Analyzing COPOM ’ s Decisions

This paper aims at modeling the conduct policy of the interest rate by the Monetary Policy Committee of the Central Bank of Brazil (COPOM), following methodologically a statistical framework developed by Engle & Russel (1998) and used by Hamilton & Jordà (2002) in studying the decisions of the Federal Reserve of the United States. The results, once we observe the period from January 2002 to July 2010 with a weekly frequency, suggest that COPOM is influenced by government expenditures and that it has a forward-looking behavior, holding onto the expectations of the Gross Domestic Product (GDP) and the official inflation of the country. In addition to this relevant evidence in the discussion of the balance between fiscal and monetary policies in Brazil, Brazilian monetary authority theoretically possesses a behavior which is aligned to that observed in practice, though it presents a greater sensitivity to macroeconomic variations.


INTRODUCTION
he Brazilian economy possessed strong inflationary levels during the military governments and even during its recent democracy in the 80s, when from 1994 it began to experience a period of economic stability with the adoption of the Plano Real, on which context it appeared propitious to the creation of the Monetary Policy Committee of the Central Bank of Brazil (COPOM) in 1996.
Following a worldwide trend towards greater transparency in the decision making process and the ease of communication with the general public, from the adoption of the target regime for the Inflation in Brazil in 1999 this committee began to define the SELIC rate value -benchmark in the fixed income segment, whose name refers to the Special System of Clearance and Custody (Sistema Especial de Liquidação e Custódia) -in order to keep inflation at the levels determined by the National Monetary Council (Conselho Monetário Nacional -CMN).
In order to support this evolution in the Brazilian economy, we observe wide, extensive and well-founded literature on the conduct of monetary policy, which was previously interested in addressing the instability period as in Holanda (1987), where the inflationary aspects vis-à-vis public finances are analyzed, or again in Holanda (1991), where the unsuccessful attempt to control inflation in the early 90s is studied.
Specifically on this less turbulent phase, Pastore & Pinotti (1999) affirm about the apparently incomplete stability, in view of the inflationary inertia driven by the exchange rate regime and the absence of a fiscal austerity counterpart.Now under the Target Regime and the Fiscal Responsibility Act, Bonomo & Brito (2002) model the inflation forecast based on a macroeconomic structural framework, holding not only to Taylor's traditional rule, as well as the effects of a combination between inflation and real exchange rate.Bonomo et al. (2003) suggest modeling through evolutionary games of the relationship between inflationary inertia and the coordination problem associated with the limited rationality of economic agents.Commenting further on the performance of this system in Brazil, it is worth mentioning Ferreira & Jayme Jr. (2005) and Teles & Nemoto (2005), whose results indicate to an evolution of the credibility of Brazil's Central Bank to fight inflation.
More recently, literature has directed greater attention on agencies responsible for deciding the respective interest rates on savings, such as the Feredar Open Market Committee (FOMC) in the United States, the Central Bank Council in Germany or COPOM in Brazil, BBR, Braz. Bus. Rev. (Engl. ed., Online), Vitória, v. 12, n. 6, Art. 2, p. 24 -47, nov.-dec. 2015 www.bbronline.com.braiming to demonstrate what influences the direction and the magnitude of these decisions.In this context, Portugal & Silva (2009) suggest, from the calibration of a backward-looking macroeconomic structural framework associated with the optimization of the loss function of the Central Bank of Brazil, that interest rates react in a contemporary way to changes in inflation, deviation of the Gross Domestic Product (GDP) relative to potential output and exchange rate fluctuations, in addition to the obvious concern with the smoothing factor on the changes of the targets for interest rates.Conceptually aligned, Caetano et al. (2011) propose the modeling of the SELIC target by COPOM, taking into consideration aspects associated with the non-stationary, obtaining a well specified model.
Quite common in this kind of study composing this empirical literature has been the use of a macroeconomic approach, based on quarterly data or even less frequently, in spite of the monetary decisions such as the definition of the interest rates being taken in meetings with characteristically regular over time spacing, taking place on a monthly basis till 2006 or every 45 days since then.Thus, this study aims to add to this literature by modeling COPOM's decision in the conduct of policy interest rates decision in Brazil from data with a weekly frequency.
Following methodologically Hamilton & Jordà (2002) Thus, without abiding in the variables of political character, but only the economic and financial ones, we make use of the Autoregressive Conditional Hazard (ACH) modeling, a generalization of the Autoregressive Conditional Duration (ACD) by Engle andRussell (1997 and1998) which is able to incorporate a instrumentalization in order to condition the framework of the probability of changing the SELIC target before the most diverse information available.Following that, we make use of the ordered probit, to predict in which direction and magnitude such changes occur.Bu applying this to the reality of an economy with different characteristics from the American one is done precisely by adaptations, especially with regard to the variables used with the supposed ability of influencing decision making processes of monetary policies.This definition for Brazil is based on the learning of other monetary studies for this economy and explicit signs in the minutes of the COPOM meetings.The study also highlights, in terms of time series, the existence of relevant forecasting series, in view of the forward-looking aspect, only available from 2002.
In short, the results suggest that SELIC's decision be taken by a committee with a behavior theoretically forward-looking with regard to future fluctuations in production and inflation in the country.Such a committee appears to still being concerned about the increased expenditures by the National Treasury.We can also observe that in thirty five out of the fifty six changes of the SELIC target rate carried out by COPOM over the analyzed period, the framework turns out to be able to properly shape the direction of change, being in general the order of magnitude change lower than the one actually practiced by COPOM.
The study is divided so that Section 2 is a review of monetary policy, addressing the adoption of the Inflation Target Regime in Brazil.The third section presents the methodology, while the results are described in section 4. In the fifth section, the final considerations are presented.

INFLATION TARGET REGIME IN BRAZIL
From the recent experience in other countries, it is evident that a gradual, predictable and controlled increase in prices of goods and services consists of an indicator of economic stability and a precondition for economic growth. 1 In short, a scenario with inflation under control allows economic agents to make more efficient allocational decisions, influencing even the distribution of income, since inflation consists of taxing in which its incidence is greatest on the poorest.
In the case of the Brazilian economy, after experiencing a long period with fixed exchange rate regime, and several crises, such as the debt crisis in the 80s and the economic downturn in the 90s, it is evident that until the mid-90s the country had presented numbers of hyperinflation, with the average annual inflation between 1986 and 1994 of 842.5%, reaching a peak of 82.39% per month in March 1990, as seen in the IPCA index (National Index of Consumer Price) in Figure 1.
1 See Mankiw (2000) on the trade-offs between inflation and real variables such as growth and unemployment.Thus, in order to take advantage of the benefits of a society with low and predictable price swings, and on account of an unsustainable and explosive scenario of the evolution of prices of the basic goods, Brazil follows, in a similar manner to various economies, the New Zealand experience in the 90s, beginning to adopt from January 1999 the floating exchange rate regime, macroeconomic foundation so that on the 2 nd of June 1999, the country would follow the Inflation Target Regime.This regime had an associated target with the IPCA, an    In this regime, the instrument used by the monetary a authority in controlling the inflation target is the systematic definition of the basic interest rate of the economy, the SELIC rate, i.e., the average rate of daily financing, backed by federal securities in the Special System of Clearance and Custody, and their possible bias, low, high or neutral.This is possibly the primary role when the scheduled or special meetings of the Monetary Policy Committee (COPOM), board established on June 20, 1996 and formed by the president of the Central Bank and its directors.With regards to the meetings of this committee (see Table 1), the first one took place on 25 and 26 June 1996, and up until July 2010, 152 meetings had taken place, of which only three were extraordinary.
As reported in this table, up until 2005 these meetings took place at an approximate monthly basis frequency, except for the months of October in 1997October in , 1998October in , 1999October in , and 2002, in , in time which these meetings occurred during periods shorter than two weeks and in March 1999 time which there was no meeting for over a month and a half.As of 2006, there were only eight annual meetings, which last for two days.Normally, during Tuesday sessions a comprehensive analysis of the economic situation is presented, covering several inflation indices, the analysis of time series for the purpose of labor market comparisons, economic activity, consumer confidence index, consumer expectations, the capacity utilization rates, credit and default, foreign trade and balance of payments, external environment and money market.On Wednesday, proposals for the interest rate are proposed, being followed by votes and a subsequent dissemination.
As reported in this table, up until 2005 these meetings took place at an approximate monthly basis frequency, except for the months of October in 1997October in , 1998October in , 1999October in , and 2002, in , in time which these meetings occurred during periods shorter than two weeks and in March 1999 time which there was no meeting for over a month and a half.As of 2006, there were only eight annual meetings, which last for two days.Normally, during Tuesday sessions a comprehensive analysis of the economic situation is presented, covering several inflation indices, the analysis of time series for the purpose of labor market comparisons, economic activity, consumer confidence index, consumer expectations, the capacity utilization rates, credit and default, foreign trade and balance of payments, external environment and money market.On Wednesday, proposals for the interest rate are proposed, being followed by votes and a subsequent dissemination.

THE AUTOREGRESSIVE CONDITIONAL HAZARD MODEL
High frequency data tend to be irregularly spaced in time.To work with this type of time series, one possibility is to derive a kind of point process, that is, a stochastic process that (1) An extreme useful generalization of the ACD consists in the Autoregressive Conditional Hazard (ACH), a framework that allows answering the same questions raised when the use of the ACD is associated with changes in the SELIC rate, but incorporating information not only intrinsic to the process in question, but others associated directly or indirectly, of financial or macroeconomic nature, that may be important in the decision making of policy makers, for example.
Without loss of generality, assume that the time interval is chosen so that no observed duration is lower than one period and the probability of occurrence of more than one event at a time is insignificant.
We can generalize the relationship (1) as an ACH (r, m) as: (2) where corresponds to a vector of known variables on t-1.model, it would be given by: (3) The robustness of the routines of maximization demands that the following restrictions on the parameters are met: e while the stationarity is associated with In relation to issues associated to the estimation of these frameworks, for a vector of parameters, we make use of the current conditional log likelihood function given by , where is a binary dummy that assumes the unit value in the event of change in the target rate at t and zero otherwise.We can write the distribution of the joint probability of and as: , (4) 2 Seeking greater clarification on the estimation of the ACD, see Engle and Russell (1998) and Pacurar (2006).
As for greater details on the formalization of relations between the specifications of the maximum likelihood of the ACD and ACH functions can be seen in Hamilton and Jordà (2002).
where correspond to the limits endogenously identified in the estimation.Take as the function that determines the probability of a standard normal variable assuming a value lower or equal to z, so that in this case the probability of the interest rate changing in a value is expressed by: 3 3 For details on the maximum likelihood estimation via this process, see Hamilton & Jordà (2002).
correspond nd nd nd nd to th  The main variable in question is the SELIC rate adopted in Brazil.Thus, Table 2 presents the schedule of changes in the target rate, where the period of validity of each rate, the rate value, the size of the change and its duration in days.With regards to the data frequency aspect, just as seen in Hamilton & Jordà (2002)  With regards to the variables used with alleged ability to influence monetary policy decision making, we propose an adaptation of the base originally used in Hamilton & Jordà (2002).The descriptions of the explanatory variables are reported in Table 3.In an empirical exercise, it would be ideal to make available a disaggregated database and the most extensive enough in temporal dimensions and cross section.For the Brazilian economy, given that the expectations series are available only from January 2002, we decided to limit this as the beginning of the sample, and the month of July 2010 as the last one.Thus, we defined a set, with a weekly frequency over the period from January 2002 until July 2010, more specifically the first week analyzed set between 03 and January 9, 2002 and the last between 22 and 28 July 2010, in a total of 447 weeks.All the time series used were taken from the Central Bank of Brazil.This adaptation to the Brazilian reality is strictly necessary, in keeping with the availability of existing proxy variables series and by expanding this group, focusing on variables used in other recent empirical work in Brazil, as well as the analyzes done in public minutes disclosed after COPOM's meetings.
In details, we made use of 17 variables: monetary in nature, capable of measuring accumulated inflation or deviations related to the target, seasonally adjusted metrics of accumulated GDP and its deviations related to its potential value, growth rates component aggregates of national accounts, public finance indicators, accumulated changes in the foreign exchange market, variables on the labor market and product and inflation expectations series for the next 12 months.BBR, Braz.Bus.Rev. (Engl.ed., Online), Vitória, v. 12, n. 6, Art. 2, p. 24 -47, nov.-dec. 2015 www.bbronline.com.br It is important to observe that the data used must be compatible with the period in which they are disclosed to the public in general.For example, we consider the Wide National Consumer's Price Index -WNCPI (Índice de Preços do Consumidor Amplo -IPCA) of a given month to be disclosed with an interval from nine to fifteen calendar days, according to the calendar made available by the Central Bank itself, as well as information about the labor market, which are made public only after three weeks and the national accounts in 70 calendar days.So when we consider the first or second week of a given month, the regarded IPCA is the one reported in the previous month.Information on production, imports, consumption, exports and Government expenditures in the last quarter 2008 for example, will only be taken into account by COPOM in the week between 12 and March 18, 2009.
Finally, figure 3 reports the evolution of the IPCA, the main variable which we believe to play an indisputable role in the COPOM's monetary decisions, in its various versions, be it a deviation from the target accumulated over the previous 12 months, or projected for the following year.A first intuitive observation based on this figure is the reduction over time of orders of magnitude and especially in the oscillation of the series of deviations from the inflation target, a heading that only begins to mark to guide the conduct of monetary policy by adopting the Inflation Target regime.In line with this evidence, a shift in one year of the accumulated inflation series would allow the observation of the apparent long-term relationship between this series and the expectation, confirming the predictive ability of this series published in the reports, Focus of the Central Bank.

RESULTS FOR THE ACH
When the estimation of an ACD or ACH model, we should observe some details, bearing in mind the trade-offs between efficiency and consistency, as well as the additional assumptions about the behavior of the variables in question.In this context, according to Pacurar (2006), just as in Engle & Russell (1998), Engle (2000), and Engle & Russell (2002), the ACD framework has common characteristics to those observed in the framework proposed in Bollerslev (1986), the Generalized Autoregressive Conditional Heteroskedastic (GARCH).That is, the ACD is considered as a counterpart of the GARCH for duration of data at high frequency.
The results by Lee & Hansen (1994) and Lumsdaine (1996) support that under conditions believed to be reasonable, the estimator properties almost maximum likelihood of a GARCH (1.1), such as obtaining consistent estimators and asymptotically normal, are maintained in case of a framework EACD (1.1) proposed by Engle & Russel (1998). 4Under certain conditions accepted in this literature, Lee & Hansen (1994) prove a corollary that allows the ACD model to be estimated as a GARCH, being necessary in order to estimate via almost maximum likelihood, impose restrictions on the conditional mean of the dependent variable, the source of duration, .
By holding onto the set of explanatory variables, we decided to carry out a procedure on Eviews of the backward type, that is, testing the variables in different subsets with all 17 together, 16 to 16, ..., 2 to 2, and with only 1. Regarding the ACH (r, m) model specification possibilities, we opted for a parsimonious strand, in which we tested the ACH (0.1), the ACH (1.0) and the ACH (1.1).Thus, as in a GARCH (1.1), which proves able to remove the dependency on the squares of returns, an ACD or a low order ACH also seems to satisfy this criterion in the case for temporal dependence on durations.Considering the three ACH's specifications in all estimations made to the accompanying sets of five or more explanatory variables, we identified insignificance in all estimated parameters, with values between -1.030 and -1.060 for the statistical log of likelihood.For virtually all sets tested this metric assumed value considerably higher in module when using the ACH specification (1.1).As in Jordà & Hamilton (2002), only estimations containing small groups of variables, among which, the cumulative inflation (infac), inflation expectations (expinf), the occurrence of the COPOM meeting (COPOM) and growth in the issuance of M2 (m2g), presented significance of the coefficients when using the specifications ACH (1.0) and ACH (0.1).Being limited therefore, only to the specifications in which there was significance of all coefficients, and following we opted for modeling with the greatest value of statistics in log likelihood, we defined by the specification that only the occurrence of the COPOM assembly was considered as relevant in a ACH (1.0) framework, as the results listed in Table 4.
The estimation of parameter α which showed insignificance at 5% for most of the specifications, suggests a negligible degree of persistence in serial correlation of durations, whilst the β associated with the discrepancy with a value of 0.875 allows evidencing that the stationary condition is satisfied, as well as the restrictions imposed on maximizing numerical routines previously presented.
The hazard rate value regardless of the instrumentation, would be 16.55%, while by incorporating the occurrence of COPOM meetings, the only relevant explanatory variable in the better specified framework, the hazard rate given by the relation (3) would be given 21,23%.Thus, we characterize an expected and intuitive increase in the probability of there being a change in the SELIC rate in weeks with the COPOM meeting vis-à-vis weeks without them.This is an increase of probability, of which the magnitude appears to be lower than expected, but not necessarily negligible.It is important to observe that there are no changes to the SELIC target on behalf of the Central Bank of Brazil, since 1998, unless it is in a week when meetings take place with the committee, unlike the previous years where all the changes observed in the years 1996 and 1997were settled in one or two weeks after the meeting taking place.
With regard to diagnostic tests considered as standard in this type of statistical modeling, there is violation of the assumptions of gaussianity and homoscedasticity residuals, which can interfere with the estimation of efficiency, being however commonly found in literature.

RESULTS FOR THE PROBIT
Most of the changes adopted for the SELIC target rate were of absolute magnitude at 0.25%, 0.50% and 0.75%, especially from 2002 onwards.Thus, we made the choice of marks, or possible theoretical changes in interest rates at 0.25% as suggested by Hamilton & Jordà (2002), however, greater amplitude becomes possible: 3 marks up or down, that is, varying from 0.75% to +0.75%.All reductions with higher order of magnitude 0.75% are accounted for in the same category, as well as all increases higher than 0.75%, regardless of the value itself, whether 1.0% or 2.0%, for example.
In order to determine the intervals corresponding to each possible variation in the change in interest rates, the explanatory variables used in the Probit model were exactly the same ones as in the exercise described in the previous subsection, reported in Table 3.
According to the results, coefficients associated with the explanatory variables were significant at 5% only when the estimation of parsimonious frameworks containing: the increase in Government expenditure (govg), GDP deviations in relation to potential GDP (dpib) and inflation expectations (expinf) and GDP (exppib) both for the next 12 months.All other variables appeared significant isolated or jointly.However, with particular regard to the specification with greater value for the statistical likelihood, we identified the relevance with significance at 1% individually and jointly of the variables that measure two expectation series and government expenditure growth only.The results of this framework are reported in Table 5. Log likelihood condition fuction of the marks:

Explanatory variables
Parameter Variable Estimated Intuitively, we identify that increases in any of these variables in question implies the expected increase in the SELIC rate, making the framework more sensitive to the forecast of production, whose average value over the period was 3.39% per annum, with a coefficient of 0.31.Sensitivity to expected inflation, which averaged 5.31% per annum, was 0.2, while the parameter associated with Government expenditure (mean of 3.11% per annum) was 0.16.More interestingly, in the framework of ordered Probit used with weekly frequency, during  There is a difficulty in establishing comparisons with results for the Brazilian economy, bearing in mind that comparison with quarterly results are neither that valid nor informative, for example.Even if we are not dealing with an adequate theoretical reference, the error obtained here was less than 0.3421% obtained when using the strategy to always use the latest value of the SELIC, as a forecast.These results are qualitatively and quantitatively robust in the scenario where we use the model every week, regardless of whether or not COPOM's meetings are being held.
In thirty five out of the fifty six changes in the SELIC target rates implemented by COPOM, the model correctly suggested the direction of change, so being in general the order of the change magnitude lower than the one actually practiced by COPOM.In 30 meetings where there was no change in the rate, the model predicted changes in directions, which would be mostly observed in future meetings, characterizing the Central Bank designed by the model, as being more sensitive to macroeconomic and financial variations, reacting more frequently and with less intensity than in practice, but aligned on the contractionary or expansionary nature of monetary policy.
, whose application analyzed the behavior of the Federal Reserve Bank (FED), this study aims to accommodate this proposed technique to the reality of a developing economy, which recently experienced a period of economic stability, making it possible to add by proposing subsidies to answer questions on the Brazilian case, such as: when, in what direction and to what extent COPOM decides to change the basic interest rate?Which variables are taken into account?What is the probability that the target rate would change the following week given the available set of information today?Would it be possible to show a backward or forward-looking behavior by the monetary authority?
indicator defined and widely disclosed by the National Monetary Council (Conselho Monetário Nacional -CMN), composed by the Minister of Finance, Minister of Planning and the President of the Central Bank of Brazil.

Figure 1 -
Figure 1 -Evolution of the monthly variation of the National Index Wide National Consumer's Price Index (IPCA) a a Source: Brazilian Institute of Geography and Statistics (Instituto Brasileiro de Geografia e Estatísticas -IBGE).According to Figure 2, this monetary regime has been characterized by the adoption of one digit targets with annual correction bands, in the order of 2% or 2.5% above and below the center, thus allowing the central bank flexibility in controlling the IPCA, a benchmark index which measures the variation in the cost of the representative consumption basket of the population with an income up to 40 minimum wages in 12 metropolitan areas.

Figure 2 .
Figure 2. Inflation targets, limits and the observable IPCA (% per year) a, b a Source: DEPEC-BACEN (Central Bank).b In the case of the year 2010, the IPCA is reported in the year up until July.
builds random points in time and are called arrival times.Thus, consider the arrival times where .In the present study, corresponds to the n-th change of the SELIC in the considered time interval.To this set of time arrivals it is common to associate it with the step function .In this study, the step function also carries out the discrete count changes in the benchmark interest rate over time.BBR, Braz.Bus.Rev. (Engl.ed., Online), Vitória, v. 12, n. 6, Art. 2, p. 24 -47, nov.-dec.2015 www.bbronline.com.brIn an objective description of the Autoregressive Conditional Duration model (ACD), define as an interval between the n-th and the n + 1-th changes in the SELIC target rate.Be the expected time of duration of considering the past observations .Engle & Russell (1998) proposed the following specification for frameworks that follow an ACD (r, m) as an use of the step function : The relation (1) summarizes a statistical framework ad hoc titled ACD which allows modeling discrete time duration of a random variable whose change, regardless of the magnitude or direction, is associated with the history of observed durations, as well as the history of durations expected by the framework itself.The relation (2) describes the ACH generalizes incorporating the possibility of testing the individual and the whole explanatory power of the exogenous observable variables to the duration to be modeled.A very useful and relevant concept in this modeling consists in the Hazard rate, which is defined as the conditional probability that the interest rates change in week t given the set of information available in , , being expressed by .Suppose initially that the only information that we have until the time are the dates of the previous changes in interest rates.Thus, the Hazard rate would not change until there was a new change in interest rates.The Hazard rate of the model ADC (1,1), for example, it 2 3.2 CHANGE FORECAST IN THE SELIC TARGET RATE: THE FRAMEWORK OF ORDERED PROBIT Once the ACH framework allows us measuring the probability of a change in the SELIC target rate taking place in the following week and that variables affect this probability, to predict the SELIC rate value is to determine in which direction this change will occur and what its magnitude in each week, being associated with this change the same set of variables used in the empirical exercise about the likelihood of change, which will be described in detail in Section 4. These changes will be seen as marks and the respective time series obtained from the ordered pairs, marked point process is such that, the first coordinate corresponds to the points where the interest rate was changed and the second coordinate corresponds to the size of the change.Thus, we define the mark or the size of the change in t, by .The set containing the information available in includes , variable associated or not with the change in the target, the very magnitude of this change, , and a vector of variables that influence the decision of the Central Bank.
where corresponds to the function of probability of change associated with the econometric model of the ACH, whilst is associated with the characteristic function of the Probit about the direction and order of the magnitude of the change.Both probability functions are parameterized by the vectors and , respectively.Whereas changes in interest, that is, the marks occur in discrete amounts, we suggest the use of the ordered response model as in Hausman, Lo & MacKinlay (1992).In this context we estimate here the Ordered Probit model, a technique used in empirical studies of dependent variables that assume only a finite amount of values that can be ordered, which consists of a generalized linear regression model for cases where the dependent variable is discreet.This framework is shown capable of capturing the impact of relevant variables considering discretization and irregular time intervals.Take as a vector of observed variables in the week before t.Assume hypothetically, the existence of a random non-observable variable that depends on in the following way: , where .When the Central Bank decides to change the interest rate, it has available k different discrete amounts to choose from.Set the possible changes in the target as of , which, without loss of generality, may be increasingly ordered.Considering ,, i.e., change in interest rates at t, the observed change in interest rates, , will be related to the random non-observable variable as follows:,

Figure 3 -
Figure 3 -IPCA (% pa): cumulative deviation throughout the year in relation to the inflation target, inflation in the last 12 months and inflation expectations for the next 12 months a a For the year 2010, the IPCA of 3.10% corresponds to the accumulated value observed from January to July this year.
test (c i = 0, i = 1, ..., 6 ): Stat.F = 87,874 (P-valor = 0,000) a Statistical log likelihood = -238,952.b Definition of the explanatory variables:corresponds to the expected revised quarterly real GDP growth for the next 12 months, corresponds to the expected revised monthly inflation measured by the IPCA/IBGE for the next 12 months, whilst corresponds to the growth rate compared to the respective previous year's period, associated with cumulative quarterly series throughout the year and seasonally adjusted government consumption.All these series are from the Central Bank of Brazil.*Parameter not significant at 5%.
Figure 4 illustrates the performance of the SELIC expected exercise period ahead, during the period from January 2002 to July 2010, being that the model used only for SELIC forecasts during weeks in which COPOM's meetings took place, minimizing the error associated with the model that allows changes within weeks without meeting.Based on the SELIC time series (gray) and the prediction by the statistical framework estimated in this study (dotted black) we obtained an order of 0.3358% forecast error.

Figure 4 -
Figure 4 -In-sample forecast one period ahead of the SELIC rate based on ACH and Ordered Probit models during the period from January 2002 to July 20105 CONCLUSIONSThe object of interest of this paper is a consequence of monetary policy in force after the Plano Real in 1994, period in which Brazil experiences an economic scenario of stability, with a controlled evolution in price developments.Thus, following Jordà &Hamilton (2002) methodologically, in order to model and understand the reaction of the Central Bank in SELIC interest rate decision, the main results obtained, suggest that the committee responsible for monetary policy in Brazil appears to have a forward-looking behavior in relation to the production and inflation in force in the country, besides worrying about the apparently absent balance in recent years between fiscal and monetary policies, since the increased expenditures by the National Treasury would also be pushing the Central Bank to adopt a contractionary monetary policy.Partially, this evidence on the forward-looking bias corroborates withCaetano et al. (2011), according to which, COPOM considers variables such as the deviation of inflation expectations as being twelve-step ahead in relation to the target, the consumer confidence index, lagged interest rate, as well as the ratio of public sector primary surplus / GDP and formal employment.
, we should consider a time unit, such that the probability of there being more than one change in the interest rate in this period of time would be very negligible.It is also essential to pay attention not to choose a time interval too great in order not to lose important date of the sample.There is the historical evidence that meetings have occurred at intervals of less than two weeks and that during the period between January 2002 and July 2010, and the period of a rate's shortest duration was of 21 days, that is, at no time two changes are observed within the same week, so therefore the time unit used here is weekly.It is possible to observe that over the fourteen years of the COPOM, more precisely from its first meeting, on 25 and 26 of June 1996, up until the 152 nd meeting on 20 and 21 of July 2010, on average, SELIC is kept constant throughout 52 calendar days, almost two months, this duration being very volatile, with the standard deviation of almost 50 days.

Table 3 -Detailed Description of the Explanatory Variables a Code Description Monetary variables copom
Total net debt of the consolidated public sector (% of GDP) Dummy variable that associates or not the ocurrence of COPOM's meetings in the current week.diTheSELICinterest rate defined by the COPOM revised weekly (% per annum)m2g The final balance growth rate for the period of the broad money supply, M2, revised monthly Inflation Metrics dinf Inflation deviation (measured by the IPCA/IBGE) accumulated over the year compared to the current inflation target, revised monthly (% per annum) infac acumulated inflation over the last 12 months measured by the IPCA/IBGE (% per annum) govg Growth rate compared to the respective period of the previous year, associated with cumulative quarterly series throughout the year and seasonally adjusted government consumption BBR, Braz.Bus.Rev. (Engl.ed., Online), Vitória, v. 12, n. 6, Art. 2, p. 24 -47, nov.-dec.2015 www.bbronline.com.brgovpib a Source: Central Bank of Brazil

Table 4 -Estimation of ACH Framework (1.0) During the Period from January 2002 to July 2010 a, b
Log likelihood statistics = -1.035,108.b Definition of the explanatory variables: COPOMN(t) corresponds to the dummy variable associated with the occurrence of the COPOM meeting in the current week.*Significant parameter at 5%.