WHAT CAN UIP TELL US ABOUT EXCHANGE RATE REGIMES? SOME EMPIRICAL EVIDENCE FROM EAST ASIA. Manfred Keil and Ling Cao *

Transcription

1 WHAT CAN UIP TELL US ABOUT EXCHANGE RATE REGIMES? SOME EMPIRICAL EVIDENCE FROM EAST ASIA by Manfred Keil and Ling Cao * * Keil (manfred.keil@claremonmckenna.edu): Claremon McKenna College, Claremon Graduae Universiy, and Freeman Program in Asian Poliical Economy a The Claremon Colleges; Cao: Claremon Graduae Universiy. Keil is corresponding auhor. We hank Tom Wille and oher paricipans of he Claremon/KIEP conference on he Poliical Economy of Regional Inegraion (Claremon, November 2005) for useful commens. The usual disclaimer applies.

2 Wha Can UIP Tell Us abou Exchange Rae Regimes? Some Empirical Evidence from Eas Asia Absrac We follow he recen approach in he lieraure o use parameers from he uncovered ineres pariy condiion o deermine exchange rae regimes in seleced Eas Asian counries. We firs find ha, in general, movemens in hese parameers mirror de faco exchange rae regime changes following he Asian Financial Crisis. However, he same holds for a group of conrol counries ha neiher changed exchange rae regimes nor were par of he Asian Financial Crisis. Furher analysis shows ha he resuls are primarily driven by inflaion convergence beween counries owards he end of he 90s. Key words: uncovered ineres pariy, exchange rae regimes, open economy rilemma JEL Classificaion: F31, F32, F33 1

3 1. Inroducion Sudens of open economy macroeconomics learn early abou he concep of he macroeconomic policy rilemma or he impossible riniy (see, e.g. Krugman and Obsfeld, 2006; , Mankiw, 2007; ). According o i, counries canno pursue domesic moneary auonomy, a fixed exchange rae regime, and a policy of allowing free capial movemen a he same ime. Insead, only wo of hese hree opions are available o policy makers. Pu differenly, once a counry decides o fix is exchange rae, i hen mus choose beween an independen moneary policy and allowing free capial flows. If one believes in he effeciveness of acive macroeconomic policy, counries poenially have anoher policy ool available o conduc sabilizaion policy: fiscal policy. However, is applicaion is limied by he presence of long inside lags and a widespread unwillingness o raise ax raes. In addiion, fiscal policy for sabilizaion purposes is infrequenly employed due o he presence of large budge deficis in many counries. This suggess he use of an independen moneary policy o conduc acive policy in he presence of free capial mobiliy. The European Union, he U.S., he U.K., and Canada come o mind as examples in having chosen his opion, alhough counries wihin he European Union earlier oped, a imes, for a fixed exchange rae regime wih capial mobiliy. Bu do all counries which allow for heir exchange rae o be flexible really gain moneary freedom? Assume ha moneary policy is se hrough a shor-erm ineres rae arge and ha capial markes are ighly inegraed. Ignoring he risk premium for he momen, or assuming a leas ha i does no vary, hen seing he domesic shor-erm ineres rae o a level differen from ha of a base rae (he U.S. reasury bill, say) should 2

4 resul immediaely in an expeced change in he exchange rae wih respec o some base rae. If he domesic counry canno somach large exchange rae swings resuling from seing domesic ineres raes independenly, hen even hose counries ha have a flexible exchange rae canno really conduc independen moneary policy. This phenomenon is referred o as he Fear of Floaing (Calvo and Reinhar, 2002). In is presence, he rilemma is reduced o a single choice: o fix or no o fix he exchange rae. Selecing one or he oher has no implicaion for moneary freedom, and as a resul, he choice of exchange rae regime does no depend on he naion s desire o pursue an independen moneary policy. In general, world financial markes have become increasingly inegraed across borders. This has imporan and esable policy implicaions, especially for smaller and lesser developed counries. In paricular, one would posulae ha irrespecive of he chosen exchange rae regime (and here are more han simply he fixed versus flexible variey), here is no difference in he behavior of domesic ineres raes wih regard o he base ineres rae. Wha is he evidence? Using he uncovered ineres pariy (UIP) relaionship and panel daa, Frankel e. al. (2000, 2002) find evidence supporing he fear of floaing hypohesis, while Shambaugh (2004) rejecs i. The purpose of his paper is o shed furher ligh on his conroversy. We do so by using he Asian Financial Crisis of as a quasi experimen. Many of he Eas Asian counries included in our sample (Indonesia, Korea, Malaysia, he Philippines, Thailand) changed heir exchange rae regime for he pos-crisis period while ohers (Hong Kong, Singapore) did no. Accordingly we would expec corresponding changes in cerain parameers of he UIP ha is consisen wih his behavior. The evidence for he 3

5 counries in our sample a firs seems o suppor Shambaugh s conclusion in general: counries wih more (less) flexible exchange raes appear o have a more (less) auonomous moneary policy. However, we observe he same paern of change in parameers for counries ha were no par of he Asian Financial Crisis, and ha did no change exchange rae regimes as a resul of i (Ausralia, New Zealand, Canada). 1 We inerpre his resul as implying ha UIP is no really well suied o es for he ype of exchange rae regime and ha i canno ell us as much abou he effec of hese regimes on moneary policy as is commonly assumed in he lieraure. Insead he resuls we observe over he sample period are more consisen wih inflaion convergence beween counries. The paper proceeds as follows: secion 2 presens he relaionship beween UIP and exchange rae regimes, and relaed specificaion issues. In secion 3, using monhly daa, we sar our empirical analysis by looking a a single eniy, Korea, a counry which is believed o have moved o a more freely floaing exchange rae regime following he Asian Financial Crisis. Secion 4 analyzes daa from a panel of Eas Asian counries and a conrol group of counries using boh annual and monhly daa. A final secion concludes. 2. UIP and Is Relaionship o Exchange Rae Regimes Following Frankel e. al. (2000, 2002) and Shambaugh (2004), we will focus on he UIP relaionship o gain insighs ino wheher counries wih more flexible (fixed) exchange raes have a higher likelihood o pursue a more (less) independen moneary policy. A more independen moneary policy here is defined as having he abiliy o se and o 1 From here on, we will use he following symbols for counries: Indonesia IDN, Korea KOR, Malaysia MYS, he Philippines PHL, Thailand THA, Hong Kong HKG, Singapore SGP, Ausralia AUS, New Zealand NZL, Canada CAN. 4

6 move he domesic ineres rae ( R ) independenly and a a level which differs from he base rae ( R ), i.e. f R R and f ΔR Δ R. 2 Generally, nominal ineres raes could be f differen beween counries for reasons oher han moneary policy. There is also no reason o expec ha changes in hese ineres raes should always be he same even under a fixed exchange rae sysem. The difference beween he wo ineres raes can be described by wo condiions, covered ineres pariy (1) and UIP (2). f R = R +( f + s ) (1) n f R = R + E ( s s ) + rp (2) + n where f +n is he forward exchange rae for n periods ino he fuure, s is he curren spo rae (boh exchange raes in logs), 3 and rp is he risk premium. The expeced exchange rae and he risk premium are unobservable variables. Boh pariy condiions basically assume for he law of one price o hold. The connecion beween (1) and (2) can be seen hrough he following decomposiion (Frankel, 1991): f f R R E ( s s ) = ( f E s )+[ R R ( f s )] (3) + n + n + n + n The firs erm on he r.h.s. is ofen referred o as he currency risk premium. The second erm is he counry or poliical risk premium. The laer is zero if CIP holds exacly. Hence if CIP canno be rejeced, hen rejecion of UIP means ha forward raes do no equal expeced fuure exchange raes. 4 2 These are nominal ineres raes on similar asses and measured in local currency. 3 Throughou his paper, he exchange rae is quoed as he domesic price of foreign currency. 4 For a survey of hese resuls, see Froo and Thaler (1990) and Engel (1995). 5

7 In principle, we could use equaion (1) o deermine he exchange rae regime prevalen in a given counry. This would involve running a regression of he following ype: R = β + β ( R + f + s ) + u (4) f 0 1 n where in he case of a fixed exchange rae regime we would expec β 0 and β 1 o be zero and one respecively coinciding wih a high regression 2 R. In he case of a flexible exchange rae regime, we would expec o find no relaionship beween he domesic and base rae plus forward premium variable. There are complicaions for inference if counries follow he same independen moneary policy following a common shock, bu we leave his for laer discussion. Mos sudies focusing on equaion (4) have used daa from indusrial counries and have been more ineresed in esing for CIP raher han choosing i as a ool o classify exchange rae regimes. Few sudies on CIP have looked a developing counries (de Brouwer, 1999, and Bansal and Dahlquis, 2000). We see he reason for his in he absence of sufficienly liquid forward foreign exchange markes in developing counries. Moreover, even when hey exis, daa is no easily available. 5 Regardless, Wille, e al. (2002) poin ou: (S)ubsanial deviaions from covered ineres pariy are a good indicaion ha capial mobiliy is less han perfec..[however]... [f]inding ha covered ineres pariy holds is consisen wih eiher high or low capial mobiliy, and here is no good reason o presume ha he magniudes of deviaions from ineres pariy will provide a reasonable proxy for he degree of inernaional capial mobiliy. In erms of modern heory, he appropriae measure of capial mobiliy is he exen o which uncovered raher han covered ineres pariy holds. (424-5) 5 We have been unable o replicae Edwards and Khan s (1985) sudy and would love o hear from anyone in he profession who has been able o do so and is willing o provide us wih he daa. See also Ahn (1994). 6

8 As a resul of hese argumens, UIP is used o exrac informaion abou exchange rae regimes. Equaion (4) is herefore replaced by: R = β + β [ R +E ( s + s )] + u (5) f 0 1 n where eiher he inercep or he error erm may conain he risk premium, depending on one s view of i. In a fixed exchange rae regime ( E( s + ) = s ), and esing, a firs glance, should be sraighforward. Wih flexible exchange raes, equaion (2) is solved for he expeced change in he exchange rae. Hence differences beween he domesic and base ineres rae drive expeced changes in he exchange rae. In his case, β 1 should be close o zero and he regression here is fear of floaing or he presence of capial conrols. 6 n 2 R should be low, unless, of course, An alernaive roue for using UIP o deermine exchange rae regimes is o find proxies for ( E( s s )) and o include hese in UIP regressions. The + n following four measures have ypically been suggesed in he lieraure: (i) perfec foresigh E ( s s ) = s s + n + n (ii) exrapolaive expecaions E ( s s ) = s s + n n (iii) saic expecaions E ( s + s ) = 0 n survey (iv) survey daa E ( s s ) = s s + n + n 6 Noe, however, ha saisical sudies of he relaionship beween ineres rae differences and laer depreciaion raes show ha he ineres rae difference has been a very bad predicor, in he sense ha i has failed o cach any of he large swings in exchange raes. Even worse, he ineres difference has, on average, failed o predic correcly he direcion in which he spo rae would change. Krugman and Obsfeld (2007; 596). 7

9 In (iv), survey daa is obained by inerviewing marke paricipans. This is regularly done by eniies such as he Economis Inelligence Uni (EIU) Currency Consensus Forecas. 7 Using any of he assumpions abou expeced exchange rae changes, we could now esimae equaion (5) were i no for some furher complicaions. The firs involves he classificaion of exchange rae regimes. Then here are poenial economeric esimaion problems frequenly encounered in ime series analysis. 2.1 Exchange Rae Classificaion One way o use equaion (5) is o es wheher or no counries wih a more flexible exchange rae sysem conduc moneary policy more independenly han hose wih a fixed exchange rae regime. To do his you firs have o idenify counry specific exchange rae regimes as more flexible (nonpegged counry) or more fixed (pegged counry). Equaion (5) can hen be esimaed separaely for boh ypes of regimes, wih he expecaion ha he pegged counries have a slope closer o uniy and a higher regression ( 2 R when compared o he nonpegged counries. For example, he erm E( s s )) does no only vanish under a fixed exchange rae regime, bu also for a + n credible peg. Boh Frankel e.al. (2000, 2002) and Shambaugh (2004) use pooled regression resuls (among oher echniques) o see if here is empirical relevance o he fear of floaing argumen. Alernaively, and more ineresingly for our se of counries, you can choose sample periods when some counries are commonly assumed o have swiched exchange rae 7 Examples are Frankel and Froo (1987, 1989), Taylor (1989), MacDonald and Torance (1990), Cavaglia e al. (1993), Chinn and Frankel, (1995). MAS (1999) uses four Asian exchange rae markes. 8

10 regimes, while ohers have no. The slope and goodness of fi parameers in equaion (5) should hen change correspondingly. As in he case discussed in he previous paragraph, he analysis here requires a specificaion of exchange rae regime. Mos commonly, he lieraure disinguishes beween he Inernaional Moneary Fund de jure classificaion and a de faco classificaion based on echniques developed by Calvo and Reinhar (2002) or Reinhar and Rogoff (2004). The IMF idenifies hree caegories and 15 furher subcaegories beween 1975 o Since 1999, i inroduced a new classificaion which also akes ino accoun acual raher han declared behavior. 8 De faco classificaions are derived from measures ypically relaed o exchange rae and reserve movemens, and exchange rae movemens ouside cerain bands. We will use boh de faco and de jure classificaions below. 2.2 Economeric Issues: Levels vs. Differences Having seled he issue of exchange rae regime classificaion in he previous secion, we could simply esimae equaion (5) by OLS. Assuming ha he base rae is exogenous, ha here are no large ouliers in eiher he dependen or explanaory variables, and ha here are no omied variables, hen he OLS esimaor is consisen if he ineres rae variables have a saionary disribuion. This resul holds irrespecive of wheher or no he errors are auocorrelaed. Inference is valid, as long as heeroskedasiciy- and auocorrelaion-consisen (HAC) sandard errors are used. 8 The basic hree caegories of he earlier specificaion were pegs, limied flexibiliy, and more flexibiliy. The newer eigh caegories were: exchange rae arrangemen wih no separae legal ender, currency board arrangemen, convenional pegged arrangemen (peg agains a single currency or a basked of currencies), pegged exchange rae wihin horizonal bands, crawling peg, crawling band, managed floaing wih no preannounced pah for exchange rae, and independenly floaing. 9

11 Assuming for he momen ha here are no srucural breaks during our sample period, hen for he domesic and base ineres raes o have a saionary disribuion requires ha neiher one of he wo series conains a sochasic rend. In oher words, we need for boh variables o be inegraed of order zero (I(0)). If, on he oher hand, boh were I(1), hen inference is problemaic, since he -saisics are no normally disribued (even in large samples), and here may be a spurious regression problem (Granger and Newbold, 1974; Phillips, 1986; for an excellen summary see, Sock and Wason (2007), chaper 14). There are differen pahs ou of his siuaion in he case of I(1) ineres raes. One is o esimae an equaion specified in differences, i.e. ΔR = [ ΔR + ΔE ( s s )] + u (6) f * β 1 + n In addiion, if he wo ineres raes are coinegraed, hen an Error (Equilibrium) Correcion Mechanism (ECM) form should be employed. The null hypohesis here is ha ineres raes are more likely o be coinegraed for pegged counries han for ohers. We would be somewha less concerned abou his issue if inference regarding he fear of floaing and moneary independence was robus irrespecive of using equaion (5) or (6). Unforunaely, his is no he case. The issue also separaes Frankel e.al. (2000, 2002), who esimae levels equaions, from Shambaugh (2004), who prefers some form of differenced variable esimaion. Boh come o differen conclusions regarding he fear of floaing argumen and Shambaugh (2004; 314) in paricular feels ha his is primarily due o esimaing (6) raher han (5). Frankel e.al. (2002) do no ignore he poenial problem resuling from dynamics and address i by saing ha hey look a equaion (5) as an equilibrium expression, alhough hey do no perform a coinegraion es for heir 10

12 monhly panel. Furhermore, hey argue ha a priori we would expec ineres raes o be I(0) variables, (Frankel e.al., 2002; 12). While we lean more owards he idea ha nominal ineres raes, real ineres raes, and inflaionary expecaions are saionary, especially over longer periods and for counries ha did no experience episodes of hyperinflaion, 9 we will esimae boh equaions (5) and (6) below o conras poenial differences and sensiiviies in he conclusions. 3. Esimaion Resuls: Korea We begin wih our empirical invesigaion by looking a a single counry, Korea. The sample period is January 1990 (1990:01) o June 2003 (2003:06). We focus on a single counry firs because many of he phenomena which we observe for our group of counries can be analyzed in more deail using a single eniy. We picked Souh Korea for his secion because i is one of he mos developed counries in our sample of seven Eas Asian counries and i experienced a change in exchange rae regime. 3.1 Choice of Exchange Rae Regime Classificaion: Korea Korea s exchange rae regimes is classified as follows: during he pre-crisis period of our sample Reinhar and Rogoff (2004) lis Korea as having a pre announced crawling band unil November 1994, and a crawling peg o he U.S. dollar unil November For he period during which Korea was heavily affeced by he Asian Financial crisis (December 9 For he U.S., we can rejec he presence of a uni roo in he CPI inflaion rae for he sample period 1974:I-2002:IV a he 1% level, and for he sample period 1947:I-2002:IV a he 0.1% level using an Augmened Dickey-Fuller (ADF) es. Furhermore, he ADF es has low power, and by using he more powerful DF-GDL es (Ellio, Rohenberg, and Sock, 1996) srenghens his resul. 11

13 June 1998), hey lis i as freely falling, 10 followed by a freely floaing exchange rae regime (July 1998 o December 2001). The de jure classificaion used by he IMF is managed floaing unil November Since we found ha resuls for Eas Asia are very sensiive o he sample beginning and end poins due o ouliers surrounding he Asian Financial Crisis, we looked a he exchange rae series more carefully and found considerable movemen during Ocober and November To avoid hese ouliers and heir consequences for OLS properies, we chose Sepember 1997 as he las observaion of our pre-crisis period. As for he saring poin of he poscrisis period, we seled on January Alhough he de faco classificaion has he won as freely floaing saring half a year earlier, his classificaion is based on exchange rae behavior only. However, here was much variaion in reserves beginning in July 1998, which suggess ha he cenral bank inervened heavily in he exchange rae marke. Sill, here is lile doub ha he won has been much more flexible during he pos-crisis period han during he pre-crisis period (McKinnon and Schnabl (2003), Hernandez and Moniel (2003), Wille and Kim (2004), Kim, Kim, and Wang (2005)). 11 There is, of course, he addiional issue of capial conrols, which has a significan effec on he UIP relaionship and is abiliy o signal moneary independence. Shambaugh (2004) allows for he ineracion of capial conrols wih he base rae variable, and we would like o do he same. However, here is no monhly capial conrol index available o our knowledge. Even if we had such an index a our disposal, mos observers dae he opening of capial markes for Korea o We doub ha 10 This caegory is used for counries wih a 12-monh rae of inflaion above 40%. 11 For a comparison of crisis/normalcy performance of UIP, see Flood and Rose (2002). 12

14 ineracing he wo variables would have much of an effec from hen on unil he end of our pre-crisis period in Sepember As a resul, we ignore he issue for now. In summary, we sele on classifying he pre-crisis period as inermediae and he pos-crisis period as floaing. We believe ha his is consisen wih he de jure and he de faco classificaion. 3.2 Esimaion Resuls Korea Wha really maers for he purpose of our analysis is ha Korea moved from a peg (precrisis) o a nonpeg (pos-crisis). Hence we expec he slope o be closer o uniy in he earlier period wih a coinciding higher regression R 2. Formally KOR US R = β + β R + u or Δ R = β0 + β1δ R + u KOR US 0 1 and H : β > β ; R > R peg nonpeg peg nonpeg Table (1) presens he regression resuls for boh levels equaions (column (1)) corresponding o equaion (5) above, and he difference specificaion (column (2)) corresponding o equaion (6). In boh cases we assume saic expecaions as presened above. Columns (3) and (4) are Shambaugh s panel resuls using differences, boh for his enire sample and for he 1990s only. We enered his slope coefficien for he pegs in he pre-crisis row and for he nonpegs in he pos-crisis row. The levels equaion (column (1)) produces he expeced resuls for he slope coefficiens. For Korea, he size of he slope coefficien decreased for he nonpeg period when compared o he peg period. For boh periods, he slopes are saisically significan, and he slope for he pre-crisis period is also significanly differen from uniy. The slope 13

15 coefficiens are pleasanly close o Shambaugh s resuls for he enire esimaion period (column (3)), alhough Shambaugh does no find a saisically significan slope for he 90s (column (4)). The Korean surprise is he regression 2 R, which increases dramaically for he pos-crisis period despie a decrease in he slope coefficien. Hence we canno rejec he null hypohesis for he slope, bu we rejec i for he regression Alhough he resuls in column (2) are closer in spiri o column (4), here is an exraordinary high slope coefficien during he peg period. I is roughly hree imes as large as he one found in column (4) for he peg. However, i is no saisically significanly differen from uniy. Even so, he regression 2 R. 2 R is surprisingly low. This coefficien drops o half he size in column (4) for he nonpeg period, alhough i remains saisically significan. When faced wih his ype of resul, he firs inclinaion is o check for daa enry errors. Having ruled ou his possibiliy, i was comforing o find anoher sudy which lised a slope coefficien of similar magniude: Kim and Lee (2004) show an even larger slope coefficien (2.946) for a similar sample period using a differen esimaor. While one can se ou o search for heoreical explanaions of coefficiens of his magniude (see, e.g. Shambaugh, 2004; 306) i is ofen useful o plo he daa. Figures 1a and 1b presen he scaer plo and he regression line for boh he pre- and pos-crisis period. Noe ha boh are drawn o he same scale. The reason for he high regression 2 R is ha here was much less variaion in Korean ineres raes during he pos-crisis period when compared o he pre-crisis period. The base rae variaion was roughly he same. However, he reason for he higher Korean variaion seems o be he resul of higher average ineres raes during he pre-crisis 14

16 period. I is well known ha he variance of he inflaion rae increases wih is level. To us i seems ha he Korean resul is driven more by inflaion convergence han by a swich from a peg o a more flexible exchange rae regime. This may also explain he size of he slope coefficien in column (2): a given change in he base rae produces much more of a variaion in he domesic rae. To be fair, we need o poin ou ha, following Frankel e.al. (2000, 2002), we have used monhly daa, while Shambaugh (2004) works wih annual (panel) daa. Our sample period is much shorer han his, and Shambaugh hope[s] he dynamics have largely seled so ha he can pool he daa across counries. (312) We could pursue his issue furher here, bu find i disracing from our main argumen. Appendix 1, however, produces ess for coinegraion, a dynamic OLS (DOLS) esimaor o idenify he coinegraing vecor and he speed of adjusmen coefficien, a general o specific specificaion search o find he ECM represenaion, he use of differen expecaions hypohesis, and allows for dynamics o sele o an equilibrium over a longer period. The upsho of his analysis is ha, according o he ess based on UIP, he pos-crisis period raher han he pre-crisis period appears o have been a peg. 4. Muliple Counry Analysis The conclusion from secion 3 was ha, quie surprisingly, UIP esimaion does no sugges ha Korea followed a more flexible exchange rae regime in he pos-crisis period when compared o he pre-crisis period. However, his does no make sense since here is srong evidence from oher sources ha i did. We herefore mus quesion he 15

17 abiliy of UIP o idenify exchange rae regimes. Perhaps here are oher reasons for observing he resuls we did for Korea. Of course, we could simply sae ha here is an ineresing sory behind every observaion, and ha heory reveals iself in daa when we looking a averages for a panel of counries. Indeed his is one of he main asses of comparaive economic performance. We will do so in his secion by using daa for seven Eas Asian counries (HKG, IDN, KOR, MYS, PHL, SGP, THA), and hree conrol counries ha are believed o have had a freely floaing exchange rae regime hroughou he sample period (AUS, CAN, NZL). Noe ha he hree non-asian counries were no as much affeced by he Asian Financial Crisis as he seven Eas Asian counries. Table 5 presens de jure and de faco exchange rae classificaions for he seven Eas Asian counries for he pre- and pos-crisis period. There is much variaion in he behavior beween counries. Four counries moved from a more pegged siuaion owards a more floaing regime (IDN, KOR, PHL, THA), one wen in he opposie direcion (MYS), while wo remained he same (HKG, SGP). Table 6 presens he esimaion resuls using annual daa for boh levels and differences. Focusing on he levels resuls firs, we find he following. For he wo counries which were expeced o mainain he same slope coefficien, i acually declined. The slope for MYS was expeced o increase, and i did. Three of he four counries for which he slope coefficien was supposed o decrease, show a decrease. However, all hree free floaing developed counries, which were no direcly affeced by he Asian Financial Crisis, have a subsanial decrease in he slope coefficien. Similar resuls hold for he differences. 16

18 Having esimaed hese equaions using annual daa, we have o confess ha we do no have much faih in regression coefficiens obained from only four observaions 12 or so. We simply use hese as a descripive device and would raher no alk abou goodness of fi (or inference for ha maer). To overcome he small sample size problem, we reesimaed he coefficiens using monhly daa. Resuls are repored in Table 7. Looking a he slope coefficiens for he levels regressions firs, we find ha HKG, IDN, KOR, THA, and MYS follow he expeced paern beween he pre- and pos-crisis period. However SKP, CAN, NZL, and AUS all show significan decreases in he slope coefficiens, wih PHL enering wih an incorrec sign. The regression 2 R display a mixed paern, wih mos of hem going agains he null hypohesis. Moving o he difference equaions nex, we observe a coefficien decrease as expeced only for PHL and KOR, and only he laer is saisically significan. However, he slope coefficien also decreases for HKG, CAN, NZL, and AUS (again, significanly so only for AUS). Only Korea s goodness of fi decreases as expeced. For mos counries i increases for he pos-crisis period. We now have a dilemma. We would like o base our conclusions on he monhly daa esimaes of Table 7, bu are worried ha, in Shambaugh s words, dynamics have no seled, and as a resul we should repor he resuls from he annual sample period, or perhaps averages across counries. However, his does no make sense when you only have very few annual observaions available. Bu wha if he annual coefficiens, for which dynamic adjusmens have been compleed, are similar o he slope coefficiens 12 The phrase Are you kidding me? comes o mind if we alked abou inference now. 17

19 based on he monhly daa sample? To invesigae his possibiliy, we ran he following regression for he level and difference regressions combined. monhly β 1 = annual β 1 + (0.08) (0.07) 2 = 1,, 20, R = 0.88 e I does appear ha, on average, he slope coefficiens from he monhly sample period esimaion are very similar o hose from he annual sample period. Figure 2 confirms his impression. Hence, and for he purpose of he problem a hand, i seems all righ o base he analysis on he monhly daa analysis. Taking he evidence from boh annual and monhly daa ogeher, as well as levels and differences, i seems fair o say ha slope coefficiens on average decreased. More imporanly, his is he case wheher hey were supposed o do so when counries moved from a peg o a nonpeg, or remained on he same ype of regime. Mos damaging o he idea ha UIP can ell us somehing abou exchange rae regimes is he fac ha slope coefficiens of he floaers AUS, NZL, and CAN also decreased. This seems o sugges ha we should search for an explanaion elsewhere. The quesion hen becomes, wha is i ha drives hese resuls? One possible explanaion is ha here has been inflaion convergence hroughou he 90s, which would explain he paern of he slope coefficiens concerned. Noe ha R = β + β R 0 1 ( ρ + π) = β + β ( ρ+ π) 0 1 Δ p = ( β + β ( ρ ρ )) + β Δp f f * 0 1 f Δ p = β + β Δp f f 18

20 where ρ is he real ineres rae, and π represens inflaionary expecaions. To ge o he final expression, we have assumed ha real ineres raes are equal beween counries (or a leas ha heir difference remains he same over ime), and ha agens have saic expecaions. Now consider a siuaion where inflaion during he pre-crisis period is subsanially higher in he domesic counry when compared o he base counry. Furhermore, hink of he inflaion rae falling for boh counries, bu more so for he domesic inflaion rae han for he inflaion rae in he base counry. For illusraion purposes, le he inflaion rae for he domesic economy fall by five percenage poins, while he base counry inflaion rae fell by 2.5 percenage poins during he pre-crisis. This would resul in a slope coefficien of Alhough inflaion raes coninued o fall during he pos-crisis period, hey did no fall by as much anymore, since hey were already a a lower level a he beginning of he pos-crisis period. Again, for illusraion purposes, consider an addiion fall of 1.5 percenage poin for he domesic inflaion rae level, and a 1 percenage poin change in he base counry. In ha case, he slope coefficien would be 1.5, i.e. i would have fallen. Table 8 presens he pre-crisis and pos-crisis inflaion raes for he counries in our sample. Clearly inflaion raes beween he U.S. and he oher counries have converged. A similar resul would hold for he formulaion in differences. 5. Conclusion We se ou o use UIP o idenify poenial exchange rae regime changes for seven Eas Asian counries for he pre and pos Asian Financial Crisis period. This seemed o be a 13 The calculaions are based on he idea ha he inflaion rae falls by his amoun in a single period, or every period over a longer sample. You can make reasonable adjusmens o find smaller slope coefficiens. 19

21 worhwhile exercise for wo reasons. Firs, hose counries which followed a more flexible exchange rae regime are supposed o have gained more moneary independence unless here is fear of floaing. Changes in parameers from he UIP equaion should be able o confirm his one way or anoher. Second, we can hink of he episode as a quasi experimen: some of he counries changed exchange rae regimes, while ohers did no. In addiion, we have a conrol group of counries ouside of Asia which is believed by mos o have coninued wih is exchange rae regime regardless of he crisis. Wha we found is ha UIP is no well suied o idenify exchange rae regime changes during he period we analyzed and for he counries we looked a. 14 In our view, auhors have been asking UIP o reveal more han can be reasonably expeced. There are various reasons for his, such as variaion in resuls depending on assumpions abou he saionariy of ineres rae variables, changes in capial mobiliy, ime frequency of daa used, assumpions abou expecaions formaion, ec. Mos imporanly, we believe ha economic phenomena oher han he choice of exchange rae regimes may explain why some auhors have found evidence agains he fear of floaing, which implies ha counries wih more flexible exchange raes can pursue more of an independen moneary policy. In paricular, i is inflaion convergence during he laer pars of he 90s which seems o explain much of he fall in he slope coefficien of he UIP relaionship. Our resuls srongly sugges ha saemens regarding moneary independence and choice of exchange rae regime should ake his ino accoun more carefully. 14 MacCallum (1994) also believes ha deviaions from he UIP may be due o moneary policy decisions of cenral banks and proposes o include a moneary policy reacion in an expression for he UID. Bird and Rajan (2001) and Rajan, Siregar and Sugema (2002) offer bank-based explanaions for persisen ineres rae differenials in Eas Asia. Also see Edwards and Khan (1985) and Wille e al. (2002). 20

22 Appendix 1: Furher Dynamics involving he Korean UIP Specificaion The analysis in his appendix coninues o analyze various dynamic aspecs of he UIP relaionship for Korea. The resuls in secion 3 sugges ha following he analysis using levels and difference specificaions, here was some doub on wheher Korea had pursued a more flexible exchange rae regime, as is commonly assumed in he lieraure. For example, he regression R 2 in he levels equaion rises dramaically from o 0.667, i.e. weny imes. Furhermore, he slope coefficien for he difference specificaion is 2 saisically significan for he pos-crisis period wih a regression R ha is five imes as high as Shambaugh s (2004) value for nonpegs. The crucial quesion now is wheher we view regression (5) as a long-run relaionship or as a regression beween wo saionary variables in he shor run. If we viewed he level regression (5) as represening a long erm moneary equilibrium and wih compleed policy reacion raher han shor-erm financial marke inegraion, hen saic regressions of his ype have made a comeback when esing for coinegraion (Engle and Granger, 1987). To deermine wheher he wo ineres raes are coinegraed, we esimae he saic regression as shown in column (1) of Table 1 and hen perform an ADF es on he residuals. For he pre crisis period he EG-ADF es saisic is Since he criical value a he 10% level is -3.12, we canno rejec he null hypohesis of no coinegraion even a his level. The resul holds when oher domesic variables such as inflaion, money and income growh are added o he coinegraing equaion. However, for he pos crisis period, he EG-ADF saisic is -3.73, resuling in rejecion of he null hypohesis of no coinegraion a he 5% (bu no a he 1% level). 15 Having found supporive evidence for he presence of coinegraion in he pos-crisis period, we use he dynamic OLS (DOLS) esimaor (Sock and Wason, 1993) o find he coinegraing vecor. This involves adding fuure, presen, and lagged changes p KOR f US in he U.S. ineres rae o he saic regression, i.e. R = β0 + θr + δ jδ R j + u. Our esimae of θ is wih a sandard error of 0.024, indicaing a long run relaionship beween he wo ineres raes, bu also one ha is some disance from uniy. These resuls are opposie o wha migh be expeced for a peg and a floa: our prior is ha ineres raes should be more likely o be coinegraed for he peg han for he floa. Shambaugh (2004, p.342) finds 29 cases where he null hypohesis of no coinegraion is rejeced. Of hese, 23 are pegs and only 3 are nonpegs. Using he DOLS esimaor, Shambaugh repors an average value of θ of 0.84, wih he majoriy lying beween 0.8 and 1.2, i.e. close o uniy. Noe ha for Korea, we can rejec he null hypohesis of θ = 1 even a he 1% level. We find i insrucive o invesigae how robus his resul is if we assumed ineres raes o be saionary, i.e. I(0) variables. In ha case, dynamics could be inroduced by j= p 15 The Johansen procedure produces he same resul, viz. no evidence of a coinegraing vecor for he firs period, bu rejecion of no coinegraion for he second period. 21

23 adding a lag dependen variable, e.g. hrough a parial adjusmen assumpion. Equaion (5) would be replaced by where R = R + E ( s s ) + rp (5a) * f + n R = R + λ( R R ) * 1 1 Equaion (5a) could be viewed as an equilibrium relaionship and he parial adjusmen could be he resul of he presence of capial conrols in he pos-crisis period. Column (2) in Table 2 shows he parial adjusmen resuls for he pre and pos-crisis esimaion period. (Column (1) is added for comparison.) Firs noe he increase in he regression R 2. As in he saic regression resuls, and according o prior expecaions, he slope coefficien for he peg is more han wice as large as for he floa. Furhermore, solving boh equaions for he saionary sae equilibrium resuls in a slope coefficien of 0.96 for he peg and 0.23 for he floa. The former is obviously close o uniy. Somewha surprisingly, he speed of adjusmen is higher for he floa han for he peg. The mean lag in he former is roughly hree monhs, while for he peg i is five and a half monhs. Also opposie o prior expecaions is he relaive magniude of he regression R 2, wih he floa allowing for less room o maneuver han he peg. Durbin s h suggess he presence of furher dynamic problems in he floa specificaion, bu no for he peg period. Finally, and perhaps mos imporanly, noe he robusness in he resuls when comparing he esimae obained from he DOLS esimaion above and he pos-crisis long-run coefficien for equaion (13) (parial adjusmen): boh are almos idenical. This is comforing given he difference in mehodology. Since we found evidence of coinegraion for he floa period, here should also be an ECM represenaion for he pos crisis period. We could follow Shambaugh (2004) here and simply esimae he relaionship l m f f f θα ( 0 1 α1 1) β1 1 γ j j δk k j= 1 k= 1 Δ R = + R R + Δ R + Δ R + Δ R + u Insead we follow Frankel e.al. (2002), who use he LSE/Hendry specificaion search of general o specific since he resul will be more parsimonious. 16 Having seled for a se of explanaory variables, a general Auoregressive Disribued lag model (ADL) is esimaed firs. Given ha we work wih monhly ineres rae daa, we fel ha an ADL(5,4) was sufficien. Table 3 shows he resuls for he General Unresriced Model (GUM). 17 As can be expeced from he number of regressors, he 16 see Hendry (1995; ) or Gujarai (1995), for a convenien summary. For deailed examples of earlier applied sudies in consumpion, see Davidson, Hendry, Srba and Yeo (1978); in he demand for money, Hendry and Mizon (1978). 17 [The GUM] is he mos general, esimable, saisical model ha can reasonably be posulaed iniially, given he presen sample of daa, previous empirical and heoreical research, and any insiuional and measuremen informaion available.... The GUM is also formulaed o conain he parsimonious, 22

24 sandard errors of he parameers are quie large, giving low -saisics. Some would argue ha his is due o high mulicollineariy among he explanaory variables, alhough on his see Hendry (1995 pp and p. 365). We are less concerned abou uncerainy surrounding he GUM coefficiens a his poin. Insead we noe ha all of he previously published ineres rae equaions are nesed in he GUM. Furhermore, we are ineresed in he paern of coefficiens o derive a more parsimonious model from he GUM, which will be funcions of hese parameers. This resuls in he following ECM represenaion: KOR Δ R = *** KOR US Δ R * Δ R ** Δ R US 1 ( ) (0.067) (0.082) (0.049) = 1999:1-2003:6, KOR US *** ( R 1 R 1) 0.184*** R US 1 (0.026) (0.020) 2 R = 0.780, SER = Since he parsimonious equaion is nesed in he GUM, we can use an F-es for he validiy of he resricions. The F(5,43) saisic is We herefore canno rejec he null hypohesis ha he resricions are valid. The solved form in Table 4 shows ha his is no surprising since he major feaures of he GUM are accouned for. The long run (saionary sae) soluion of he ECM equaion is R KOR = R US Again, i is comforing o see how close his resul is o he long run soluion calculaed from he P.A. equaion and he DOLS esimaion of he coinegraing vecor. Finally, we experimened wih alernaive expecaions hypohesis regarding he expeced change in he exchange rae. The resuls are repored in columns (3) and (4) of Table 2. Clearly, in comparing hese wo specificaions o column (1), he esimaed relaionship becomes subsanially weaker. For he pre crisis period, he slope coefficien is insignifican in boh columns (3) and (4). For he pos crisis period, i is reduced in size o beween one and eigh hundredh of is size in column (1). The same is rue for he pos crisis period, alhough he slope coefficien remains saisically significan. Looking a a graph of he relaionship, i becomes clear ha large changes in he proxied expeced exchange rae change are associaed wih relaively small movemens in he Korean ineres rae. Furhermore hese movemens in he exchange rae seem o dominae any change in he U.S. ineres rae. The resuls in columns (3) and (4) are quie discouraging when faced wih having o find a proxy for exchange rae expecaions. inerpreable, and invarian economeric model a which i is hoped he modeling exercise will end. Hendry (1995 p. 361) 23

25 References Ahn, B. (1994). Moneary Policy and he Deerminaion of he Ineres Rae and Exchange Rae in a Small Open Economy wih Increasing Capial Mobiliy, Working Paper No A, Federal Reserve Bank of S. Louis. Bansal, R. and M. Dahlquis (2000). The Forward Premium Puzzle: Differen Tales from Developed and Emerging Economies, Journal of Inernaional Economics, Bird, G. and R. Rajan (2001). Banks, Financial Liberalizaion and Financial Crises in Emerging Markes, The World Economy, Calvo, G. and C. Reinhar (2002). Fear of Floaing, Quarerly Journal of Economics, Cavaglia, S., Verschoor, W. and C. Wolff (1993). Furher Evidence on Exchange Rae Expecaions, Journal of Inernaional Money and Finance, Davidson, J., Hendry D., Srba, F., and S. Yeo (1978). Economeric Modeling of he Aggregae Time-Series Relaionship beween Consumers Expendiure and Income in he Unied Kingdom, Economic Journal, de Browuer, G. (1999). Financial Inegraion in Eas Asia, Cambridge: Cambridge Universiy Press. Edwards, S. and M. Khan (1985). Ineres Rae Deerminaion in Developing Counries, IMF Saff Papers, Ellio, G., Rohenberg, T. and J. Sock (1996). Efficien Tess for an Auoregressive Uni Roo, Economerica, Engle, R. (1995). The Forward Discoun Anomaly and he Risk Premium: A Survey of Recen Evidence, Working Paper No.5312, NBER. Engle, R. and C. Granger (1987). Coinegraion and Error Correcion: Represenaion, Esimaion, and Tesing, Economerica, Flood, R. and A. Rose (2002). Uncovered Ineres Pariy in Crisis, IMF Saff Papers, Frankel, J. (1991). Quanifying Inernaional Capial Mobiliy in he 1980s, in B. Bernheim and J. Shoven (ed.), Naional Saving and Economic Performance, Chicago: Universiy of Chicago Press. 24

26 Frankel, J. and K. Froo (1987). Using Survey Daa o Tes Sandard Proposiions Regarding Exchange Rae Expecaions, American Economic Review, Frankel, J. and K. Froo (1989). Forward Discoun Bias: Is I an Exchange Rae Premium? Quarerly Journal of Economics, Frankel, J., Schmukler, S., and L. Serven (2000). Wha Happens when he Fed Tighens? Ineres Rae Sensiiviy and Currency Regime, paper presened a he World Bank conference on Lain America and he Caribbean from Frankel, J., Schmukler, S., and L. Serven (2002). Global Transmission of Ineres Raes: Moneary Independence and Currency Regime, Working Paper No.8828, NBER. Froo, K. and R. Thaler (1990). Anomalies: Foreign Exchange, Journal of Economic Perspecives, Granger, C. and P. Newbold (1974). Spurious Regressions in Economerics, Journal of Economerics, Gujarai, D.N. (1995). Basic Economerics. 3 rd Ed. New York: McGraw-Hill. Hendry, D. (1995). Dynamic Economerics, Oxford: Oxford Universiy Press. Hendry, D. and G. Mizon (1978). Serial Correlaion as a Convenional Simplificaion, No a Nuisance, A Commen on a Sudy of he Demand for Money by he Bank of England. Economic Journal, Hernandez, L. and P. Moniel (2003). Pos-crisis exchange rae policy in five Asian counries: Filling in he hollow middle?, Journal of he Japanese and Inernaional Economies, 2003, Kim, S., S.H. Kim and Y. Wang (2005). Fear of floaing in Eas Asia?, Working Paper, Tufs Universiy. Krugman, P. and M. Obsfeld (2006). Inernaional Economics: Theory and Policy, Boson: Addison Wesley, 7 h ediion. McCallum, B. (1994). A Reconsideraion of he Uncovered Ineres Pariy Relaionship, Journal of Moneary Economics, MacDonald, R. and T. Torrance (1990). Expecaions Formaion and Risk in Four Foreign Exchange Markes, Oxford Economic Papers, Mankiw, N. (2007). Macroeonomics, New York: Worh Publishers, 6 h ediion. 25

27 McKinnon, R. and G. Schnabl (2003). Eas Asian Dollar Sandard, Fear of Floaing and Original Sin Paper, Working Paper No.03001, Sanford Universiy, Deparmen of Economics. Moneary Auhoriy of Singapore (MAS) (1999). How Well Did he Forward Marke Anicipae he Asian Currency Crisis: The Case of Four ASEAN Currencies, Occasional Paper No.13, Economics Deparmen. Phillips, P. (1986). Undersanding Spurious Regressions in Economerics, Journal of Economerics, Rajan, R., R. Siregar and I. Sugema (2003). Capial Flows and he Credi Transmission Channel in Souheas Asia, Journal of Inernaional Developmen, Reinhar, C. and K. Rogoff (2004). The Modern Hisory of Exchange Rae Arrangemens: A Reinerpreaion, Quarerly Journal of Economics, Shambaugh, J. (2004). The Effec of Fixed Exchange Raes on Moneary Policy, Quarerly Journal of Economics, Sock, J. and M. Wason (2007). Inroducion o Economerics, Boson: Addison Wesley, 2 nd ediion. Sock, J. and M. Wason (1993). A Simple Esimaor of Coinegraing Vecors in Higher-Order Inegraed Sysems, Economerica, Taylor, M. (1989). Expecaions, Risk and Uncerainy in he Foreign Exchange Marke: Some Resuls Based on Survey Daa, The Mancheser School, Wille, T., and Y. Kim (2004). Assessing Korea s Pos Crisis Managed Floa, Working Paper, Claremon Graduae Universiy, Bank of Korea Workshop, Seoul, Augus Wille, T., M. Keil and Y. Ahn (2002). Capial Mobiliy for Developing Counries May No be So High, Journal of Developmen Economics,

28 Figure 1a Korea M oney M arke Rae vs. US Treasury B ill Rae, 1990: : KRR USR Figure 1b Korea Money Marke Rae vs. US Treasury Bill Rae, 1999: : KRR USR 27

29 Figure 2: Regression of Monhly Slope Coefficiens on Annual Slope Coefficienss Monhly Slope Coefs Annual Slope Coefs Table 1 Effec of U.S. Ineres Raes on Korean Ineres Raes, Level and Differences Dependen Variable columns (1) and (2): Korea Money Marke Rae, 1990: :06. (4) Explanaory (1) (2) (3) Shambaugh Variable Levels Differences Shambaugh 90 U.S. In. Rae pre 0.320** (0.180) pos 0.230** (0.025) 1.736*** (0.598) 0.179** (0.100) 0.46** (0.04) 0.27*** (0.08) 0.56*** (0.06) 0.35 (0.25) Consan Summary Saisics 2 Adj. R pre 0.010*** (0.001) pos 0.003*** (0.0001) *** ( ) ( )???? pre pos Noe: Columns (1) and (2): Pre-crisis and pos crisis periods for columns are 1990: :09 and 1999: :06. Numbers in parenhesis are Newey-Wes HAC sandard errors. *** indicaes significance a 1% level, ** a 5% level, and * a 10% level of a one-sided -es. Columns (3) and (4) are from Shambaugh (2004) and for comparison only. The sample is a panel of 103 counries for he sample period , where he dependen variable is he local ineres rae. The explanaory variable is no always he U.S. ineres rae. The resul for he pre-crisis period is Shambaugh s peg slope coefficien, while he pos-crisis period is his resul for nonpegs. Column (4) presens Shambaugh s resuls for he 1990s. 28

30 Table 2 Effec of U.S. Ineres Raes on Korean Ineres Raes (Level) Dependen Variable: Korea Money Marke Rae, 1990: :06 (1) (2) (3) (4) Explanaory Saic Saic Perfec Exrapolaive Variable Expec. Expec. foresigh Expecaions USA R pre 0.320** (0.180) 0.170*** (0.065) (0.019) (0.027) pos 0.230** (0.025) 0.078*** (0.012) 0.006*** (0.002) 0.004** (0.002) Consan pre 0.010*** (0.001) 0.001** (0.0008) 0.011*** (0.0003) 0.011*** (0.0003) pos 0.003*** (0.0001) 0.001*** (0.0002) 0.004*** (0.0001) 0.004*** (0.0001) R 1 Summary Saisics 2 Adj. R pre *** (0.070) pos *** (0.042) pre pos Noe: Pre-crisis and pos crisis periods are 1990: :09 and 1999: :06. Numbers in parenhesis are Newey-Wes HAC sandard errors. *** indicaes significance a 1% level, ** a 5% level, and * a 10% level of a one-sided -es. 29

31 Table 3: General Unresriced Model (GUM) of he Ineres Rae Equaion lags j j (0.10) (0.152) (0.144) (0.093) (0.035) Δ R US Δ R j (0.101) (0.129) (0.101) (0.103) (0.072) Noe: Sample Period 1999:1-2003:6; HAC sandard errors in parenhesis, consan no repored 2 here, R = 0.98, SER = Table 4: Solved Form of he Ineres Rae Equaion lags j j 1 KOR Δ R US Δ R j