Credibility and Reputation : an Application of the “ External Circumstances ” Model for the Real Plan *

This paper analyzes credibility and reputation aspects of the Brazilian economic policy between August 1994 and December 1998. It uses an “external circumstances” model, which can be applied to countries with fixed or crawling-peg exchange rate policies. The model assumes that no government can conduct its economic policy with the single objective of inflation control, thoroughly ignoring the unemployment and growth paths. Therefore, in the presence of “external circumstances” (unexpected exogenous shocks) even a strong anti-inflationary government can be forced to devalue its exchange rate. The results here show that the government followed a consistent policy with inflation control while allowing for a gradual recovery of the competitiveness level.

After several unsuccessful attempts to stabilize the Brazilian economy during the 19 Os and early 1990s, t he Real P lan has fina lly managed to reduce infla tion systematically and consistently.Init ially, economic authorities a llowed the real to be cons iderably overvalued, since reducing infl ation was a priority.Hence, between July 1994 and February 1995, t here was a floating exchange rate policy that allowed for such appreciation.After t he Mexican crisis, from March 1995 onwards, an excha nge rate band regime was implemented, wh ich, in fact, resembled a crawli ng-peg regime.During the Asian crisis in October 1997, t here was a reduction in foreign reserves, which was reversed after a few months, causing reserves to accumulate again.Wit h t he Russian crisis in August 1998, extensive foreign reserve losses, and t he speculation against t he real, the government decided to abolish the exchange rate band regime a t the beginning of J anuary 1999.
In this paper we analyze credi bili ty and reputation aspects of the Brazilian economic policy between August 1994 a nd December 1998 using an "external circumstances" model.We emphasized the period between March 1995 and J anuary 12, 1999, in which exchange rate band regimes were in use in Brazil.
An "external circumstances" model, developed by Drazen and Masson (1994 ), 1 Masson (1995) and Masson and Agenor (1996), was used to analyze credibility and reputation factors in a fixed or pegged exchange rate system.T he model assumes t hat, in general, no government can conduct its economic policy with the single objective of inflation control , thoroughly ignoring t he employment and growth paths.T herefore, in t he presence of "external circumsta nces" (unexpected exogenous shocks) even a tough, anti-inflationary government can be forced to devalue its exchange rate.
We observed that the government built up reputation and credibility for its policy during the exchange rate regime implementation.Based on this theory, the government signaled its exchange rate commitment, causing unemployment to increase and competitiveness to recover gradually.A general conclusion is t hat t here is no evidence "based on t his model" to affirm t hat flawed economic fundaments led to t he exchange ra te regime colla pse.
T he art icle is divided into fi ve sections, including t his introduction: Section 2 consists of a theoret ical review of t he external circumstances model.Section 3 discusses the econometric procedure to be used and t he estimation of the model.Section 4 presents the results, taking into consideration the economic policy implemented at t hat t ime, while sect ion 5 presents a few conclusions and remarks.

The Theoretica l Model
Alt hough an exchange rate band was adopted by Brazil between March 1995 and early 1999, in practice it was not so different fr om a fixed, managed, or pegged exchange rate sys tem , as some analysts have pointed out.2T herefore, the analysis of exchange rate behavior is not based upon t he extensive literature on exchange rate ba nd originated by Krugman's model (1991). 3T he t heoretical approach used in this ar ticle complies wi th the credibility models tha i use economic policy games as their cornerstone. 4art of this li teratu re, namely the one produced by Backus and Driffill (19 5), has underscored t he role of the "type" of policymaker (weak, Lough, or innumerous ty pes) in determining t he cred ibili ty of a given economic policy.As far as this ma tter is concerned, a weak policy maker thin ks it is nice to mimic t he behavior of a tougher one in order to build his reputation.Therefore, a tough policymaker who wants to signal his type should, for instance, allow high increases in unemployment (a bove his des irable levels) right from t he start in order to improve his reputation and raise t he credibili ty for the implemented policy.
Later on, escape clause models were created.T hose models aimed a t structuring econom ic policy conditional on exogenous shocks or unpredicted contingencies .
In Lohmann (1992), for example, a policymaker delegates monetary policy to a conservative central banker who s hould establish a certain inflation rate in norma l circumstances, a nd implement a fl exible escape cla use in extreme situations.If he does not do t hat , he will be fired. 5ollowing in the same path, Drazen a nd Masson (1994) devised an "external circums tances" model.According to them, "Whether or not a n announced policy is carried out, however, often reflects more than t he policymaker's intentions.The situation in which he finds himself can be as important: even a "tough" policymaker ca nnot ignore the cost of very high unemployment, he may renege on an anti-inflation commitment in sufficiently adverse circumstances, that is, in times of weak activity, when pressures to restore high employment are strong.In short, t he cred ibility t he public assigns to an announced policy should t herefore reflect external circums ta nces as well" (Drazen & Masson, 1994:736).Masson (1995) made a mult i-period generalization of Drazen and Masson (1994), in which credibili ty evolves over time and t he updating of beliefs is explicit ly modeled.Masson and Agenor (1996) developed a slightly different version from t ha t of Masson's (1995), in which t he trade-off tha t a policymaker is faced up with occurs between inflation and economic competitiveness.Finally, Agenor and Masson (1999) also developed a slightly different version from that of Masson 's (1995) , in which the tradeoff that a policymaker is faced up with occurs between domestic interest rates a nd exchange rate stability.
In a formal manner, in Drazen and Masson's (1994) model,6 a stochas tic shock is applied to unemployment in Barro and Gordon's (19 3) model, in a way that the selection of economic policies by the government will depend on the realization of shock a nd also on the weight t he government attributes to inflation in relation to unemployment.Furt hermore, t here is uncertainty about t he policymaker's preferences a nd the shocks tha t impinge upon economy.We ta ke for granted that surprise inflation reduces unemployment rate Ut in relation to t he natural unemployment rate ii and that Ut is affected by its lagged 'Ut -l value and subject to stochastic 'f/t shocks: (1) where 7rt is the current inflation rate, 1rf the expected inflation ra te and o 2:: 0 is a measure of persistence for unemployment fluctuations (thus, !:::.= 0 fo is the autoregressive coefficient on unemployment, and if o = 0 there is no persistence) .
At the initia l stage of t he two-period model, we assume t hat uo -u = 0 a nd, t herefore, persistence only affects unemployment rate in period 2. 7T he government aims at minimizing an expected two-period quadra tic loss function (£ 1 ) .This L t fu nctio n in a single period depends on t he unemployment deviation from a target level below t he natural rate, • u-K (where K cap tures t he dis tortions that lead to an excessively high natural unemploymen t rate), and also to current inflation rate (2) where x represents t he weight attributed to infl ation s tability in relation to unemployment.T here may be difl' erent ty pes of government, which t ra nslates into uncertai nty a bout the gove rnmen t's objective function .T here are two types of government: tough government (wit h superscript T) , which cares a bout infla tion with the weight xT, and weak government (with s uperscript W ), which assigns a lower weight (xw ) to inflation in its objective function, therefore xT > xw.
In period 1, the type of government is unknown, so the government 's choice of policy in period 1 may influence the expectation of a devalua tion in period 2. Analyzing only two periods, and assumi ng t hat exchange rate (in log) et is t he political tool used to influence price level, and t ha t t he price level is equiva lent to t he excha nge rate,8 t he i-type government 's objective function cond itioned on information availa ble at t = 1 may be defined as where 6.e 1 is the exchange rate change.To simplify matters we defined K = J(j ..j(i and Et-'f/t/ ..j(i. T he val ues of t he objective funct ion in (3) depend on the government 's a bility to offset shocks in periods 1 and 2, and a lso t he signaling of its type t hrough its act ions in period 1, provided that the unemploym ent rate in period 2 is dependent on exchange rate expectations .T he t iming of events is as follows: private agents esta blish their exp ectations as to t he exchange rates and set wages before shock Et is carried out; t he policymaker observes t he shock on unemployment, and then he determines t he exchange rate (devaluing it or keeping it fixed).
T he question to be analyzed is how the probability of a devaluation in period 2, denoted /12, depends on government 's action in period 1, since not only the standard signaling of an unknown type of government but also the effect of unemployment persistence are considered.Taking into consideration the uncertainty about the types, we can write 112 as (4) where: w2 = proba bility the government is of type W ; p}]' = probability a government of type W will devalue (given the distribution of E2); pfprobability a government of type Twill devalue; argument j (= D or F) indicates t he government's action or choice, whether it devalued t he exchange rate (D) or it kept t he exchange rate fixed (F) in period 1; t he calcula tion of t hese probabilit ies depends on the information structure.
To calculate p~, we should first solve t he government's problem in period 2, for given expectations of a devaluation 112(j).After t hat, subst ituting p~ in equation ( 4), we obtain 11 2 (j).It is assumed t ha t private agents know the values xT and xw, but do not know about t he type of government in addit ion to not observing shock Et.9 Denoting by L;v(j) (where j was the action in period 1) the single period loss function of an i-type government if it devalues in period 2, t hen the government will devalue in period 2 if L;v (j) -L;F (j) < 0, that is, if the expected loss with devaluation for period 2 is less t han the expected loss res ul ting from t he maintenance of a fi xed excha nge rate in p eriod 2. T his defines a critical value for shock E;(j): (5) 9 Drazen and Masson (1993) also analyze t he case in which t he shock is obser vable or may be inferred.
where s is t he fi xed devaluation size.Therefore, critical value E;(j) depends on the type of government (via x i) and on the previously observed policy.If the realization of E2 is below this critical value, the policy of maintaini ng t he exchange rate fixed is optima l; if it is above it due to external shock, a devalua tion is optimal.Assuming tha t t he distribution of Et is uniform between -v and +v, and symmetric around 0 (for an interior solut ion), we have: (6) To calculate t he probability of government type, we assume t hat private agents use the Bayesian approach to assess t he type of government, starting wit h uniform priors over the two types.Expectat ions are conditio ned on whether or not the government will devalue in period 1, but not on t he rea lization of shock E 1, since it is assumed that private agents do not observe it.T he probability that the govern ment is weak condi tional on its act ions in period 1 may then be written as when uniform priors are used as a star t.Observe t ha t w2(D) > w2(F) as long as pf > pf, t hat is, as long as t he probability t hat a weak government devalues in period 1 is greater t han t he probability t hat a toug h government deva lues in period 1.
T he probability that a given typ e would devalue in period 1, p\, is derived in an ana logous way to t he previous calculat ion for p~.We calculate a cr itical value of the shock in period 1, t ha t is, E~, in a way t hat Ai (D) = Ai (F) .l 0Consequently, the proba bili ty that E 1 > Ei 1 , may be calcula ted ass uming t he same uniform dist ribut ion.
T he last step is to establis h t he rela tions hi p between t he credibility of the no-devaluation commitment in period 2 a nd t he policy act ion observed in period 1.This allows us to show that no-devaluation in period 1 may raise instead of reducing priva te agents' expectations of a devaluation in period 2. For t hat reason, we have to derive t he difference in the probabilities of a devaluation in period 2 as a function of t he policy action taken in period 1, that is, /-L2(D) -/-L2(F).To calculate J.t2(D) -/-L2(F ), we have to combine equations ( 5), ( 6) and ( 7) so that, after some algebraic manipulation, we obtain: where we used 1-L t(D) -~-t 1 (F) = -.jas.ote t he sign in expression (8).The first part on the right-hand side is always posit ive, as we assume that 1 -sj2v > 0, because t he sole objective of devaluation is to offset t he shock; thus, devalua tion size should not exceed twice the maximum size of t he shock.
T he unemployment persistence parameter 8 will affect both terms inside the brackets.The effect on p~ ('i = T, W ) arises because critical level E~ of the shock in period 1 depends on the welfare in both periods and, as a consequence, on 8. T here are two possibilities.F irst, in the case of no pers istence of t he efl"ects of unemployment between periods ( 8 -0), as only one signaling effect exists, the fi rst term disappears from within the brackets, and the expression in ( ) is certainly posit ive.T he standard result on t he sig naling of types will t hen be in effect: if a tough policy was observed (no-devaluation) in period 1, the probability for no-devaluation will increase in period 2. Usi ng equation ( 8) , we observe t hat 1-t2(D) > M(F) as long as Ptv > pf, that is, as long as the probability that a. weak government would devalue in period 1 is greater than tha t of a tough government, which is t rue from the moment we assu med t hat xT > xw .T his wa:y, t he a bsence of persistence and the existence of different preferences over inflation imply that the signaling motive alone contributes to the credibility for fixed exchange rate, which shows improvement in p eriod 2, if no devaluation is observed in period 1. T he second possibility considers t here exists persistence in unemployment, thus 8 > 0. The dependence of f.1.2(D)-f.1.2(F)on 8 is quite complex, reflecting the cont ribution of both terms.Drazen and Masson (1994) , solving equation (8) t hrough numeric methods, showed that for sufficiently high 8 values, t he persistence effect tends to domina te t he signaling effect and (8) will become negative.Therefore, t he probability for devaluation in period 2 will be increased , which implies that the government 's credibility will be reduced.Based on t his Drazen & rvi asson (1994:742-3) concluded as follows: "To summarize, pos itive pers istence of unemployment implies that no devaluation in t he first period may raise rather than lower the public's expectation of a. devaluation in t he second period .Shocks t hat are not offset t hrough a devaluation in period 1 have further unfavorable efl:' ects in period 2, increasing the probability that a government of either type will devalue.If these persistence effects are sufficiently s trong (8 large) , not devalua ting in t he first period will raise the probabili ty of a devaluation in t he second .Thus, credibili ty will not necessarily be enhanced by 'playing tough' in period 1." As previously mentioned , this is just a two-period model, a nd signa ling is carried out in period 1. Masson (1995) made a mult iperiod generalization of Drazen a nd tvlasson's model (1994), in which credibility evolves over time a nd t he updating of beliefs is explicitly modeled.In a formal ma nner, for simplicity, the model is written in t erms of • ur 1 , the deviation from t he n atural unemployment rate (tha t is, using the previous nota tion, ur 1 = u 1 -u ): (9) where e is t he log of exchange rat e; Et-t e t , t he expected exch ange ra te, and 'IJt an exogenous s hock on unemployment.
It is believed t hat the government minimizes a loss funct ion of a single period dependent on the quadratic de viations of unemployment from na tw•al ra tes a nd on the (squared) change in the exchange rate: l1 (10) T here are two types of government: a tough government ( with weigh t xT) , which attributes a higher weight to exch ange rate stability than a weak government (weight xw) , t hus XT > xw.P rivate agents know about these values but t hey do not know which of them is used by the government, a nd , as a consequence, they crea te proba bility assessments of the type of government.T hey specifically upda te t heir assessments of the probability tha t a government is of a weak ty pe, w 1 , based on previously presented beha vior.
Again, the government devalues the exchange rate wh en shock 1Jt is so strong that the costs of keeping the exchange rate at a fixed level are high er than the costs of an increased inAation.If L P is the value for t he loss function when t he excha nge rate is kept at a fi xed level, and L 0 when it is devalued by a given s size, then t he government devalues when L 0 < LF.T his implies t hat the government will devalue if a nd only if pf is defined as the probab ili ty tha t a weak government will devalue in period t and Pt as the proba bili ty tha t a tough government will devalue.As Wt is the assess ment by t he private sector of the proba bility t hat a government would be weak, we can write the expected devaluation ra te as where Pt is the ex pected devalua tion probabi li ty and (12) .
' xis where Assuming that 1Jt has a uniform distribution withi n [-v,v] and t hat there is an interior solut ion, we have . .
We can then solve equa tion (13) for Pr' and pf, and calculate in equation ( 12): = In equation ( 15), we can separat e the t ime-varying part from the time-independent part, and also decompose t he second term into steady-state devaluation probabilities, pw and pT.Assuming that private agents know about the type of government, we can write a realignment probabili ty in period t j ust as 12 ote t ha t, for a given assessment of type (weak or tough), hig her unemployment rates increase t he expected devaluation in the next period, since they enhance the probabili ty that a positive shock on unemployment will push it into the zone where devaluing is more interesting t han keeping the exchange rate fixed .
T he probability estimates t ha t t he government is of type W or Tare ini t ia lly calculated t hroug h a prior esti mate w 1 _ 1 , supposing t hat t he government has not devalued in period l -1.
Substituting the values in equation ( 3) into W and T , aligning and adding an error term ~t.we obtain:13 where a and {3 are parameters to be estimated.It is expected t hat {3 ~ 0, t hat is, higher unemployment rates will reduce the assessment of t he probability t hat a government is of t he wea k type.In other wo rds, t he disposi t ion to accept increases in unemployment without resorting to devalua tion enhances t he govern ment 's reputation for toughness.
In econometric terms, t he model is estimated using the Kalman filter, which consists of a transition equation given by equation ( 1) and of a measurement equat ion given by equa tion (1.6), which, under some assumptions, is simpli fied as (1.9)where a (t error term is added.It is expected that a t > 0 and 1 > 0. T he unobservable and time-varying parameter is t he probability t ha t the government is of the weak type, Wt• It is expected t hat t he lower the value for the parameter is, the lower its reputation for weakness will be, and, therefore, the higher the fixed exchange rate policy credibility will be.T his way, higher unemployment reduces the probability that t he government is of the weak type through t he transition equation (sig naling efl' ect); t hus, a future devalua t ion proba bility increases t hrough t he measurement equat ion, in t he event of sufficiently strong exogenous shocks (external circumstances effect).Masson (1995) applied this model to analyze the credibility of G rea t B ri tain's exchange rate system during the excha nge rate mechanism (ERM) period.
In Drazen and Masson (1994) and Masson (1995) unemployment is used to signal the policymaker 's toughness or weakness, and the shocks increase the probability that the fixed exchange rate system will be discont inued .Masson and Agenor (1996) present a version that is simila r to Masson's (1995) model, with the main difference that in those models t here is a t radeoff between inflation control on t he one hand, and t he maintenance of a reasonable level of competit iveness on the other hand, in order to maintain t he economic activity, avoiding t he deterioration of foreign accounts . 14T herefore, the concerns a re not related to unemployment, but to the loss of foreign competitiveness and its effect on foreign reserves.Shocks on domestic prices also tend to increase inAation and reduce competit iveness.
Formally, in Masson and Agenor's (1996) model, iL is ass umed that t here is a loss funcLion guiding t he policymaker's acLions in t he face of ex ternal or domestic shocks.The policymaker has two choices: no devaluation (or previous announcement of a path for excha nge rate devaluation) or devaluation (relatively to the previous announcement of a pa th for exchange ra te devaluation).The policymaker specifically attempts to minimize the following loss function of a single Lt period: (20) where 1rt is t he infla tion ra te 15 (in terms of consumer price index, and defined as 1rt = Pt-Pt 1, where Pt is the price level log); Ct is t he level of competitiveness (defined as Ct = et +Pi -pf ,16 where et is t he nominal excha nge rate, Pt is t he level of foreign prices, a nd pf is the domestic product price log), a nd x is t he weight attributed to inflation by the policymaker.T here are two types of policymakers: XT, if t he policymaker is of the tough type, and XW, if the policymaker is of t he weak Lype, therefore xT > xw.
Once again devaluat ion expectations depend on the probabilities that the policymaker will be weak or tough, as well as the ex ante probabilities t hat a type of policymaker will decide to devalue in the event of a shock on domestic inflation.Following a process that is identical to Masson's (1995), Masson & Agenor (1996 : -11) show that the expected devaluation rate can be written as (21) where at > 0, a2 < 0, a 3 < 0 and 'Ut is a wh ite noise error term.Therefore, devaluation expecta tions depend directly on t he probability that t he policymaker will be weak, (wt) , and, inversely on foreign inflation (6.pt) , and also on the level of competitiveness (ct) -Similarly, Masson & Agenor (1996:11) derive t he following updating equation:17 (22) where 0 < bt < 1, bz > 0, b3 > 0 and Vt is a process of white noise error.
Observe t hat, although foreign inflation and competit iveness have negative signs in equation ( 21), t he opposite occurs wit h t heir lagged values in equation ( 22) .T his happens because t he disposition to accept a competitiveness loss and lower foreign infla tion without deva luation means that p olicymakers are less likely to be weak; consequently leading to a lower 'Wt value (via equation ( 21)), tha t, in its turn, implies an increase in cred ibility.However, greater competitiveness losses in t he face of external shocks increase t he probabili ty of devaluation (via equation ( 22)) .Thus , t he s igna ling and externa l circumstances effects reappear .T he model is estima ted by means of a Kalman filter, in which equa tion (2 1) represents the measurement equation and equa tion ( 22) is t he transition equa tion.
In short, these external circumstances models point out two aspects of credibility: first, t he signaling of policymaker's type and, second, t he probabili ty for any ty pe (weak or tough) that devaluation will occur if circumstances are sufficiently adverse, since no devaluation wo uld be incons istent with the government's goals.T hus, high unemployment or compet iti veness loss signal a toug h government, but also increase t he probability t hat even a tough policymaker will devalue.

Econometric Procedure
T he model implies t hat the correlation between the changes in unemployment rates and the devaluation expectation will be very different, depending on the dominant factor: the signaling factor or the "external circumsta nces" factor.If, ini tially, there is much uncertainty about the type of government, t hen, high unemployment ra tes may convincingly signal that the government is toug h and determined to carry out its announced policy; consequently, policy credibili ty would be improved.On t he other hand, if t he type of government is known (d ue to its historical record or because it has convincingly signaled its type) or if t he types are very much a like, and t here are negative external shocks, high unemployment rates reduce credibili ty, since one or other type is more likely to relinquish the policy commitment.T he situation of competit iveness L oss is quite similar.
Alt hough Masson's (1995) a nd Masson and Agenor's (1996) models are theoret.icallymore appropriate once the evolution of credibility is explicitly modeled and estimated through the Kalman filter, they are not directly used on account of their econometric limi tations.T he aut hors t hemselves admit t here are two maj or limitations: firstly, for estimating the models, a linear approximation is used for the updating equation (which, indeed, is highly nonlinear); secondly, t he measurement equation may produce W t va lues for the estima tion t hat are not found wit hin t he [0, 1] interval, or t he equation's right-hand-side variables take extreme values or the realizations of ~t (which a re ass umed to have a Gaussia n distribution) are extremely large. 18In addition, it is necessary t hat excessive restrictions be imposed on the model so that consistent estimates can be obtained .Even so, it is hard to obtain significant parameters with the esti mations.
T his way, t he procedure used is more simila r to that of Drazen and Masson's (1994), who adopted two procedures to analyze the changes in t he relationship between unemployment and deva luat ion expectations: first, t hey used d ummy variables t hat correspond to t he historical periods in which the change s upposedly occurred, and then they t ested its significance; 19 second, and more econometrically appropriate, they d id not take t he breakpoints in t he relationship for granted and performed tests of structural stabili ty, spli tting t he whole sample into two sub periods, by successively t rying different breakpoints.If breaks are significant at severa l dates, t he one that g ives the maximum li kelihood ratio value is chosen; after t ha t, each of the s ubsamples is fu rther tested for breakpoints in a similar way.In both cases, t he coefficients of the unemployment rat e and the constant term are expected to vary between t he subsamples.
Alt hough dummy variables or tests of structural stability may be used to analyze t he changes in a relationship, this is not the best a lternative.As we expect t he parameter that measures the relationship to vary over t ime, the adequate stra tegy is to use techniques wit h time-varyi ng parameters.Thus, we will use t he Kalman fil ter for t he estimation. 20 Different ly from Drazen and Masson (1994), the competi t iveness variable is explicit ly included in the model's estimation.An interest matter here is to check whether competit iveness, measured t hro ugh the rea l exchange rate, has eA'ects on t he trade balance.P astore et a l.(1998) showed the sensit ivity of the trade bala nce to real exchange ra te by using equations that explai n the behavior of imports and exports .Consequently, the model to be estimat ed is 18 T he same would happen if t he general Kal man fi lter were used to esti mate t he nonlinear form (Masson, 1995:577).
19 T his was t he method used by Arbex and l.<' ontes (1999) in t heir application of Drazen e Masson's (1994) model to analyze t he monetary policy credibility between 199 1 and 1998 .T hey concluded t hal Brazilian monetary policy was nol credible from January 199 1 to December 1992 and li.•om J uly 1996 to J uly 1998 and was credible for t he period June 1994 to J une 1996.The rest of t he sample period it was not possible to get a clear cut a nswer. 20See Port ugal (1993) for the analysis of Kalman fil ter s uperiority.See Ha rvey (1989), C ut hberson et al. (1992) and Hamilton (1994) for t he presentation or Kalman illter.
(23) f3t = f3t-l + J.l,t and At= At-1 + Vt (23') where dift is the expectation of exchange rate devaluation for the period; Ut, is the current unemployment rate in terms of natural unemployment rate deviation; Ct, is the measurement of economic competitiveness in terms of deviations from a competitiveness target level, and ~t is a white noise error term.Equation ( 23) is the measurement equation and equations in (23') are the transition equations, which show that parameters f3t and At arc time-varying.
Considering a situation in which there is uncertainty about the actual type of government, if the government wants to show it is tough, thus increasing its credibility, we expect the coefficient on unemployment to be negative, as the government will be signaling its commitment to a fixed exchange rate (and, consequently, to low inflation).Therefore, the expectation of exchange rate devaluation should decrease, disregarding any shocks, and, as a result, policy credibility should increase.If the government possesses or has built a reputation for toughness and, therefore, has created relatively firm credibility for its policy, in a context in which strong exogenous shocks take place, higher unemployment rates may increase devaluation expectations.This way, a positive relationship is expected, indicating reduced credibility to exchange rate commitment.
A similar relationship is expected in the case of competitiveness.At the beginning, the policymaker t ries to signal his commitment to inflation control by maintaining the exchange rate at a fixed level, accepting competit iveness loss, and thus showing how tough he is .However, after some time, and in the event of external shocks, greater competitiveness losses lead to reduced credibility to fixed exchange rate commitment.Although competitiveness losses may be associated with an increase in unemployment, the major concern is the deficit infringed upon the t rade balance and the reduction of foreign reserves.
Note that this separation is important due to the fact that the government may continue to signal its anti-inflation commitment in the face of external shocks, disregarding unemployment but showing concern with competitiveness, for instance, slowly devaluing the real exchange rate.The next step is to estimate the variables that will be inserted into the model.Interest variables are devaluation expectations, deviation of unemployment rate from the natural rate, and deviation of competitiveness from a competit iveness target level.These estimations are shown next.

Estimation of deva luation expectations
Devaluation expectations are measured through the difference between the overnight Selic interest rates applied by the Brazilian Central Bank and the prime interest rates, even though this is a flawed measurement of exchange rate expectations since it does not take into account the risk premium and other factors that may influence exchange rate expectations. 21  The time path for the interest rate differential is shown in figure 1.We can sec there is a tendency towards a reduction of such differential over time, interrupted by the Mexican crisis (December 1994), the Asian crisis (October 1997), and the Russian moratorium (August 1998) .In other words, external shocks that reversed the capital flow and forced economic authorities to increase the internal interest rate.After a high increase following the crises, there was again a gradual reduction in the interest rate differential.
21 This method for measuring exchange rate expectations is criticized as it calls for t he application of uncovered inLeresL rate parity.One of Lhe criticisms is LhaL Lhere is a peso problem when testing the hypothesis (for example, Sachsida ct al. (1999)).However, this pro blem docs not imply the rejection of the hypothesis (sec a discussion on this topic in Agcnor and :Ylasson ('1999), chapter 6); in addition, there is no genera] agreement on the proposed alternatives, which have their own limitations.Another way of measuring exchange rate expectations may be seen, for instance, in :Yli,inch (1998).

Estimation of the deviation of unemployment ra te from t he natura l rate or NAIRU
The concept of natural rate of unemployment or the Non-Accelerating Inflation Rate of Unemployment (NAIRU) is part of the monetary policy structure in theory and in practice as well.It is essential that we know whether the current unemployment rate is below or above the natural rate or N AIRU to conduct monetary policy.The problem is that they are not directly observable, and thus need to be estimated.Frequently, the natural rate of unemployment is simply approximated through the unemployment series mean during the study period, or by calculating its tendency without considering its relationship with inflation.The NAIRU is an interesting concept due to the fact that its estimation explicitly takes into account its relationship with inflation.In this case, a Phillips curve or some of its versions is estimated.
Although Phillips' original article suggested that the Phillips curve was actually a curve, it has been usually estimated as if it were linear (or only presenting little discreet changes).Even though some other studies regard the Phillips curve as linear, they have allowed the NAIRU to vary over time . 22owever, some authors have confronted this viewpoint.For example, Clark and Laxton (1997), Debelle and Laxton (1997) and Debelle and Laxton (1997) have estimated nonlinear and convex Phillips curves in which the NAIRUs vary over t ime. 23Clark and L~'<ton (1997) derive an equation for the nonlinear and convex Phillips curve using a price and wage determination model.
Usually, in the case of a linear Phillips curve, the estimated models are presented as follows: where 1ft is the inflation rate; 1ri is the expected inflation rate; Ut is the current unemployment rate; N AI RUt, in this case, the time-varying non-accelerating inflation rate of unemployment, and Et, a white noise disturbance term .In the linear case, although the N AIRU varies over time, there is no difference between the N AIRU and the natural rate of unemployment, therefore, if • u; is the natural rate of unemployment, then 7.t; = N AI RUt• In the case of the nonlinear and convex Phillips curve, the model to be estimated is presented as follmvs: This functional form docs not impose a very steep convexity since we assume the lower unemployment limit to be zero.This formulation, however, allows us to perceive the differences between the linear and nonlinear forms.A convex Philips curve implies that a certain reduction in current unemployment below the natural rate triggers a bigger increase in inflation t han the reduction of inflation that would be caused by an increase in unemployment above the natural rate.In other words, as excess demands are more inflationary than excess supply are deflationary, allowing the economy to enter an excess demand zone implies that the economy will have to operate for a longer time period in the excess supply and high unemployment zone in order to prevent inflation from accelerating. 24 Another implication of this specification is the difference between the natural rate of unemployment and the NAIRU .As previously shown, when the model is linear, the natural rate of unemployment equals the NAIRU, but when the model is nonlinear, the rates arc different.Dcbclle and Laxton (1997) have shown that, in the nonlinear and convex model, 25 the natural rate of unemployment (u;) equals the N AI RUt + .Xva.r( Ut) / 2. Consequently, aggregate demand or labor policies that reduce (increase) unemployment variance, reduce (increase) t he natural rate of unemployment .
In terms of economic policy, nonlinearity has several implications: • stabilization policies t hat do not succeed in reducing the variability of the business cycle may have undesirable effects not only on unemployment variance but also on the natural rate of unemployment; • it highlights the necessity for forward-looking policies that act preventively towards offsetting inflationary pressures; 24 For a numeric example, ta.ke two periods, and suppose the inflation rate intended by economic authorities is 3% and that, at t his rate of inflation, unemployment is within t he N AIRU.T herefore, for instance, in period 1, a given reduction in unemployment may increase the inflat ion rate by 4%; if economic a uLhorities inLend Lhe average inflaLion lo be equal Lo 3%, inflaLion in period should be reduced to 2%; however, to obtain such a reduction in inflation, the increase in unemployment must be higher (in terms of absolute value) than the reduction obtained in period 1.The graphic examples can be seen in Debelle and Laxton (1997:254-6) and Clark and Laxton (1997:17). 25Assuming that, for simplicity, t.he nonlinear, convex Phillips curve is 7rt = 1ri + exp['YN AIRU, .-'Ut) ]l + <;t .
• it suggests t hat strong recessions may only have an anti-infla tion impact that is narrowly higher than weaker recessions; • it emphasizes t hat policymakers need to act cautious ly, especia lly if the economy is close t o its potent ial (Debelle and Laxton,1!197:256), and (Debelle and Vickrey, 1997:24 -6).
In t he case of Brazil , recent estimates of t he linear, ti me-varying • 1 AIRU based upon t he estimation of a P hillips curve have been carried out by Portugal et al . (1999) and Port ugal and Madalozzo (2000), having produced consistent results.However, t heir estimates of the NA IRU using only the unemployment rate and estimating its structural component (which is identified as the NAIR.U) have not shown good results.T his shows t he need to estimate a N A T R U in rela tion to t he infl ation rate.
Following Clark and Laxton (19!:17), Debelle and Laxton (1997), and Debelle and Vickery (1997), we estima te equation ( 25) , a llowing the NAIRU to vary over t ime.To estima te this equation, it is necessary to estima te t he N AI RUt that is not directly observable.With this objective, we use the Kalman filter, which estimates t he models as presented next: where vector f3t is varia ble over time a nd determined by the transit ion equa tion (26') .To estimate equation ( 25) , we define Yt = 1rt , Xt = (7rt-L, 1/ut, 1) a nd f3t = (8, -yN A I RUt, --y).We assume that 8 and -y are constant and t ha t only the -yN AI RUt varies over t ime.T he r AIRU estima tes at each time period ar e calculated t hrough t he negative value of t he ratio between t he second and t hird element of f3t • T he model was estimated by means of monthly seasonally adj us ted data on the current unemployment ra te obta ined from lpeadata and the inflation rate measured by lGP -Dl (general price index), obtai ned from FG VDados.As the • 1 AIR.U is expected to vary in t he long run, estimation encompassed t he period between J anuary 19 2 and December 199 .To estimate t he model, t he seasonally adj usted infla tion rate was used, wi th the same fil Ler a pplied Lo the unemployment series. 26 he use of the Kalman fi lter produces two AIRU time series.T he first, called filtered time series, originates from the model's recursive estimation, which uses data that are only available until t he current period.At each t ime period, the filter uses t he new information to review its estimates of t he model's parameters and of the NAIRU.T his exercise mimics, to some extent, the process that a policymaker wo uld carry out when using this a pproach to determine infla tionary pressures.The second t ime series, called smoothed , uses the data from t he whole (full) sample to estimate t he • 1 AIRU time series .T his way, we are able to assess, in a retrospective manner, whether recursive estimates produce a different path for in flationary pressure level when full-sample estimates, which incorporate more information, are used (Debelle and Laxton, 1997:268-9).T he time p::.,th.for the nnemployment r::.,te ::.nci filtereci AIRU e. t ;im::.tecitime series (denoted ' AIRUCC) a nd s moothed 1 AIRU (denoted NAIRUCCSM) are shown in fig ure 2. T he results are cons istent wit h historical evidence.In t he 1980s and mid-1990s, a period wi th high and growing inflat ion , the unemployment rate was almost always below filtered and smoothed NAIRU.After t he Real P lan, in comparison with t he fi ltered 1 AIRU, t he unemployment rate was almost always below it.Using the smoothed estimation, the unemployment ra te was ini tia lly above the I A IRU a nd below it in pa rt of 1995 a nd early 1996 .Later, the unemployment rate was always above the AIRU .This is consistent wit h t he price stabilization occurring aL LhaL period.V.,Te can also observe L haL, ruLer the opening of the economy in 1990, the unemployment and TAIRU rates increased at first and then declined .After the Rea l Plan , there was a reduction in t he unemployment rate; however, some time after t hat, t his rate, as well as the AIRU, cl imbed remarkably.
Even though it is not possible to directly compare the AIRU estimated in t his section wit h t he linea r tim e-v~ryin g ATRU estimated by Portugal ~nd r~.dalozzo (2000), ue~;au~e th~e au thor~ u~e<lttuartedy anc.l ~easonally unadjusted <lata, L>oth estimations arc cons istent with historical experience, and estimated paths usually followed a similar pattern.However, a n interesting difference is t hat t he AIRU estimoted by t hese outhors was overly vo riable, which is not very consistent with t he concept of AIRU as a long-run variable, expected to vary, but not so fast.T he smoothed AIRU est imated in t his section had a very smooth behavior, as t he name itself suggests.
A~ the ~timation period of the external circLUnstan~;es model goes from A ugu~t 1994 to December 199 , only t he data tha t correspond to t his period arc used for later analysis.T he devia tions of the current easonally adjusted unemployment rate are used in relation to the fil tered AIRU, which mimics t he behavior of a policymaker whose aim is to reduce infla tion .

Estimation of the deviation of competitiveness from a t arget level
T here has been a lot of debate on the best way to measure competitiveness.Even if we decide tha t we should use~.rea l exch ~nge mte, t here will be discussion on whi~;h the L>~t rate ~hould be. 27Sin~;e the model aSSLUnes that the poli~;ymaker is concerned with t he external sector, we opted for a general indicator, which includes not only the US but al o t he mo t important partners.Therefore, in this article, the competitiveness level is measured by the real effective exchange rate of exports calcula ted by Ipeadata.Again, what matters is the behavior related to a path for or a LargeL L evel of competitiveness, which may be identified with a real equilibrium exchange rate (calculated by several methods) .
In this subsection, two methods were initially used.T he first method was developed by Goldfajn and Valdes {1999), who calculate a series of overestimation or tmderestimation as deviations of the exchange rate from a filtered series using a Rod rick and Prescott fi lter (Rodrick and Prescott, 1997), where the filtered series captures the series' stochastic tendency, which is identified as t he predicted real equilibri um exchange rate.onalignments are identified as t he series' cyclic component, since t hey are exp ected to occasionally correct themselves.The second l' lll<l simplP-st metho<l SltpposP.sthat the rP-l' ll eqn ilil>riu1n exehl'l.nge rl'l.t,P. is given hy the base-year or by the index mean of the real exchange rat e during the study period. 28Since t he estimates made with the H.odrick-Prescott fil ter29 produced less consistent results, the results are presented through the deviations of t he log real exchange rate in relation to the series mean during that time period.
T he inspection of figure 3, which presents the Lime path for the real excha nge rate, shows that it is ha rd to defi ne a s ingle level for the real equilibrium excha nge rate.We can observe that t here is a strong apprecia tion of the real exchange rate, which is in accordance with the economic openness that occurred in the 1990s and, also, a s trong a ppreciation a fter the implementa tion of t he Real P lan, especially in t he period of free floating exchange ra te, in which the real exchange rate achieved a remarkable a ppreciation between July and ovember 1994, remaining constant nearly up to February 1995.After the Mexican crisis a nd t he implementation of t he exchange rate ba nd regime in March 1995, there was a tendency towards t he recovery of t he real excha nge rate, which accumulated a depreciation of nearly 10% 11fitil .l11nP.199fl, t-"~lthon gh rP-vP.rsed to some extent, <l11ring 1997.After .ll' l ,nllt-'lry 199 , there was again a gradual recovery of the real exchange rate.
Considering t hat the estimation period in this article started in August 1994, we can observe that the real exchange rate was much appreciated at that moment.Most analysts agree wit h that, despite using several methods to measure t he real exchange rate.30Depending on t he method used, a fter this strong, initial a ppreciation, there was a depreciation of the real exchange rate or the level of appreciation was maintained .T he discussion a t t he t ime dwelt upon whether and how much the appreciation implied a overvaluation in relation to a certain equilibrium value.
• rever theless, we can affirm that mos t economic agents expected no additional appreciation, but in fact a recovery of competitiveness.T his fact, which will be discussed later on, is important for analyzing the resul ts of t he model's estimation.As previously observed, these models were developed to analyze fi xed or pegged exchange rate regimes .Alt hough the exchange ra te regime band formally began in March 1995, the model was estimated through mont hly data on the variables f>.~tiffil'll.~cl in th~ pr~viOII S snhs~etion for the r~rio<l h~l.w~enAngus t 1994 ;:mel D~ cember 1998 (exactly one month before the exchange ra te regime was abolished) .T he inclusion of information abou t the previous months aims a.t determining the type of government before the implementation of the new regime.The estimation resul ts of model ( 23) , by means of t he Kalman filter, were the following (t-statistics is fo und within parenthe!:ie!:i):As we can observe in the t-statistics, the coefficient on the unemployment variable is t ime-varying, while the coefficient on t he competiti veness variable is a lmost cons tant.Figure 4 shows the time paths for the estimated filtered coefficient (Beta) and the smoothed coefficient (BetaSM) on the unemployment rate, observing t hat this is a time-va rying coefficient, almost always negative throughout t he period.T here is sign reversion in part of 1997, period t hat coincides with the reduction in the unemployment rate, signa ling, according to the model, a weaker exchange rate commitment.In addit ion , in this period , unemployment was below the filtered I AIRU, indicating inflationary pressm e and reduced credibility. 31ater , the unemployment rate became negative again, a tendency only reversed during the external shock periods, Asian and R ussian crises.In these two cases, sign reversion was due to the increase in interest rate d ifferential , which originated from the increase in the internal interest rate as a. response to reversed capital iiLAows, fln<l no t hec:fl,use of unemployment rfl.te negfl.tive< levifltions, flS t hey hfl,d always been positive from early 1997 onwards.

z:
F igure 4 Unemployment variable coefficient( 1994-98) co "' "' oc ;;;: "' ..j. ., ., .;., .;., .J, .... .... .... ,.!. ,.!. ,.!. ,.!.T he time paths for t he estimated filtered (La1nbda) a nd smoothed (Lmnbda SM ) coefficient on competitiveness are s hown in figure 5, which is practica lly constant (it has a final estimated value of -4.34) .T he filtered est imates are consistently negative t hroughout the estimation period.In add it ion , the smoothed parameter estimates are practically cons tant.T his indicates t ha t the slow recovery of competi tiveness reduced t he interest rate differentia l, conferring higher credibility to the government.As mentioned, in August 1994 there was a common agreement that the real exchange rate was overvalued and that such overvaluation should be reversed.T herefore, the agents did not expect higher competi t iveness losses to get to know t he government level of commitment, as these losses would reduce t he government's reputation and its p olicy credibility.Consequently, the gradual recovery of competitiveness would reduce the interest rate differential and was consistent wit h the economic policy adopted .

Credibility and reputation during the exchange rate regime
T he objective of t his section is to analyze whether t he a bandonment of t he exchange rate regime can be explained by economic fundamentals, or, more specifically, to analyze whether the government has los t its reputation and whether the policy losL its creu iuili ty b efore the excha nge ra.Le regime was lifteu .IL is irnportant to stress t hat t he interest is not on t he t iming of speculati ve at tacks or on t he exchange rate crisis.T herefore, the literat ure on sp eculat ive at tacks and exchange rate crises is not t he ma in topic here. 32 he externa l circums tances model presented in the theoretical section aims at explaining t he reputation and credibility of a fixed or pegged exchange rate regime.The basic idea is that by allowing for an increase in unemployment and competitiveness losses, t he government's reputation and its policy credi bility are enhanced.However, a fter external shocks, further increases in unemployment and greater competitiveness losses reduce the government 's reputation and its policy credi bi I i ty.
After several unsuccessful attempts to stabilize the economy during the 19 Os and early 1990s, the Real P lan has fin ally managed to reduce inflation systematically and consistently.At t he beginning, a monetary anchor was announced, but the idea was soon cast aside, and a n exchange rate a nchor was t hen adopted.
Initially, economic authori ties allowed the real to be considerably overvalued, as reducing inflation was a priority.Hence, between July 1994 and February 1995, there was a floa ting exchange rate policy that a llowed for such appreciation.After the Mexican crisis, from March 1995 onwards, an exchange rate band regime was implemented, which , as previously mentioned, resembled a. crawling peg regime.After the Asian crisis in October 1997, there was a reduction in foreign reserves, which was reversed a fter some mont hs, causing reserves to accumulate again.After the Russia n crisis in August 199 , extens ive foreign reserve losses, and the speculation agains t the rea l, t he government decided to abolish the exchange rate band regime at the beginning of Ja nuary 1999.
As previously mentioned , in t he second half of 1994, t here was a general agreement that the real exchange rate was very much appreciated.Economic aut hori t ies had already shown t heir commitment to price reduction by accepting the strong exchange ra te appreciation in the period between July and November 1994, when the Central Bank decided to intervene in t he exchange market in order to prevent an increased a ppreciation, although t here was no fixed exchange ra te regime yet.Aft er the Mexican crisis, in March 1995, an exchange rate band regi me was implemented, and t he exchange ra te began to recover, especially t hrough gradual increases in the nominal exchange rate.Therefore, it is hard to believe that economic agents were waiting for greater competitiveness losses to show t he government's commitment to inAa t ion control.It would be more coherent with t he model to await a grad ua l recovery of the real exchange rate, thus allowing economic competitiveness to be restored and, consequently, not jeopardizing the reduction in t he inflation rate .As expedeu, the results show that l:iU<:h l:iituation was really taking place.The recovery of competitiveness, or at least t he unchanging level of competitiveness, was fo llowed by t he reduction in the interest rate differential, conferring higher policy credibility and increased reputation to the government.
[n the case of unemployment , t here was ini tially a strong economic expansion and a reduction in t he unemployment rate, which may be explained by consumption boom models in stabilization plans with an exchange ra te anchor. 33T his is consistent wi th the positive value found a t the beginning for the model's coeffi-33 See, ror exa mple, Agenor and Masson (1999), cha pter L2. cient on unemployment.T hus, we ca nnot affirm that the positive sign indicates lack of policy credibility.Later on, the unem ployment rate, however, began to gradually increase, rising above the filtered AIRU from April 1995 to mid-1996, thus signaling a commitment to the exchange rate regime and better reputation to the government.After that, unemployment declined and, in the period between September 1996 and April 1997, a positive value was obtained again.T his was caused by the reduction in unemployment itself and also by the unemployment r:'l,te below the AIRU in m i<l-199() Mld e::lrly 1997.We h::l;ve, therefore, :'1, period in which the reputation and credibility of the exchange rate regime were reduced .After mid-1997, t he unemployment rate increased cont inually, rising a bove t he filtered NAIRU from the beginning of 1997, and growing even more at the end of 1997, signaling, again, the commitment to the exchange rate regime, and thus enhancing the govern ment's reputation and its policy cred ibili ty.T his was precisely t he period in which monthly inflation rates fell below 1%.However, after t he Russian crisis, t he strong speculation against t he real and the quick reduction in foreign reserves caused t he exchange rate regime to collapse at the beginning of 1999, making t he exchange rate oscillate, and forcing the government to implement inflation rate targets.
How can we therefore explain the abandonment of t he exchange rate policy at the beginn ing of 1999?As previously observed , the interest rate d ifferential, despite the reversions on its reduction path due to external shocks, continued to slowly decrease until some time before the exchange regime was discontinued.Apparently, the adopted policy was consistent with the prescriptions for the model, with the goal of gaining or maintaining credibility.Based on the estimated model, we cannot affirm that fundamental factors (unemployment and competitiveness) ~iguale<.lthe on~et of an unbearable situation.Unemployment was t:onsis teutly above t he NAIRU and competitiveness was gradua lly recovering, conferring credibility to the adopted policy.On top of the fundaments analyzed by the model, there were some other worrying factors, especially fiscal vulnerability and the consequent increase in the internal public debt/ GDP ratio, whose major part consisted of short-term financing.However, these factors are not full y blamable for the exchange rate collapse.At least, t here is no econometric evidence showing that .
T herefore, it is believed that, in great part, the abandonment of the Brazilian exchange rate regime was neither explained by flawed economic fundaments 34 nor predicted by economic agents. 35It might be better explained t hrough secondgenerat ion specu lation models, which have nothing to do wit h fundaments. 3GFor example, Baig and Goldfajn (2000) have shown that speculative at tacks were triggered by a contagious effect from the Russian crisis.However , this kind of explanation is beyond the scope of this article.
Also, we can always affirm tha t, a posteriori, the initially strong apprecia tion of the real was an error or tha t the recovery of competit iveness was too slow and should have been accelerated .Based on the results obtained, it seems that the government acted correctly.Although it is possible to affirm that the economy found itself in a stable situation, a strong devaluation or a faster devaluation would not bring infiaLion back to an explosive behavior. 37In t hat moment, it was risky to make a decision in that direction due to Brazil's inftationary history. 38We should also bear in mind that, due to the uncertainty over t he AIRU value, slackening the monetary policy and expecting not to have heavily inflationary effects was too risky.Maybe these facts help to explain why economic aut horities preferred to be cautious and delay as much as possible t he devaluation of the exchange rate or t he regime ma lleabili ty.

Conclusion
In this article, some basic credibility and rep utation models were analyzed so as to understand the behavior of exchange rate expectations during the exchange rate band regime implemented in Brazil.For t hat reason, a n "external circumstances" model was presented .This model assumes t hat, in general, no government should worry only about inflation control, totally ignoring the path for growth and job market.Therefore, in t he presence of external circumstances, even an ant i-inflationary government can be forced to deva lue the exchange rate.Hence, a government can signal its commitment to a fixed exchange rate regime, by causing increases in unemployment and competitiveness losses, enhancing its reputation as well as the credibili ty of t he exchange rate regime.T hus, in the event of external s hocks, furt her increases in unemployment or greater competitiveness 3 ~For example, Coldfaj n and Valdes (1998) a nd Berg and Pattillo (1999) have s hown t hat exchange crises a re not predictable.
36 See some references in footnote 31 of t his article. 37See, for example, Fiorencio and Moreira (1999a,b). 38T he fear of a n inflation backlash was so strong t hat, even after the excha nge rate was lifted, t.he possi bility for fi xing t he excha nge rate was cogitated , considering t hat.many ob.ervers forecasted very high inflation rates.See Fraga (2000) for the opinion of an economic a uthori ty.losses reduce the government's reputation and t he credibility of t he regi me, possibly result ing in t he abandonment of t he exchange rate regime.
We observed that the government built up reputation and credibility for its policy during the exchange ra te regime implementation.Based on this theory, the government signaled its exchange rate commitment, causing unemployment to increase and competiti veness to recover gradually.A general conclusion is that there is no evidence based on this model to affirm that flawed economic fundaments led tn the exc:hMlge r l' !l.e regime enlll'l,f>Se.
T his way, the first contribution of t his article was analyzing the credibility and reputation factors throughout t he implementation of t he exchange rate band regime.As previously shown, the economic policy was being conducted in a way that was consistent with the model.The government took on the risk of gradually recovering competiti veness, especially after the Mexican and As ia n crises .However, after the Russian crisis, which caused a strong and fast capita l outfl ow and successive attacks against the real, the government had to discontinue the exchange rate regime and ado pt a free floa ting rate of exchange.
A second cont ribution , yet only a by-product , was estimating a nonlinear, convex P hilli ps curve.An additional line of research may intensify the discussion about this Phillips curve type and, especially, insert credibility into a nonlinear, convex Ph illips curve model, and finally allow t he model to be estimated. 39However , this is still an area open for t heoretical and empirical studies.