers 32, pp 377:3 Emr, O.Y., A. Karasoy and K. Kunter, (00), Monetary Polcy Reacton Functons n Turkey, mmeo, October Engle, Robert F., (1982), Autoregressve Condtonal Heteroskedastcty wth Estmates of the Varance of U.K. Inflaton, Econometrca, 50, pp 987 8. Fane, G., (00), Captal Moblty, Exchange Rates and Economc Crses. Edward Elgar Frankel, J. and C. Okongwu, (1995), Lberalzed Portfolo Captal Inflows n Emergng Markets: Sterlzaton Expectatons, and the Incompleteness of Interest Rate Convergence, Workng Paper 5156, NBER Glck R. and M. Hutchson, (00), Foregn Reserve and Money Dynamcs wth Asset Portfolo Adjustment: Internatonal Evdence, Journal of Internatonal Fnancal Markets, Insttutons and Money, 10, pp 22947 Godfrey, L.G., (1988), Specfcaton Tests n Econometrcs, Cambrdge Unversty Press.
Haque, N. and P. Montel, (1991), Captal Moblty n Developng Countres : Some Emprcal Tests. World Development 19:10, pp 7059 Hutchson, M., (02), The Role of Sterlzed Interventon n Exchange Rate Stablzaton Polcy, mmeo, June Kour. P. and M. Porter, (1974), Internatonal Captal Flows and Portfolo Equlbrum, Journal of Poltcal Economy 82:3, pp 44367 Kumhof. M., (00), Sterlzaton of ShortTerm Captal Inflows Through Lower Interest Rates?, Mmeo, Stanford Unversty Kwack, S., (1994), Sterlsaton of the Monetary Effects of Current Account Surpluses and ts Consequences: Korea 19861990, n Glck R. and M. Hutchson (eds), Exchange Rate Polcy and Interdependence: Perspectves from the Pacfc Basn, Cambrdge Unv Press (01), An Emprcal Assessment of Monetary Polcy Responses to Captal Inflows n East Asa Before the Crss, Internatonal Economc Journal, 15, pp 95113 McCallum, B.T., (1994), A Reconsderaton of the Uncovered Interest Party Relatonshp, Journal of Monetary Economcs, 33, pp 10532 McLeod R.H., (1998), Indonesa, n McLeod and Garnaut, (eds), East Asa n Crss: From Beng a Mracle to Needng one, Routledge McLeod, R.H. and R. Garnaut, (eds), (1998), East Asa n Crss: from beng a mracle to needng one. Routledge Mark, N.C., (01), Internatonal macroeconomcs and Fnance: Theory and Econometrc Methods, Blackwell Montel. P and C. Renhart, (1999), Do Captal Controls and Macroeconomc Polces Influence the Volume and Composton and Captal Flows? Evdence from the 1990s, mmeo Moreno, R., (1996), Interventon, Sterlzaton and Monetary Control n Korea and Tawan, Federal Reserve Bank of San Francsco Economc Revew, 3, pp2333 Obstfeld, M., (1982), Can We Sterlze? Theory and Practce, AEA Papers and Proceedngs, 72, pp 459 Rajan, R.S. and R. Sregar, (01), Prvate Captal Flows n East Asa: Boom, Bust and Beyond, n G. de Brouwer (ed) Fnancal Markets and Polces n East Asa, Routledge, Forthcomng Rajan, R.S., R. Sregar and I. Sugema, (01), Why Was There a Precrss Captal Inflow Boom n Southeast Asa?, mmeo, August 21
Resen, H. and H. Yeches, (1993). Tme Varyng Estmates on the Openness of the Captal Account n Korea and Tawan. Journal of Development Economcs 41, pp 285305 Warr, P.G., (1998), Thaland, n McLeod and Garnaut (eds), East Asa n Crss: from beng a mracle to needng one. Routledge 22
Table 1 Captal Inflows as a percentage of GDP INDONESIA MALAYSIA PHILIPPINES THAILAND FDI flows 1.7 7.2 1.8 1.6 Portfolo flows 0.5 0.0 0.2 1.4 Other 3.0 2.9 2.1 8.5 Total 5.1 10.2 4.1 11.5 Reserves* 1.7 5.1 1.8 4.3 Source: Rajan and Sregar (01) and IMF Note: Negatve sgn means ncrease n reserves Drect Investment Portfolo Flows Other Total Table 2 Net Prvate Captal Inflows 1994 04 1994 * 1995 ** 1996 1997 1998 1999 00 01 02 03 04 44.7 55.8 53.4 56.5 56.1 66.4 67.4.5 53.8 70.0 77.2.8 22.3 32.5 6.7 8.1 56.1 19.8 56.9 59.6 5.5 12.0 4.7 21.2 33.5 25.5 116.4 113.9 91.7 6 31.2 22.8 9.4 70.3 99.3 119.4 37.6 52.2 8.6 4.5 9.6 25.4 52.8 79.8 Source: IMF World Economc Outlook September 04, except for September 03, * and September 02, **. Data ncludes Emergng Asa and newly ndustralsng Asa categores of the IMF. All fgures are n US Bllons. 23
Table 3 Sterlsaton Coeffcents KOREA THAILAND INDONESIA MALAYSIA PHILIPPINES Sterlsaton Coeffcent 1.11 (3.81) 0.91 (8.15) 0.76 (7.07) 0.94 (12.70) 0.97 (10.24) Source: Cavol and Rajan (05). Results based on a sterlsaton equaton as follows: DA t = λ 0 + λ 1 FA t +λ 2 FA t1 + ε t The estmates presented are those for the contemporaneous sterlsaton coeffcents, λ 1 as ths s qute standard n estmatons of ths type. Sample: 1990.1 1997.3 (monthly observatons) ( ) denotes 1% sgnfcant level Table 4 Interest Rate Model Captal Moblty and Domestc Influences Dependent Varable: Money market Rate, t, 0.02 0.11 KOREA THAILAND INDONESIA MALAYSIA PHILIPPINES OLS TSLS OLS TSLS OLS TSLS OLS TSLS OLS TSLS Const 19.17 65.86 12.32 16.97 21.71 0.82 5.17 90.77 (2.67) 14.99 (0.96) (1.34) (0.27) (1.11) (1.10) (0.19) (0.51) (2.56) ** 101.89 (2.72) * 0.34 0.13 0.15 0.12 (2.65) 0.39 (1.95) * 0.59 (2.35) ** 0.83 (2.77) 0.36 (2.38) ** 0.56 (1.66) * 0.03 (0.82) (0.59) (0.21) (0.17) f t 0.06 0.05 0.03 0.03 0.27 0.03 0.001 0.01 (2.08) ** (0.10) (0.44) (0.) (0.97) (1.75) (3.68) (0.92) (0.52) (0.94) m t1 6.47 3.53 1.92 7.98 0.17 0.81 3.41 3.83 (4.70) 6.37 (4.71) 8.75 (1.43) (0.62) (0.62) (1.58) (0.70) (1.00) (0.51) (0.55) π e t+1 0. 0.37 0.44 0.33 1.52 7.72 0.002 0.05 0.55 0. (4.) (2.95) (2.03) ** (1.34) (0.59) (1.96) ** (0.10) (0.58) (1.91) * (1.84) * p t 13.22 6.25 3.96 16.84 0.17 0.76 24.25 30.46 (3.22) 14.94 (2.12) ** 24.83 (1.) (0.37) (0.52) (1.43) (0.12) (0.31) (1.16) (1.) y t 2.23 (1.03) 3.38 (0.76) 0.43 (0.06) 0.70 (0.09) 3.94 (0.89) 5.17 (1.11) 0.35 (0.79) 1. (0.91) 4.17 (0.43) 7.64 (0.) Adj 0.78 0.76 0. 0.52 0.68 0.42 0.95 0.88 0.09 0.02 R 2 DW 1.76 1.77 1.73 1.82 1.94 1.59 2.14 2.16 1.98 1.91 Obs 89 89 88 77* 89 88 89 89 89 89 *(**)( ), 10% (5%)(1%) sgnfcant levels, respectvely 24
Table 5 Test for Lagged Effects usng VAR Model. Wald Test for Coeffcent Restrctons Effect of Lagged * Effect of Lagged f Effect of Lagged I χ 2 Prob χ 2 Prob χ 2 Prob KOREA 12 lags 14.24 0.29 11.50 0.49 42.97 0.00 6 lags 15.47 0.02 6.77 0.34 74.88 0.00 3 lags 12.61 0.01 5.71 0.13 92.78 0.00 THAILAND 12 lags 6.14 0.91 7.42 0.83 11.55 0.48 6 lags 7.48 0.28 6.21 0. 14.19 0.03 3 lags 1.61 0.66 8.36 0.04 13.24 0.00 INDONESIA 12 lags 6.75 0.87 16. 0.17 26.33 0.01 6 lags 7.58 0.27 12.94 0.04 31.12 0.00 3 lags 10.16 0.02 0.49 0.92 28.75 0.00 MALAYSIA 12 lags 9.72 0.64 15.76 0..43 0.00 6 lags 6.27 0.39 23.66 0.00 281. 0.00 3 lags 2.88 0.41 19.17 0.00 621.50 0.00 PHILIPPINES 12 lags 14.56 0.27 19.65 0.07 17.04 0.15 6 lags 4.54 0. 5.00 0.54 10.39 0.11 3 lags 2.29 0.51 1.08 0.78 0.89 0.83 Note: Based on the VAR model for, f and * n equaton (12). The sample s 1990:1 to 1997:5 25
Fgure 1 Fgure 2 Impulse Response Functons Varance Decomposton of 1.00 Korea Korea 0. 0. 0. 0. 0. 0. * * Mo nths Mo nths 2.00 Thaland Thaland 1.50 1.00 0.50 0.50 1.00 * * Mo nths Mo nths 1.00 Indonesa Indonesa 0. 0. 0. 0. 0. * * 0. Mo nths 0.30 Malaysa Malaysa 0. 0.10 0.10 0. 0.30 * * Mo nths Mo nths 2.50 Phlppnes Phlppnes 2.00 1.50 1.00 0.50 0.50 1.00 * * Mo nths Months 26
Appendx 1 1A. Augmented DckeyFuller (ADF) Test for Money Market Rates Country ADF 1% CV 5% CV 10% CV Notes Test Statstc Korea 3.22 4.06 3.46 3.16 Intercept, trend, no lagged dfferences Thaland 3. 3.51 2.89 2.58 Intercept, no trend, no lagged dfferences Indonesa 2.96 3.51 2.89 2.58 Intercept, no trend, no lagged dfferences Malaysa 1.72 3.51 2.89 2.58 Intercept, no trend, 1 lagged dfference Phlppnes 7.92 4.06 3.46 3.16 Intercept, trend, no lagged dfferences 1B. ADF Test for Resduals of Interest Rate Regresson (2.12) Country Korea (OLS) Korea (TSLS) Malaysa (OLS) Malaysa (TSLS) ADF Test Statstc 1% CV 5% CV 10% CV Notes 8.43 2.58 1.94 1.62 No ntercept, no trend, no lagged dfferences 8.33 2.58 1.94 1.62 No ntercept, no trend, no lagged dfferences 7.99 2.58 1.94 1.62 Intercept, no trend, 2 lagged dfferences 8.42 2.58 1.94 1.62 Intercept, no trend, 1 lagged dfference 27
Appendx 2. Impulse Response Functons for 12, 6 and 3 lag model 0. Korea, 12 lags 1.00 Korea, 6 lags 1.00 Korea, 3 lags 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. * 0. * 0. * 0. 0. 0. 0. 0. 0. 1.50 Thaland, 12 lags 2.00 Thaland, 6 lags 2.00 Thaland, 3 lags 1.00 1.50 1.50 0.50 * 1.00 0.50 * 1.00 0.50 * 0.50 0.50 0.50 1.00 1.00 1.00 1.00 Indonesa, 12 lags 1. Indonesa, 6 lags 1. Indonesa, 3 lags 0. 1.00 1. 0. 0. 0. 0. * 0. 0. 0. 0. 0. * 1.00 0. 0. 0. 0. 0. * 0. 0. 0. 0. Malaysa, 12 lags 0.30 Malaysa, 6 lags 0.30 Malaysa, 3 lags 0.15 0.10 0.05 0. 0.10 0. 0.10 0.05 0.10 0.15 * 0.10 0. * 0.10 0. * 0. 0.30 0.30 2.50 Phlppnes, 12 lags 4.00 Phlppnes, 6 lags 4.50 Phlppnes, 3 lags 2.00 3.00 4.00 3.50 1.50 1.00 0.50 * 2.00 1.00 * 3.00 2.50 2.00 1.50 1.00 0.50 * 0.50 1.00 1.00 2.00 0.50 1.00 28
Appendx 2. Varance Decompostons for 12, 6 and 3 lag model Korea, 12 lags Korea, 6 lags Korea, 3 lags * * * Thaland, 12 lags Thaland, 6 lags Thaland, 3 lags * * * Indonesa, 12 lags Indonesa, 6 lags Indonesa, 3 lags * * * Malaysa, 12 lags Malaysa, 6 lags Malaysa, 3 lags * * * Phlppnes, 12 lags Phlppnes, 6 lags Phlppnes, 3 lags * * * 29