Monetary Transmission in a Crawling Peg Regime: Evidence from Costa Rica. by: Abstract

Transcription

1 Prelmnary, do not quote wthout permsson of the authors. Monetary Transmsson n a Crawlng Peg Regme: Evdence from Costa Rca by: Melana Flores, Alexander W. Hoffmaster, Jorge Madrgal, and Lorely Vllalobos (Frst draft, 6/19/00, Ths draft 8/18/00) Abstract Ths paper dscusses monetary transmsson n Costa Rca that has successfully mantaned a crawlng peg regme for ffteen years. Ths exchange rate regme coupled wth captal moblty hnders the central bank s ablty to conduct monetary polcy. Abstractng from perfect captal moblty, however, the central bank has some room to set nterest rates n the short-run. The evdence suggests that nterest rate shocks have farly standard effects on the economy, that vansh when external shocks are accounted for. In short, Costa Rca s ablty to conduct monetary polcy s lmted and n large part mports ts monetary polcy ts tradng partners. JEL Classfcaton: a, b, and c. e-mal addresses: florespm@bccr.f.cr, hoffmasteraa@bccr.f.cr, madrgalbj@bccr.f.cr, vllalobosml@bccr.f.cr. The authors thank... Correspondng author. Respectvely, Banco Central de Costa Rca (BCCR) Research department, IMF Research department and BCCR Offce of the Presdent, BCCR Research department, and BCCR Research department. The vews expressed n ths paper are those of the authors and should not be assocated wth the BCCR or the IMF.

2 Introducton Monetary polcy and ts transmsson have been the subject of consderable attenton n the past decade. A farly detaled account of the effects and channels through whch monetary polcy exerts s effect on the economy has been provded by among others Bernanke, Clarda, Chrstano, Echenbaum, Evans, Gal, Gertler, Sms, and ther co-authors (see Bernanke and Gertler, 1995 and references theren). And whle there s a broad consensus that monetary polcy s neutral n the long-run, there s consderable evdence that monetary polcy has substantal effects on economc actvty that last up to two years. Wth few exceptons, notably Mshkn and Savastano (2000), the lterature has focused prmarly on developed countres wth flexble exchange rates. There s consderable less understandng and emprcal evdence of the workngs of monetary polcy n developng countres. Asde for the wder dversty of economc structures and nsttutonal frameworks, monetary authortes n developng countres tend to be more concerned wth nomnal exchange rate fluctuatons than ther developed country counterparts. Ths concern for the nomnal exchange rate condtons the central bank s ablty to conduct ndependent monetary polces, and the transmsson of monetary polcy. Abstractng from perfect captal moblty, the lower degree of flexblty n the exchange rate reduces the scope of ndependent monetary polcy, but does not entrely elmnate t n the short-run. Ths paper examnes monetary transmsson n Costa Rca where the exchange rate has followed successfully a crawlng peg for ffteen years. Not surprsngly, nterest rate movements are condtoned by nternatonal condtons and the defense of the exchange rate crawlng peg. For nstance, nterest rates and the rate of the exchange rate crawl were adjusted to restore the uncovered party condton, when nternatonal reserves fell sgnfcantly n (Fgure 1). 1 The surge n nternatonal reserve that followed was 1 The party condton s obtaned usng the sx month US dollar lbor rate, the actual exchange rate deprecaton for the correspondng sx month perod, and a constant three percentage pont country rsk.

3 - 3 - assocated wth a declne n domestc nterest rates and of crawl of the exchange rate. A smlar pattern s observed throughout the 1990's, the last epsode when nterest rates and the rate of crawl were adjusted to arrest the declnes n nternatonal reserves was n These stylzed facts makes Costa Rca a partcularly nterestng case to examne how the exchange regme n developng countres mpnges upon the central bank s ablty to conduct ndependent monetary. Moreover, Costa Rca provdes a relatvely clean data set for econometrc analyss as t has not been subjected to a major macroeconomc or exchange rate crss snce the Latn Amercan debt crss n the early 1980 s. To ths effect ths paper estmates a seres of VAR models to examne the effect of nterest rate shocks on economy, especally on economc actvty and prces. The evdence suggests that the effects of nterest rate shocks are farly standard, although these effects contrast starkly wth the effects when world nflaton controls for external shocks. In partcular, the mpact of nterest rate shocks on the economy vrtually dsappears. Ths paper contnues by examnng the mportance of the credt channel that could play an mportant role n the transmsson of monetary gven underdeveloped captal markets. Also, evdence s provded to document the response of banks loan and depost rates when monetary polcy s tghtened, and verfes the dfferences n the responses of state-owned versus prvate banks. The rest of the paper conssts of three sectons. Secton 2 provdes a bref descrpton of the VAR technques used. Specfcally, generalzed VAR technques are used to document the hstorcal correlaton n the data. Secton 3 contans the bulk of the emprcal evdence for the seres of models examned. The dscusson centers on the responses to nterest rate shock, although a full set of mpulse responses are provded. Secton 4 summarzes the man results.

4 Measurng the effects of nterest rate shocks VAR framework. The models used to study the macroeconomc effects of nterest rate shocks can be expressed by the followng reduced-form VAR model: x = d = d ( L) ( L) t 1 t d d ( L) x ( L) x t 1 t µ µ x where, and x correspond respectvely to the nterest rate (BEM, sx months), and a vector of k-1 domestc varables (n logs) that are specfed below. The lag polynomals of p th order are denoted by d sj (L), the shocks to each equaton are denoted by µ s, and the vector of shocks of the system µ s such that E[µ]=0 and E[µµ ]=Ω. Generalzed mpulse responses. Followng Koop et al. (1996) and Pesaran and Shn (1998) ths paper calculates generalzed mpulse responses (GIR) that account for all of the hstorcal correlaton n the data. These responses are not obtaned by mposng (zero) restrctons on the contemporaneous correlaton as n standard VAR analyss, and they are unque snce they are not subject to the composton effect are unque. Ther dsadvantage s that they cannot be nterpreted as reflectng the effect of pure (orthogonal) shock, rather they fully reflect the hstorcal correlaton n the data. The GIR are defned as the dfference between the expected value of a varable condtoned on a partcular shock and ts expected value n the absence of that partcular shock. The expected values are obtaned assumng that µ s dstrbuted as a multvarate normal, wth zero mean, and covarance matrx Ω. Assumng that the shock to the nterest rate equals one ½ standard error, µ,0 =σ,, the GIR of yt =[, x] can be expressed as: GIR y 1/ 2 1 ( t, µ, 0 = σ, ) = D( L) Ω σ, where D(L), Ω, σ, are respectvely the lag polynomal matrx wth typcal element d sj (L), the column of Ω that corresponds to the nterest rate, and the varance of the nterest rate (see Pesaran and Shn, 1998).

5 - 5 - The functonal form of the GIR are related to standard mpulse responses but are conceptually dfferent constructs. Contrary to standard mpulse responses, GIR do not assume that the shocks of nterest are orthogonal. Thus, GIR are not desgned to dentfy pure shocks rather they are desgned to capture all the hstorcal correlaton n the data. Note that GIR are not obtaned usng tmng restrctons that typcally lead to assume a recursve system. Thus, GIR are not subject to the composton makng them unque. Specfc models. The GIR are calculated for two models, each comprsed of three varants that successvely add varables of nterest. Model 1 focuses on macroeconomc responses, and the frst varant defnes x=[ p, y] thereby examnng the basc macroeconomc effects on prces and output. The second varant adds the nomnal exchange rate, e, so that x=[e, p, y]. Ths varant accounts for the movements n the exchange rate that have accompaned shocks to the nterest rate (see Fgure 2). The thrd varant adds world prces, as proxed by the PPI n the US, p*, so that x=[p*, e, p, y]. Ths varant completes the exchange rate determnaton, as the crawlng peg has been roughly determned as the dfference between world (US) nflaton and programmed domestc nflaton. For ths varant the GIR to shock n p* are of partcular nterest. These shocks are dentfed usng the assumpton of a small open economy, so that domestc shocks do not affect p* (see Hoffmaster y Roldós, 2001). Model 2 focuses on the role of credt n the transmsson of the nterest rate shock. The frst varant defnes x=[ cr, p, y] where cr s prvate sector credt. As n the frst model, the second and thrd varants of the second model are obtaned by successvely addng the exchange rate and world prces, so that x s defned respectvely as x=[cr, e, p, y] and x=[p*, cr, p, e, y]. To quantfy the mportance of the credt channel, the mpulse responses n Model 2 are also computed assumng that credt s an exogenous varables,.e., the level of credt does not vary followng an nterest rate shock. Usng the notaton above, the GIR assumng credt s exogenous can be expressed as: GIR y 1/ ~ ~ 2 1 ( t, µ, 0 = σ,, crt = cr0, σ, cr = 0) = D( L) Ω σ,

6 - 6 - where ~ denotes the system where the prvate sector credt s exogenous,.e., cr t =cr 0, and the correlaton between the nterest rate and credt s set to zero. Comparng these responses to the standard GIR, provdes a farly straghtforward way to quantfy the role of the credt channel. As above responses to p* are of nterest n the thrd varant. In addton, ths paper consders a thrd model that focus on the bankng sector, so that x=[r, r_dep, r_loan, cr], where r_dep and r_loan are respectvely the nterest rates on deposts and loans. Ths model s estmated thrce, usng nterest rate and credt data for the bankng sector, stated-owned banks, and prvate banks. These results are partcularly useful to examne the delays that may exst between changes n the nterest rate and movements n banks loan and depost rates. Moreover, potental dfferences n the behavor of publc and prvate banks--gven the structure of the bankng sector, consstng of a few large stateowned banks and several smaller prvate banks--can be examned. 3. Impulse responses to nterest rate shocks. 2 The mpulse responses of the models dscussed above are contaned n a seres of fgures. The fgures are arranged so that columns correspond to specfc shocks and rows correspond to specfc varables n the models. Fgures numbered ( ) contan the mpulse responses of the three varants of Model 1 (Model 2). For nstance, followng a shock to the nterest rate (e_r) the response of nflaton (Dlp) n Model 1 varant 1 are contaned n the frst column and second row of Fgure 2.1. The mpulse responses to a shock n the nterest rate for the bankng sector model are presented n Fgure 4, where columns contan the responses respectvely of the bankng sector, state-owned, and prvate banks. As noted before the dscusson of the mpulse responses centers on the shock to the nterest rate. 2 Detals of the data defntons are contaned n the appendx Table A1. The emprcal evdence n ths paper s based on VAR models usng sx lags. Ths evdence, however, s not unduly senstve to the number of lags used. The mpulse responses were also computed for VAR models nne lags, and where approprate the man dfferences are note. Note that the model tend to exhbt unstable behavor wth more lags.

7 Impulse responses for Model 1. 3 Consder the mpulse responses for Model 1 varant 1 n Fgure 2.1. A shock to the nterest rate s assocated wth an ncrease n nflaton, and a declne n economc growth both persstng more than two years. These responses confrm the conventonal wsdom for developed countres that output tends to respond more quckly than prces. Note that although the response of nflaton s perverse, t has been documented n the early work on monetary polcy for the US economy (Ltterman and Wess, 1985, and Sms, 1986). The postve correlaton between nterest rates and prces, known as prce puzzle, was solved by addng commodty prces to the model because those prces tend to forecast both nflaton and the pre-emptve nterest rate movements (for a crtcal revew of the prce puzzle, see Hanson, 1999). Pre-emptve nterest rate movements have for the most part not characterzed monetary polcy here. A more lkely explanaton s assocated wth the exchange rate. Monetary authortes have tended to ncrease nterest rate when the exchange rate tends to deprecate more than programmed (see Flores, et al., 2000). Ths coupled wth some exchange rate pass-through to prces could explan the postve correlaton of nterest rates and prces. Ths s dscussed below. Consder the mpulse responses for Model 1 varant 2 n Fgure 2.2. A shock to the nterest rate s assocated wth an exchange rate deprecaton as posted above, and contnues to be assocated wth an ncrease n nflaton and wth a declne n economc growth. The response of nflaton and output growth are smlar to those above, although they are somewhat smaller. Thus, the pass-through of the exchange rate to prces s partally assocated wth the 3 The mpulse responses for Model 1 usng nne lags are qualtatvely the same as those n Fgures In partcular for varants 1 and 2, output tends to fall, and the prce puzzle contnues to be assocated wth the deprecaton that accompanes an nterest rate shock. Also, the mpact of nterest rates on the economy s drastcally reduced when world condtons are accounted for n varant 3. The effect on the exchange rate deprecaton, however, s a bt larger than n Fgure 2.3 but stll much less than n varants 1 and 2.

8 - 8 - prce-puzzle. Note that to a large extent the exchange rate peg has been set usng a purchasng power party (PPP) rule vs-à-vs the US, ths s ssue s consdered below. Before turnng to the next set of responses, note that the smaller declne n output growth s assocated wth a deprecaton of the exchange rate, that for most of the frst sx months s a real deprecaton (nomnal deprecaton exceedng nflaton). Consderng the responses to a shock n nomnal exchange rate (e_dle, column two), the exchange rate deprecaton leads to an ntal ncrease n economc growth followed by a smaller expanson. These results are nterestng because they seem to be assocated wth an ncrease n real nterest rates and a real deprecaton. The latter s at odds wth the contractonary effects of devaluaton n Latn Amerca (Edwards, 1989). Consder the mpulse responses for Model 1 varant 3 n Fgure 2.3. A shock to world nflaton (e_dlp*, column 1) s assocated wth qualtatvely smlar responses to those observed n the two prevous varants of Model 1 for a shock n the nterest rate. Namely, the nterest rate, exchange rate deprecaton and nflaton ncrease, and output growth falls. The man dfference, however, s n the tme path of the nomnal varables. The responses of nterest rates and exchange rate deprecaton are delayed respectvely by about two and four months, and the ncrease n nflaton materalzes earler. Note that the responses to a shock n world nflaton conform well to the vew that the exchange rate essentally follows a PPP rule. An ncrease n the world nflaton translates nto domestc nflaton va a (relatve) PPP rule as the exchange rate remans essentally unchanged for about four months. The authortes seem to rse nterest rate frst before adjustng the exchange rate to offset the real apprecaton. In ths scenaro, output growth declnes are assocated wth a real apprecaton. Perhaps the most revealng mpulse responses of the exchange rate regme are those assocated wth shocks to the nterest rate (e_r, column 2). Although the shock and the dynamc path of the nterest rate s essentally the same as n the two prevous varants of Model 1, the effect on the economy as markedly reduced. The exchange rate response s

9 - 9 - vrtually flat for up to sx months, the nflaton response s smaller and substantally less persstent, and the output response s small and undefned. These responses suggest that nterest rate shocks have lost vrtually all ther mpact on the economy when external shocks are explctly accounted for n the model. The path of real nterest rates followng a shock n nflaton (e_dlp) n Model 1 s worth notng. In general, durng the frst three months followng the shock, real nterest rates tend to fall as the response n nterest rates s less than that of nflaton. For a few months afterward, real nterest rates tend to ncrease as the response n nterest rates exceeds that of nflaton. Ths path, when averaged over tme, s consstent wth Corbo (1999) who fnds a small (long-run) response of nterest rates to nflaton n hs estmated nterest rate reacton functons. Moreover, the larger response of nterest rates when nflaton exceeds ts targeted level (see Flores et al., 2000) seems to capture ncreases n real nterest rates observed a few months followng the nflaton shock. 3.2 Impulse responses for Model 2. 4 Consder the responses for Model 2 varant 1 n Fgure 3.1. A shock to the nterest rate s assocated wth a declne n credt growth that perssts for about two years, and wth an ncrease n nflaton, and a declne n economc growth both persstng more than two years (sold lnes, e_r, column 1). When credt s taken as an exogenous varable the nterest rate shock and the assocated nflaton lasts longer. 5 The output growth response s essentally the 4 The mpulse responses for Model 2 usng nne lags are qualtatvely the same as those n Fgures In partcular for varants 1 and 2, output tends to fall, and the prce puzzle contnues to be assocated wth the deprecaton that accompanes an nterest rate shock, and the effect of credt s small. Also, the mpact of nterest rates on the economy s drastcally reduced when world condtons are accounted for n varant 3. The effect of credt remans small. 5 Note that there are two dfferent ways to compute the GIR when credt s exogenous. The frst s assumng a once and for all ncrease n credt, so that the GIR reflects the effect of a sustaned ncrease n the growth of credt. These responses would assume that the growth of (contnued )

10 same durng the frst twelve months, and ts tends to last longer when credt s exogenous. The output growth responses are puzzlng as they suggest a perverse effect of credt: hgher credt s assocated wth lower output growth (broken lne). Ths perverse output response, however, do no show up n the other varants of Model 2. Consder the responses for Model 2 varant 2 n Fgure 3.2. A shock to the nterest rate s assocated wth a deprecaton of the exchange rate, an ncrease n nflaton, a declne n credt and output growth. These responses are comparable to those n Model 1, although the deprecaton and nflaton are less persstent, and the declne n output growth s a bt more persstent. When credt s taken as an exogenous varable the persstence of nflaton and of the exchange rate deprecaton s smlar to that n Model 1. Perhaps a more nterestng result s that declne n output growth s short-lved, and confned to the frst two or three months followng the nterest rate shock. Ths response s suggestve of a credt channel operatng on the real sector of the economy. However, ths result s weaker n fnal varant of Model 2. Consder the responses for Model 2 varant 3 n Fgure 3.3. A shock to world nflaton (e_dlp*, column 1) s assocated wth qualtatvely smlar responses to those observed n the two prevous varants of Model 2 for a shock to the nterest rate. Namely, the nterest rate, the exchange rate, and nflaton ncrease, and output growth falls. As above, the man dfference s n the tme path, whereby the responses of nterest rates and exchange rate deprecaton are delayed, and the response of nflaton manfests ts self earler. The responses when credt s taken as exogenous (broken lnes) are smlar to those n Model 1. In partcular, they suggest that the output growth declne s less and the exchange rate deprecaton s greater after about sx months. credt s essentally a constant plus a shock. The second s to characterze credt as an AR process that does not depend on the rest of the varables n the system, as assumed here. Thus, followng a one tme shock to credt, credt s allowed to follow a dynamc path that depends only on tself. The man advantage of allowng credt to follow an AR process s that credt remans a statonary varable makng these responses easer to compare to those when the credt equaton reflects the lags of other varables.

11 As above, the mpulse response to a shock to the nterest rates (e_r, column 2) contnue to be suggestve of how the exchange rate regme hnders the effect of ndependent monetary polcy. To a large extent, the mpact of the ncrease n the nterest rate are dmnshed substantally, wth the excepton of the declne n credt. The declne n credt s smaller than those that follow a shock to world nflaton (column 1) but that are smlar to those observed n prevous verson of Model 2. It s possble that domestc banks ncrease loan rates when nterest rates ncrease, even after controllng for the effect of external factors, the demand for domestc credt declnes. In ths case, the small declne n output growth could be nterpreted as evdence of a lmted credt channel or f the credt channel s mportant, then t would be consstent wth a swtch toward external credt sources. Fnally, consder brefly the mpulse responses for shocks to nflaton. In all varants, an ncrease n nflaton s assocated wth ncreases n nterest rates, and lagged declnes n credt that vary from about two to four months. As before real nterest rates tend to fall durng the frst few months followng the shock and tend to ncrease for some months afterward. As before, the long-run response of real nterest rates to shock to nflaton wll tend to be small as t wll average ths response. Also, the declnes n credt seem to concde wth ncreases n real nterest rates and declnes n output growth. Ths suggests that the demand for credt could be drvng the movements n credt followng a shock to nflaton. 3.3 Impulse responses for bankng sector model. 6 Consder the responses for the bankng sector to a shock n domestc nterest rates n Fgure 4. It s nterestng that both state-owned banks and prvate banks ncrease (on mpact) loan and depost rates by a smaller amount that the nterest rate shock. In partcular, an nterest rate shock of roughly 130 bps leads (on mpact) to ncreases that are roughly half as large. And although banks contnue to ncrease rate gradually for the frst three or four months, the 6 The mpulse responses for Model 3 usng nne lags are not only qualtatvely the same, they are vrtually ndstngushable from Fgure 4.

12 ncrease never match the ncrease n domestc nterest rates. Nonetheless, prvate banks seem to ncrease depost rates more and for a longer perod than do state-owned banks. Note that the ncreases n loan rates, whch are very smlar across banks, are less than the ncreases n depost rates. Ths suggests that spreads of loan over depost rates fall n both types of banks, but the fall s greater n prvate banks. Note that to a large extent, the credt responses observed, confrm the lagged responses observed n Model 2, and suggest that they are roughly two and three months respectvely for state-owned and prvate banks. These responses appear to be consstent wth the structure of the bankng sector and the dfferences n the composton of banks' asset and lablty postons (Table 1). In partcular, note that state-owned banks domnate the bankng system, wth and percent respectvely of total deposts and prvate sector loans. Ths suggests that prvate banks are lkely to "follow the lead" of state-owned banks. Ths s partcularly true for loan rates, because ncreases beyond those of state-owned banks wll translate nto decreased market shares. Ths could be behnd the smlar responses of loan rates. What s partcularly ntrgung s the magntude of the responses, as one would expect that rates move n unson under normal crcumstances. Ths could be assocate to the fact that ncreases have sgnfcant effects on the assets sde of banks balance sheets, partcularly state-owned banks, as they the rato of portfolo nvestments exceed n about 15 percent ther prvate sector loans. That s, the mpact of ncreases n the nterest rate on the cost of the banks operatons are partally offset by ncreases n the return of ther portfolos. It s not clear, however, why prvate banks ncrease depost rates more than state-owned banks. Perhaps ths s a strategy to capture a larger share of deposts n the bankng sector. 4. Concludng remarks. Monetary polcy and ts transmsson have been the subject of consderable attenton n the past decade. Lttle attenton, however, has been pad to examnng these ssues n developng countres, that are characterzed by a more dverse set of economc structures and

13 nsttutonal frameworks, than developed countres. Moreover, monetary authortes n developng countres are typcally more concerned wth fluctuatons n nomnal exchange rates than ther developed country counterparts. The role of these factors n monetary polcy and ts transmsson, and how the exchange rate regme can effectvely hnder the pursut of ndependent monetary polcy has not yet been subjected to exhaustve emprcal examnaton. Ths paper examnes monetary transmsson n Costa Rca where the exchange rate has followed successfully a crawlng peg for ffteen years. Ths makes Costa Rca a partcularly nterestng case to examne how the exchange regme n developng countres mpnges upon the central bank s ablty to conduct ndependent monetary. Moreover, Costa Rca provdes a relatvely clean data set for econometrc analyss as t has not been subjected to a major macroeconomc or exchange rate crss snce the Latn Amercan debt crss n the early 1980 s. In general, the emprcal evdence can be summed up as follows. Interest rate shocks appear to be assocated wth a prce-puzzle, whereby nflaton ncreases followng a shock to nterest rates. The puzzle can be partally explaned by the pass-through of nomnal exchange rate deprecaton that s assocated wth the nterest rate shock. However, wthout dentfyng a pure monetary shock (usng standard tmng assumptons) t s cavaler to push ths nterpretaton here. More mportantly, the mpact of nterest rate shocks on the economy vanshes when external shocks are accounted for. Ths suggests that nterest rates essentally follow world condtons and have very lmted mpact on the economy when they do not. Ths s qute revealng of how the exchange rate regme hnders the central bank s ablty to pursue ndependent monetary polcy. Regardng the credt channel of monetary polcy, some evdence s found that the declne n credt that follows a shock to nterest rates s assocated wth a declne n output growth. In general the mpact on output growth s concentrated n sx to 24 months followng the shock. Surprsngly, however, the credt channel effect appears to be relatvely small. It s possble that the responses here underestmate credt channel effects because the declne n credt s lkely to be less than that capture by the mpulse responses. Ths s because facng the

14 observed ncreases n loan rates, frms could substtute away from domestc credt to foregn credt. Unfortunately, the lack of data on foregn prvate credt s weak, makng t dffcult to construct the requred tme seres to verfy the extent to whch frms swtch to foregn sources of credt followng an nterest rate shock. The response of the bankng sector to an ncrease n domestc nterest rates dffers somewhat between state-owned and prvate banks. Although both loan and depost rates ncrease followng an ncrease n domestc rates, these ncreases are lmted to about half of the ncrease n domestc rates. Ths could be assocated to the structure of the bankng sector, wth prvate banks followng the lead of state-owned banks, and the large share of portfolo nvestments n the banks asset poston. Moreover, the ncreases n depost rate exceed those of loan rates, suggestng that followng an ncrease n nterest rates the spread of loan over depost rates fall for both types of banks.

15 References Bernanke, Ben S., and Mark Gertler, Insde the Black Box: The Credt Channel of Monetary Polcy Transmsson, Journal of Economc Perspectves, Vol. 9, Fall 1995, pp Corbo, Vttoro, Monetary Polcy n Latn Amerca n the 90 s, presented at the Thrd Annual Conference of the Central Bank of Chle: Monetary Polcy: Rules and Transmsson Mechansms, September 20-21, 1999 Dhrymes, Phoebus J., Introductory Econometrcs, Sprnger-Verlag, (Nueva York, 1978). Edwards, Sebastán, Real Exchange Rates, Devaluaton, and Adjustment: Exchange Rate Polces n Developng Countres, Cambrdge, Mass.: MIT Press, Flores, Melana, Alexander W. Hoffmaster, Jorge Madrgal y Lorely Vllalobos, Una Funcón de Reaccón para el Banco Central de Costa Rca, Banco Central de Costa Rca, Nota de Investgacón #2-00, August Hamlton, James D., Tme Seres Analyss, Prnceton Unversty Press (Prnceton, New Jersey, 1994). Hanson, Mchael S., "On the Identfcaton of Monetary Polcy: the "Prce Puzzle" Reconsdered, NBER Monetary Economcs Conference, November, Hoffmaster, Alexander W., y Jorge E. Roldós, The Sources of Macroeconomc Fluctuatons n Developng Countres: Brazl and Korea, Journal of Macroeconomcs, (forthcomng, 2001). Koop, Gary, M. Hashem Pesaran, y Smon N. Potter, Impulse Response Analyss n Nonlnear Multvarate Models, Journal of Econometrcs, Vol. 7, 1996, Ltterman, R. B., and L. Wess, "Money, Real Interest Rates, and Output: A Renterpretaton of Postwar U.S. data, Econometrca, Vol. 53, pp Mshkn, Frederc S., and Mguel Savastano, Monetary Polcy n Latn Amerca n the 90 s, IMF mmeo, February, Pesaran, M. Hashem, y Yongcheol Shn, Generalzed Impulse Response Analyss n Lnear Multvarate Models, Economc Letters, Vol. 58, 1998, Sms, Chrstopher A., "Are Forecastng Models Usable for Polcy Analyss," Federal Reserve Bank of Mnneapols Quarterly Revew, Wnter, 1986, pp

16 Table 1. Stylzed Facts of the Bankng System, State-owned banks Prvately held banks Share of total deposts n the bankng system Share of prvate sector loans Rato of total deposts to prvate sector loans Rato of portfolo nvestments to prvate sector loans Rato of demand to tme deposts Note: State owned banks are comprsed of three state owned banks (Banco Naconal, Banco de Costa Rca, and Banco de Crédto Agrícola de Cartago) and Banco Popular. Prvately held banks are the 20 banks recognzed by the bankng supervsory board (SUGEF). Total deposts are comprsed of all bankng sector labltes (checkng accounts plus tme deposts). Prvate sector loans are defned as the stock of commercal and mortgage credt from bankng system to prvate agents. Portfolo nvestments are comprsed of dollar tme deposts plus other nvestments n BCCR and banks' holdngs of domestc publc debt, wch s aproxmatedly a half of total nvestments. The data for prvate banks demand deposts are net,.e. banks' sght labltes mnus banks' sght assets. Also, the rato of demand to tme deposts of prvate banks has doubled when computed over the past three years. In part ths reflects the growth n sght labltes that now exceed sght assets.

17 Fgure 1. Stylzed Macroeconomc Fact 31.5 Inflaton 10 Output Growth % 17.5 % World Inflaton 600 Internatonal Reserves % 2 0 Mll.US$ Devaluaton and Inflaton Dfferental 56 Interest Rate and Uncovered Party Rate % 24 % Tasa de nterés BEM Devaluacón p-p* Tasa de nterés de ndferenca

18 Fgure 2.1. Impulse responses, model 1 varant 1 µ_ µ_dlp µ_dly Dlp Dly

19 Fgure 2.2. Impulse responses, model 1 varant 2 µ_ µ_dle µ_dlp µ_dly Dle Dlp Dly

20 Fgure 2.3. Impulse responses, model 1 varant 3 µ_dlp* µ_ µ_dle µ_dlp µ_dly Dle Dlp Dly Dlp*

21 Fgure 3.1. Impulse responses, model 2 varant 1 µ_ µ_dlcr µ_dlp µ_dly Dlcr Dlp Dly

22 Fgure 3.2. Impulse responses, model 2 varant 2 µ_ µ_dlcr µ_dle µ_dlp µ_dly Dlcr Dle Dlp Dly

23 Fgure 3.3. Impulse responses, model 2 varant 3 µ_dlp* µ_ µ_dlcr µ_dle µ_dlp µ_dly Dlcr Dle Dlp Dly Dlp*

24 Fgure 4. Impulse responses to an nterest rate shock n: Bankng sector State-owned banks Prvate banks _Dep _Loan Dlcr Spread

25 Table A1 Descrpton of Varables Varables Descrpton Interest rate of BEM, sx months. Dlp 12-month rate of growth of consumer prce ndex (CPI), 1995=100 Dly Dle 12-month rate of growth of the monthly ndex of economc actvty (IMAE), 1991= month rate of growth of nomnal exchange rate (e). Dlp* 12-month rate of growth of the US producer prce ndex (PPI), 1995=100 Dlcpr Dlcpr_e Dlcpr_p _dep _loan 12-month rate of (end of perod) growth of total prvate sector credt. 12-month rate of (end of perod) growth of prvate sector credt of state-owned banks. 12-month rate of (end of perod) growth of prvate sector of prvately held banks. Monthly average of nterest rates on sx month deposts. * US T bll rate, (sx months). Spread Party nterest rate Inflaton Dfferental Monthly average of loan nterest rate, weghted by new credts of state-owned and prvate banks dfference of _loan and _dep + * + et+ 1 2 et 1 6 dfference of Dlp and Dlp* 1 *100 * 2 + φ ; φ, country rsk = 3