Monetary policy surprises and jumps in interest rates

Transcription

1 Monetary policy surprises and jumps in interest rates Roberto Meurer Department of Economics Universidade Federal de Santa Catarina, Brazil André A. P. Santos Department of Economics Universidade Federal de Santa Catarina, Brazil Douglas E. Turatti Department of Economics Universidade Federal de Santa Catarina, Brazil Abstract We consider a monetary-jump model to measure the contribution of jumps to the total volatility of interest rates in the Brazilian interbank market and to assess the extent to which central bank s unanticipated monetary policy decisions are driving these jumps. For that purpose, we estimate a mixture GARCH-jump model that disentangles two components of interest rate volatility: a GARCHtype specification that models conditional heteroskedasticy accounts for the volatility during normal times and a Poisson process that models the occurrence of abrupt changes in interest rates. Our empirical application suggest the contribution of jumps to the total volatility is substantial, and that monetary policy decisions partly explains the occurrence of those jumps. In particular, we find that the likelihood of a jump occurring during a meeting day of the central bank s monetary policy committee (COPOM) is higher in comparison to that of a non-meeting day. ˆ Keywords: Interest rates, jumps, monetary policy surprises ˆ JEL Classification: E43, E52, C32 ˆ Submission Area: Applied Macroeconomics 1

2 1 Introduction In this paper, we shed light on the market reaction to changes in the Brazilian central bank s target interest rate. Such policymaking consists of deciding whether or not to change the target rate and, if so, to what extent. This decision not only affects short-term market rates, but also long term rates, thus affecting the value of interest-rate-linked assets and influencing spending in interest-rate-sensitive sectors of the economy such as housing, consumer durable goods, and investments. In this sense, studying the market reaction to monetary policy is important not only to assess the effectiveness of monetary policy but also to assess the central bank s predictability in executing monetary policy. Most of the empirical evidence on the market reaction to monetary policy is based on the so-called event-study approach introduced by Cook and Hahn (1989). This approach, although useful and widely employed, has at least one major limitation: the unanticipated component of the monetary policy is measured by a single, time-invariant coefficient. This limitation is important since one expects time variation in the market reaction to monetary policy to be due to changes in market expectations or in macroeconomic conditions. We overcome this limitation by employing a novel approach proposed by León and Sebestyén (2012) to measure monetary policy surprises. The model follows closely the approach proposed by Chan and Maheu (2002), Maheu and McCurdy (2004), and Rangel (2011), as it aims not only to capture the time series characteristics of the series and estimate the effects of monetary policy surprises, but also to evaluate the extent to which jumps contribute to the volatility of interest rates. An appealing feature of this model is that it disentangles two components of total conditional volatility, namely a GARCH component that captures variations in normal times and a jump components that accounts for abrupt changes. Johannes (2004) and León and Sebestyén (2012) argue that news is incorporated into asset prices either smoothly or abruptly, inducing sharp movements in the latter case. It is a well-known empirical phenomenon that asset returns are subject to infrequent jumps, and that these jumps play a dominant role in interest-rate dynamics; see Das (2002). According to the general view, jumps occur when new and unexpected information arrives in the markets, which makes investors revise their expectations. This adjustment may occur in an abrupt way, since this new information is generally some piece of public news such as macroeconomic announcements or monetary policy decisions (Maheu and McCurdy, 2004; Beber and Brandt, 2010; Rangel, 2011). The distinction between normal and abrupt information flows is important, since central banks try to pursue their monetary policy and shape market expectations in a smooth way. Large surprises cause uncertainty in the markets and reduce central banks predictability. 2

3 This paper amends the literature on the Brazilian interest market s reaction to monetary policy surprises by considering a more robust approach than the event-study approach adopted by Tabak (2004) and Buchholz et al. (2012). One major advantage of the approach employed in the current paper is that it enables us to obtain time series properties of the market reaction to monetary policy in Brazil. For that purpose, we estimate a GARCH-jump model proposed by León and Sebestyén (2012) in which both jump intensity and the jump mean are time-varying. Our empirical evidence is based on daily closing yields of the DI-Pre swap rates contracts with maturities of 30, 60, 90, 120, 180 and 360 days from January 2003 to June DI-Pre swaps are futures contracts in which a party pays a fixed rate over an agreed principal, and receives a floating rate over the same principal, while the reverse occurs for his or her counterpart. DI-Pre swap rates are currently established as the rate on which interbank lending operations in the Brazilian financial system are based. The main results of this paper can be summarized as follows. First, we find that jumps are an important component of the total volatility of interest rates. In particular, the contribution of jumps to the total volatility reaches very high levels, specially for shorter maturity rates. Second, we observe that while the GARCH component of total volatility seems to increase with maturity (i.e. longer maturities tend to be more volatile), the jump component of volatility seems to be more pervasive in the case of shorter rates. Third, accounting for time variation in both jump intensity and jump mean leads to an improved model fit and to a larger role of jumps in explaining the volatility of daily variation in interest rates. Fourth, we find that even though the contribution of jumps to total volatility in meeting and nonmeeting days are similar, the jump intensity and the probability of at least one jump are substantially higher during COPOM meeting days than during non-meeting days. This result is broadly in line with the intuition that unanticipated monetary policy decisions are likely to induce jumps in interest rates and that one of the main sources of jumps is unanticipated monetary policy decisions. Finally, our evidence indicates that monetary policy surprises in Brazil are more likely to induce jumps in interest rates than in other interest-rate markets, such as the European evidence reported in León and Sebestyén (2012). The paper is organized as follows. Section 2 provides a review of the related literature. Section 3 describes the GARCH-jump model specification and the estimation procedure. Section 4 provides an empirical implementation and provides a detailed discussion of the results. Finally, Section 5 concludes. 3

4 2 Literature review The predominant approach to assess market reaction to monetary policy is based on the event-study approach introduced by Cook and Hahn (1989). In this approach, one-day changes in interest rates are regressed against the respective changes in the target rate, i.e. Δrt i = α + βδtarget t + ε t, where Δrt i is the change in the market interest rate with maturity i and ΔTarget t is the change in the central bank s target rate. The parameter β measures the surprise of policymaking, i.e. the unanticipated impact of changes in the target rate. In the limiting case that the monetary decision is fully anticipated, β would be zero. Cook and Hahn (1989) show that changes in short- and long-term interest rates can be explained by changes in the United States (US) Federal Reserve s (FED s) federal funds rate, and that this influence weakens the longer the term of the market interest rate. Similar approaches using US data are employed in Roley and Sellon (1995), Balduzzi et al. (1998), and Kuttner (2001), whereas international evidence for the British and German interest-rate markets based on the event-study approach are found in Dale (1993) and Hardy (1998), respectively. Variations of the event-study approach can also be found in Haldane and Read (2000) and Bohl et al. (2008). As for the Brazilian interest-rate market, the empirical evidence suggests that the term structure of market interest rates is able to anticipate to some extent central bank s monetary decisions. However, Oliveira and Ramos (2011) show that monetary policymaking in Brazil also has an unanticipated component that surprises the market and affects the term structure. 1 The event-study approach to test the market reaction to changes in the target rate has been also employed by Tabak (2004) and Buchholz et al. (2012). Tabak (2004) tests the effect of decisions made by the COPOM concerning the target rate on various market interest rates from 1996 to 2001 and finds a positive response that decreases with the maturity of the interest rate. The author also studies the response of market interest rates to changes in the official rate following the adoption of Brazil s inflation-targeting framework in He finds that the adoption of inflation-targeting led to a dampening effect of interest rate surprises along the term structure. Buchholz et al. (2012) use a more recent sample period ( ) to assess the role of important macroeconomic and political changes in the Brazilian economy. The authors find that the reaction to surprises in monetary policy became less abrupt after the introduction of Brazil s inflation-targeting framework in 1999 and the reduction of macroeconomic instabilities and political uncertainties in Unlike the event-study approach proposed by Cook and Hahn (1989), the approach adopted in this 1 Marçal and Pereira (2007) show that not only expectations and economic fundamentals affect the term structure in Brazil, but also variations in risk premia. 4

5 paper proposes a characterization of the jump distribution over time and allows for the examination of each individual monetary policy decision as well as the impact of alternative surprise variables in a time-varying fashion. This approach considers that monetary policy decisions can be expressed in two dimensions that help us understand how central banks shape market expectations across the yield curve. These dimensions can be expressed in two main factors. The first one is the level factor, which corresponds to decisions that shift the yield curve level. The second one is the slope factor, which reflects monetary decisions causing changes in the slope of the term structure. Both are key variables, since the level of the yield curve is consistent with the market expectation of long-term inflation and the slope is found to be a predictor of the business cycle (Estrella and Hardouvelis, 1991; León and Sebestyén, 2012). 3 A monetary-jump model for interest rates In this section we review the monetary-jump model used in this paper. We consider that interest rates follow a process of the form Δr τ t = μ t + ε 1,t + ε 2,t, where p μ t = α 0 + α i Δrt i, τ (1) i=1 ε 1,t = σ t z t, (2) n t ε 2,t = J t,k μ j λ t. (3) k=1 Δr τ t in (1) is the change in the market interest rate with maturity τ, and α i coefficients in (1) stand for possible autoregressive terms in the conditional mean dynamics. In our empirical implementation, we consider a first-order autoregression and set p = 1 in (1). ε 1,t in (2) is a zero-mean normal innovation representing a diffusive information flow, z t is an i.i.d. standard normal variable, and σ t is the conditional volatility of ε 1,t. In (3), ε 2t is the jump innovation term and represents the impact of abrupt information arrival, J t,k is the jump size which is (in our baseline model) assumed to be normally distributed with constant mean and variance μ j and σj 2, respectively. The term μ jλ t in (3) adjusts the jump innovation to have a conditionally zero mean. We assume that both {z t, J t } and {z t, n t } are independent. Finally, n t refers to a Poisson process with time-varying conditional intensity parameter λ t for the number of jumps (n t {0, 1, 2,...}), occurring in the interval {t 1, t}. The density of n t conditional on the information 5

6 set I t 1 is Pr (n t = j I t 1 ) = exp ( λ t) λ j t j! j = 0, 1, 2,.... The jump intensity can also be written as λ t = E(n t I t 1 ), and therefore can be interpreted as the ex ante assessment of the expected number of jumps over the period {t 1, t}. In order to incorporate the change in our conditional forecasts as the information set is updated (i.e. the ex post assessment of the number of jumps), we use Bayes rule, thus yielding Pr (n t = j I t ) = f (Δr t n t = j, I t 1 ) Pr (n t = j I t 1 ) f (Δr t I t 1 ) j = 0, 1, 2,.... (4) The filter in (4) is very informative and useful for inference purposes, since the probability that at least one jump occurred on a given day is simply Pr(n t 1 I t ) = 1 Pr(n t = 0 I t ), which can be directly calculated from (4). Finally, the ex post expected number of jumps can be also calculated as E (n t I t ) = 3.1 Conditional variance dynamics j Pr (n t = j I t ). j=0 In order to specify the dynamics of the conditional variance for the model in (1) to (3), it is useful to write down the first two conditional moments of Δr t : E(Δr t I t 1 ) = μ t, Var(Δr t I t 1 ) = σ 2 t + λ t (σ 2 j + μ 2 j). (5) The second term in (5), λ t (σj 2 + μ2 j ), is of main importance as it is the jump contribution to the (total) conditional variance. 2 As for the first term, σ 2 t, we consider a variance-targeting GARCH process of the form σ 2 t = (1 α β) σ 2 + αε 2 t 1 + βσ 2 t 1, (6) where σ 2 is the unconditional variance of ε t and ε t 1 = ε 1,t 1 +ε 2,t 1 denotes the total innovation observed at time t 1. Therefore, σt 2 includes not only the impacts of past diffusive innovation but also the effects of past jump innovations on interest-rate changes. 2 Higher order conditional moments for this model are obtained by Maheu and McCurdy (2004). 6

7 3.2 Modeling time-varying jump intensity and jump mean The specification for the jump process discussed above is based on a Poisson process with constant jump intensity and constant jump mean. This specification, however, can be extended in order to allow for a jump arrival process that evolves over time. For that purpose, we follow León and Sebestyén (2012) and model both jump intensity and jump mean as functions of monetary policy surprises as measured by the first two PC of the term structure. In the first case, the intuition is that the larger the monetary policy surprise, the larger is the probability of a jump. Therefore, the intensity of the jump arrival process is given by λ t = λ 0 + λ 1 S 1,t + λ 2 S 2,t, (7) where S 1 and S 2 denote the level and slope factors, respectively. Since the model nests the constant jump intensity specification (with the constraint λ 1 = λ 2 = 0), standard likelihood ratio tests can be carried out to test whether the monetary-jump model is superior to the constant λ model. One can argue that a dummy variable for the monetary authority meeting is sufficient to account for jumps induced by those meetings. However, as León and Sebestyén (2012) point out, the mere occurrence of a monetary meeting should not increase the probability of a jump, but sharp changes are more likely if the COPOM causes a larger surprise. Thus a simple dummy variable is unable to accurately capture the dynamics of the jump arrival process; instead, jump intensity depends on the surprise itself. The authors also defend that it is not the sign of a surprise that matters in inducing jumps, but rather its magnitude. As for the jump size mean, we follow Beber and Brandt (2010) and consider a conditional specification built upon the intuition that monetary surprises can also affect the size of jumps, as very large surprises tend to be associated with large daily changes in interest rates. In this sense, it seems adequate to let the jump size mean depend on monetary surprises for the interest rates. We also follow León and Sebestyén (2012) and allow for asymmetric reactions, i.e. negative surprises have different impacts than positive surprises: μ j,t = β 0 + β1,i S i,t + β+ 2,i S+ i,t, (8) where i = 1, 2 stands for the level and slope factors, respectively, and S + i,t max(s i,t, 0) and S i,t min(s i,t, 0). 7

8 3.3 Estimation of the model Given the set of equations in (1) to (3), the distribution of Δr t is Normal conditional on I t 1 and to j jumps. 3 The corresponding probability density function is written as: f (Δr t n t = j, I t 1 ) = 1 ( ) exp (Δr t μ t + μ j λ t jμ j ) 2 ( ). (9) 2π σt 2 + jσ2 2 σt 2 + j jσ2 j Integrating out n t yields the conditional density function of Δr t in terms of the observables, i.e. f (Δr t I t 1 ) = f (Δr t n t = j, I t 1 ) Pr (n t = j I t 1 ). j=0 León and Sebestyén (2012) point out that this is a mixture-normal model, where the mixture coefficients are Poisson probabilities driven by monetary surprises. Moreover, the likelihood in (9) involves an infinite summation over the possible number of jumps. For feasible estimation we follow León and Sebestyén (2012) and truncate the summation at 20. Given our model parameter estimates, the probability of more than ten jumps is essentially zero. In order to obtain estimates for the unknown parameters, the log-likelihood function, given by lnf (Δr t I t 1 ), has to be maximized. 4 Empirical implementation In this section we describe the empirical application of the monetary-jump model discussed in Section 3. We provide a description of the data set, implementation details and, finally, discuss the results. 4.1 Data The data employed in the paper consist of daily closing yields of the DI-Pre swap rates contracts with maturities of 30, 60, 90, 120, 180 and 360 days. There are no intermediate cash flows and the contracts are only settled on maturity. The reason for using swap rates instead of government bond rates is that the latter is more frequently transactioned than the former. The DI-Pre futures contracts are traded over-the-counter and are currently established as the rate on which interbank lending operations in the Brazilian financial system are based; see Buchholz et al. (2012). Data are obtained from the web site of the Brazilian Central Bank ( Each contract has 2767 observations ranging 3 Rangel (2011) provides a formal derivation of the likelihood function of the GARCH-jump model. 8

9 from January 2003 to June We report in Table 1 the mean and variance of the target and DI-Pre rates in both levels and daily variations for each of the contracts considered in this paper. We observe in Table 1 that the variance of daily variations increases with maturity. We also identify in the sample the meeting days of the monetary authority committee (COPOM). Table 1 reports the instances in which COPOM decisions are either to increase, decrease or to maintain the target rate (refereed to as the Meta Selic) unchanged. 4 [Table 1 about here.] 4.2 Results This section reports the results of the monetary-jump model discussed in Section 3. We take as baseline model the GARCH-jump model with constant jump intensity and focus subsequently on the GARCHjump model with time-varying jump mean and jump intensity. GARCH-jump model with constant jump intensity Table 2 reports parameter estimates of the GARCH-jump model with constant jump intensity along with an estimate of the average contribution of jumps to total volatility as well as the portion of jumps occurring on COPOM meeting days. We observe that the jump intensity parameter, λ, ranges from 0.60 to 0.70 across maturities and is statistically significant in all cases, thus suggesting that jumps play an important role in explaining the daily variation in DI-Pre rates. The parameter estimates of the jump mean and jump variance, μ j and σ 2 j respectively, as well as the GARCH parameters, α and β, are also significant in the majority of the cases. We also observe that jumps account for a substantial portion of the total volatility, since the average contribution of jumps to total volatility ranges from 15% in the 180-day rate to 40% in the 30-day rate. A pattern that arises is that the jump contribution to total volatility decreases with maturity, which suggests that unexpected decisions in COPOM meetings affect shorter maturity rates more heavily, which is is line with the evidence for the Brazilian interest rate market in Buchholz et al. (2012). Moreover, it is worth noting that these values are substantially larger than those found by León and Sebestyén (2012) for the case of European interest rates, as well as those found by Beber and Brandt (2010) for the case of US bond returns. More specifically, León and Sebestyén (2012) find that 4 COPOM meetings took place on a monthly basis until Since then, meetings take place every month and a half. We also take into account an important change implemented in August 2003 regarding the announcement of the target rate. Up to that date, decisions were announced in the same afternoon of the decision date, with the market (partly) reacting to the announcement itself. From August 2003 onwards, the announcement occurs in the evening, with the full market reaction reflected on the day after the announcement. 9

10 the jump contribution to total volatility ranges from 7% to 24%. Our evidence for the Brazilian market, on the other hand, suggests that the contribution of jumps is more pronounced. It is also worth noting that a non-negligible portion of the total number of jumps occurs on COPOM meeting days. We find that approximately one out of nine jumps in the 30-day and 60-day rates occur on meeting days, whereas one out of six jumps in the 120-day and 360-rates occur on meeting days. The results are even more striking when we restrict our analysis to the ratio of the number of jumps on meeting days to the total number of meeting days. We find that on around 60% to 70% of COPOM meeting days at least one jump occurred in the 30-, 60-, 90-, and 180-day rates, whereas the same figures for the 120- and 360-day rates vary from 30% to 45%. [Table 2 about here.] GARCH-jump model with time-varying jump intensity We now consider the estimation results of the GARCH-jump model with time-varying jump intensity, which is a variant of the constant jump intensity model discussed above. Table 3 reports key parameter estimates along with the results of the likelihood ratio test with respect to the constant jump intensity model. The first striking result is that the GARCH-jump model with time varying jump intensity registers a higher log-likelihood function in all cases. Moreover, the likelihood ratio test is significant in all cases, suggesting that this model statistically outperforms its constant jump intensity, constant jump mean counterpart. [Table 3 about here.] Table 4 reports jump characteristics of DI-Pre swap rates according to the GARCH-jump model with time-varying jump intensity. The first key result is that the average contribution to total volatility is substantially higher than the one previously obtained with the GARCH-jump model with constant jump intensity (see Table 2). For instance, the jump contribution in the 30-day rate is 30% in the constant jump intensity model and 54% in the time-varying model. Similar results are found for the remaining rates and are also in line with previous evidence such as by León and Sebestyén (2012). This result suggests that accounting for time variation in both jump intensity and jump mean leads to a larger role of jumps in explaining the volatility o the daily variation in interest rates. The second striking result is that the number of jumps in both meeting and non-meeting days is also higher in the time-varying jump intensity model with respect to its constant jump intensity counterpart. Finally, the third striking result in Table 4 10

11 refers to the differences in volatility and jump characteristics in meeting and non-meeting days. We find that the average total volatility during meeting days is higher in comparison to non-meeting days in all cases. Even though the contribution of jumps to total volatility in meeting and non-meeting days tends to be similar, we observe that the jump intensity and the probability of at least one jump are substantially higher during meeting days, suggesting that jumps are very likely to occur during meeting days. This result is broadly in line with the intuition that unanticipated monetary policy decisions are likely to cause jumps in interest rates. Moreover, the likelihood of a jump occurring in a meeting day is higher (lower) for shorter (longer) maturity rates. It is also worth relating the results in Table 4 with those reported in the existing empirical evidence for other interest rate markets. For instance, the results for the European interest rate market obtained by León and Sebestyén (2012) with a very similar econometric framework to the one adopted in this paper reveal that the probability of at least one jump in the 1-month and 2-month rates during a meeting day is 0.29 and 0.37, respectively, whereas our estimates for the Brazilian interest rate market case are 0.60 and 0.56, respectively. Our evidence, therefore, indicates that meetings are more prone to be the sources of surprises, and henceforth to induce jumps, in comparison to the European evidence. [Table 4 about here.] Summary of main results and policy implications The results of the GARCH-jump model reported in Tables 2 to 4 can be summarized as follows. First, we find that jumps are an important component of the total volatility of interest rates. In particular, the contribution of jumps to total volatility reaches very high levels, specially for shorter maturity rates. Second, we observe that while the GARCH component of the total volatility seems to increase with maturity (i.e. longer maturities tend to be more volatile), the jump component of the volatility seems to be more pervasive in the case of the shorter rates. Third, accounting for time variation in both jump intensity and jump mean leads to an improved model fit and to a larger role of jumps in explaining the volatility of the daily variation in interest rates. Fourth, we find that even though the contribution of jumps to total volatility in meeting and non-meeting days are similar, the jump intensity and the probability of at least one jump are substantially higher during COPOM meeting days than during non-meeting days. This result is broadly in line with the intuition that unanticipated monetary policy decisions are likely to induce jumps in interest rates. Fifth, our evidence indicates that monetary policy surprises in Brazil are more likely to induce jumps in interest rates than is the case in the other interest rates markets, such as 11

12 the European evidence reported by León and Sebestyén (2012). One important policy implication from our results is reinforcing the importance of expectations coordination in monetary policymaking. Communication between the central bank and the market will affect not only expectations but also macroeconomic fundamentals and asset values. If the central bank is able to generate fewer jumps, the variance of interest rate-linked assets will also be reduced. This could mean that fewer shocks are transmitted to asset markets and that the effects of monetary policy can be smoothed. On the other hand, jumps are, in a certain way, unavoidable because of discretion over the monetary policy instruments. 5 Concluding remarks The importance of jumps in explaining the volatility of interest rates depends intrinsically on how central banks shape market expectations regarding the target rate (Roley and Sellon, 1995). One expects that the higher the central bank s predictability, the lower the contribution toward jumps should be, since in this case market expectations regarding monetary policy are incorporated in a smoother way. Conversely, one expects that the contribution toward jumps is higher the less predictable the central bank is. Specifically for the Brazilian economy, these market expectations were largely shaped by the introduction of the inflation-targeting framework in Tabak (2004), for instance, finds that the adoption of inflationtargeting led to a dampening effect of interest rate surprises along the term structure. Furthermore, the political stability and the macroeconomic equilibrium that Brazil enjoyed after 2003 also affected agents reaction to monetary policy. In particular, Buchholz et al. (2012) find that surprises in policymaking for the period before this macroeconomic equilibrium are considerably higher. The evidence provided in this paper amends the existing literature in several aspects. First, we employ a more robust econometric framework with respect to the event-study approach adopted by Tabak (2004) and Buchholz et al. (2012). Our approach, which is based on a mixture GARCH-jump model, disentangles the total volatility of interest rates in a GARCH component that accounts for the normal volatility and a jump component that accounts for the abrupt changes that are likely to be induced from unexpected changes in monetary policy. Second, our empirical evidence based on daily DI-Pre swap rates with post-2003 data and with five distinct maturities shows that the jump contribution to total volatility varies substantially over time, and that the probability of a jump occurring during a meeting day is higher than the probability of a jump occurring in non-meeting day. We also find that monetary policy surprises in Brazil are more likely 12

13 to induce jumps in interest rates in comparison to the European evidence reported by León and Sebestyén (2012). References Balduzzi, Pierluigi, Bertola, Giuseppe, Foresi, Silverio, and Klapper, Leora Interest Rate Targeting and the Dynamics of Short-Term Rates. Journal of Money, Credit and Banking, 30(1), Beber, Alessandro, and Brandt, Michael W When It Cannot Get Better or Worse: The Asymmetric Impact of Good and Bad News on Bond Returns in Expansions and Recessions. Review of Finance, 14(1), Bohl, Martin T, Siklos, Pierre L, and Sondermann, David European Stock Markets and the ECB s Monetary Policy Surprises. International Finance, 11(2), Buchholz, Anna, Cupertino, Cesar, Meurer, Roberto, Santos, Andre Portela, and Da Costa Jr, Newton The market reaction to changes in the Brazilian official interest rate. Applied Economics Letters, 19(14), Chan, Wing H, and Maheu, John M Conditional jump dynamics in stock market returns. Journal of Business and Economic Statistics, 20(3), Cook, Timothy, and Hahn, Thomas The effect of changes in the federal funds rate target on market interest rates in the 1970s. Journal of Monetary Economics, 24(3), Dale, Spencer The effect of changes in official UK rates on market interest rates since The Manchester School, 61(S1), Das, Sanjiv R The surprise element: jumps in interest rates. Journal of Econometrics, 106(1), Estrella, Arturo, and Hardouvelis, Gikas A The term structure as a predictor of real economic activity. The Journal of Finance, 46(2), Haldane, Andrew G, and Read, Vicky Monetary policy surprises and the yield curve. Bank of England Working Paper. Hardy, Daniel C Anticipation and surprises in central bank interest rate policy: The case of the Bundesbank. Staff Papers-International Monetary Fund, Johannes, Michael The Statistical and Economic Role of Jumps in Continuous-Time Interest Rate Models. The Journal of Finance, 59(1), Kuttner, Kenneth N Monetary policy surprises and interest rates: Evidence from the Fed funds futures market. Journal of monetary economics, 47(3), León, Ángel, and Sebestyén, Szabolcs New measures of monetary policy surprises and jumps in interest rates. Journal of Banking and Finance, 36(8), Maheu, John M, and McCurdy, Thomas H News arrival, jump dynamics, and volatility components for individual stock returns. The Journal of Finance, 59(2),

14 Marçal, Emerson Fernandes, and Pereira, PL Valls A Estrutura a Termo das Taxas de Juros no Brasil: testando a hipóteses de Expectativas. Pesquisa e Planejamento Econômico, 37(1), Oliveira, Fernando N, and Ramos, Leonardo Choques não antecipados de política monetária ea estrutura a termo das taxas de juros no Brasil. Pesquisa e Planejamento Econômico, 41(3). Rangel, José Gonzalo Macroeconomic news, announcements, and stock market jump intensity dynamics. Journal of Banking and Finance, 35(5), Roley, V Vance, and Sellon, Gordon H Monetary policy actions and long-term interest rates. Federal Reserve Bank of Kansas City Economic Quarterly, 80(4), Tabak, Benjamin Miranda A note on the effects of monetary policy surprises on the Brazilian term structure of interest rates. Journal of Policy Modeling, 26(3),

15 Tables Table 1: Summary statistics for the target and DI-Pre swap rates and COPOM decisions The table reports mean and variance for the target rate (Meta Selic) and DI-Pre swap rates with maturities of 30, 60, 90, 120, and 360 days from January 2003 to June 2014 (2767 observations). Summary statistics are reported for the rates in both levels and daily variations. The table also reports the number of instances in which COPOM decisions are to increase, decrease or to have the target rate unchanged along with the average increase or decrease in the target rate (in percentage points). Target rate 30-day 60-day 90-day 120-day 180-day 360-day Levels Mean Variance Daily variations Mean Variance COPOM decisions for the target rate Increase Decrease Unchanged # of decisions Average change in the target rate

16 Table 2: GARCH-jump model with constant jump intensity The table reports key parameter estimates for the GARCH-jump model with constant jump intensity. t-statistics appear in parentheses and LogL denotes the value of the likelihood function evaluated at the estimated parameters. The data consists of time series of daily closing yields of the DI-Pre swap rates with maturities of 30, 60, 90, 120, and 360 days from January 2003 to June 2014 (2767 observations). The model is defined as Δr t = μ t + ε 1,t + ε 2,t, where μ t = α 0 + p α 1Δr t 1 with p = 1, ε 1,t = σ tz t, ε 2,t = n t k=1 J t,k μ jλ t, z t N (0, 1), J t N (μ j, σ 2 j ), and n t is a Poisson process with constant jump intensity λ. σ t follows a variance-targeting GARCH(1,1) process, σ 2 t = (1 α β) σ 2 + αε 2 t 1 + βσ 2 t 1. It is assumed that z t, J t, and n t are independent. The # of jumps is the number of days on which the ex-post probability of at least one jump, Pr (n t 1 I t), is bigger than 0.5. The average jump contribution to total volatility is the average ratio of λ t(σ 2 j + μ 2 j) to the total conditional variance, Var(Δr t I t 1). The number of jumps on COPOM meeting days over all jumps is the ratio of the number of jumps on COPOM meeting days to the total number of jumps, and the # of jumps on COPOM meeting days over all meeting days is the ratio of the number of jumps on COPOM meeting days to the total number of meeting days. 30-day 60-day 90-day 120-day 180-day 360-day λ ( ) (3.13) (2.69) (16.94) (11.53) (4.39) μ j ( 8.58) (4.80) (0.13) (2.46) (1.09) (2.35) α (130.89) (2.69) (4.54) (2.26) (1.15) (2.96) β (129.15) (4.50) (7.06) (3.76) (4.10) (6.66) σj (63.76) (3.94) (2.30) (4.95) (2.03) (2.03) LogL Jump contribution to total volatility Number of jumps Number of jumps on COPOM meetings days over all meeting days Number of jumps on COPOM meetings days over all days i=1 16

17 Table 3: GARCH-jump model with time-varying jump intensity The table reports key parameter estimates for the GARCH-jump model with time-varying jump intensity along with likelihood ratio tests with respect to the GARCH-jump model with constant jump intensity. t-statistics appear in parentheses and LogL denotes the value of the likelihood function evaluated at the estimated parameters. The data consists in time series of daily closing yields of the DI-Pre swap rates with maturities of 30, 60, 90, 120, and 360 days from January 2003 to June 2014 (2767 observations). The model is defined as Δr t = μ t + ε 1,t + ε 2,t, where μ t = α 0 + p α 1Δr t 1 with p = 1, ε 1,t = σ tz t, ε 2,t = n t k=1 J t,k μ j,tλ t, z t N (0, 1), J t N (μ j,t, σ 2 j ), and n t is a Poisson process with time varying jump intensity defined as λ t = λ 0 +λ 1 S 1,t +λ 2 S 2,t where S 1 and S 2 denote the level and slope factors, respectively, obtained via principal component analysis (PCA). μ j,t = β 0 + β 1,i S i,t + β+ 2,i S+ i,t where i = 1, 2 stands for the level and slope factors, respectively, and S+ i,t max(s i,t, 0) and S i,t min(si,t, 0). σt follows a variance-targeting GARCH(1,1) process, σ2 t = (1 α β) σ 2 +αε 2 t 1 +βσ 2 t 1. It is assumed that z t, J t, and n t are independent. 30-day 60-day 90-day 120-day 180-day 360-day α ( ) (2.6963) (1.5832) (3.9544) (1.2029) (4.0672) β (4.0489) (3.3198) (2.0894) (5.0985) (1.9428) (8.9065) λ (1.0209) (6.8611) (0.0445) (0.9818) (0.3942) ( ) λ ( ) (4.2263) (2.5716) (1.9979) (0.7050) (0.4858) λ ( ) (0.0000) (0.0000) (0.6362) ( ) (0.0000) β ( ) ( ) ( ) ( ) ( ) ( ) β (4.9354) (6.3122) (1.1289) (0.1741) ( ) (0.3289) β ( ) ( ) ( ) ( ) (2.8285) ( ) β ( ) ( ) (3.2486) ( ) ( ) (1.2050) β ( ) ( ) (4.0787) ( ) ( ) ( ) σj ( ) ( ) (6.6547) (9.7591) (3.1176) (2.9515) LogL Likelihood ratio test p-value i=1 17

18 Table 4: Volatility and jump characteristics for the daily changes in DI-Pre swap rates The Table reports the main volatility and jump characteristics of the GARCH-jump model with time-varying jump intensity and time-varying jump mean described in Section 3.2. The number of jumps is the number of days on which the ex-post probability of at least one jump, Pr (n t 1 I t), is bigger than 0.5. The average jump contribution to total volatility is the average ratio of λ t(σ 2 j + μ 2 j,t) to the total conditional variance, Var(Δr t I t 1). 30-day 60-day 90-day 120-day 180-day 360-day Average total volatility All days Meeting days Non-meeting days Average jump contribution to total volatility All days Meeting days Non-meeting days λ t = E(n t I t 1 ) All days Meeting days Non-meeting days P r(n t 1 I t ) All days Meeting days Non-meeting days E(n t I t ) All days Meeting days Non-meeting days Number of jumps All days Meeting days Non-meeting days