A brief analysis of aggregate measures as an alternative to the median at Central Bank of Brazil's Survey of Professional Forecasts

de Carvalho, Fabia A. Article A brief analysis of aggregate measures as an alternative to the median at Central Bank of Brazil's Survey of Professional Forecasts EconomiA Provided in Cooperation with: The Brazilian Association of Postgraduate Programs in Economics (ANPEC), Rio de Janeiro Suggested Citation: de Carvalho, Fabia A. (2013) : A brief analysis of aggregate measures as an alternative to the median at Central Bank of Brazil's Survey of Professional Forecasts, EconomiA, ISSN 1517-7580, Elsevier, Amsterdam, Vol. 14, Iss. 2, pp. 11-26, https://doi.org/10.1016/j.econ.2013.08.001 This Version is available at: https://hdl.handle.net/10419/179544 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/3.0/ Available online at www.sciencedirect.com ScienceDirect EconomiA 14 (2013) 11–26 A brief analysis of aggregate measures as an alternative to the median at Central Bank of Brazil’s Survey of Professional Forecasts夽 Fabia A. de Carvalho Research Department, Central Bank of Brazil Abstract This paper presents a brief analysis of representative measures of inflation expectations from Central Bank of Brazil’s Survey of Professional Forecasts that are alternative to the median response. We build time series with the mode and core measures of inflation expectations from the panel of professional forecasts surveyed from January 2002 to September 2012. We compare them to the median response with respect to their predictive power in a 12-month-ahead horizon. We also compare the predictive power of the alternative measures with the realized core of consumer price inflation. © 2013 National Association of Postgraduate Centers in Economics, ANPEC. Production and hosting by Elsevier B.V. All rights reserved. JEL classification: E37; E58 Keywords: Inflation expectations; Market forecasts; Median; Core inflation; Trimmed core inflation; Brazil Resumo Esse artigo apresenta uma breve análise de medidas representativas alternativas à mediana da pesquisa de expectativas conduzida pelo Departamento de Relacionamento com Investidores e Estudos Especiais (Gerin) do Banco Central do Brasil. A partir do painel de expectativas coletadas de janeiro de 2002 a setembro de 2012, construímos séries históricas de núcleos e da moda das expectativas para a inflac¸ão, que, a seguir, são comparadas com a mediana em relac¸ão ao seu poder preditivo para o horizonte de 12 meses à frente. Comparamos também o poder preditivo dessas medidas com respeito ao núcleo de inflac¸ão do IPCA. © 2013 National Association of Postgraduate Centers in Economics, ANPEC. Production and hosting by Elsevier B.V. All rights reserved. Palavras-chave: expectativas de mercado; expectativas de inflac¸ão; mediana; núcleos aparados; Brasil 夽The author is thankful to Adriana Soares Sales and Andre Minella for comments and suggestions. Obviously, errors and omission are the author’s sole responsibility. The views expressed in this work do not necessarily reflect those of the Central Bank of Brazil. Peer review under responsibility of Peer review under responsibility of National Association of Postgraduate Centers in Economics, ANPEC. E-mail address: [email protected] Production and hosting by Elsevie r ELSEVIER 1517-7580 © 2013 National Association of Postgraduate Centers in Economics, ANPEC. Production and hosting by Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.econ.2013.08.001 12 F.A. de Carvalho / EconomiA 14 (2013) 11–26 1. Introduction Market forecasts surveyed by the Central Bank of Brazil since 1999 are reported daily at the institution’s website and published weekly at the “Focus” reports. They have become an important reference to the discussion of macroeconomic prospects in Brazil, especially with respect to inflation. The univariate measures that are usually chosen to represent the panel of forecasts both at the “Focus” reports and at the Inflation Report (Chapter 6) are the median forecasts and their standard deviations.1Throughout the rest of the world, the use of the panel median or mean is also widely disseminated.2 Carvalho and Minella (2012) present a detailed study of the predictive power of the median forecasts surveyed by the Central Bank of Brazil for a 12-month-ahead horizon. They show that in the analyzed period, the median response does not present systematic bias, which implies a reasonable predictive power, in spite of their failure in efficiency tests. Other studies have investigated the predictive power of the median responses of Central Bank of Brazil’s survey for varied forecast horizons.3 This paper assesses the predictive power of other measures representative of the panel of inflation forecasts surveyed by the Central Bank of Brazil. In particular, we build series of modes and core measures of inflation forecasts for a 12-month-ahead horizon. Except for the symmetric trimmed mean core and for one of the asymmetric core measures, all measures that we investigate are statistically different from the median response. In terms of predictive power, all measures present systematic bias in the complete sample. The evidence of bias is slightly smaller for the median. However, this conclusion does not hold for shorter sub-samples. Furthermore, for the complete sample, the investigated measures are more appropriate proxies of the smoothed trimmed core inflation index than of the actual headline consumer price inflation. Except for the mode, the investigated measures do not present systematic bias when compared to the core inflation. Notwithstanding, for subsamples beginning in January 2003 or January 2004, which are less contaminated by the confidence crisis that had hit the economy in previous years, all measures present systematic bias when compared to the inflation core. 2. Building measures that are representative of the panel of inflation forecasts Aggregate measures such as the mean, median and standard deviation of the expectations panel surveyed by the Central Bank of Brazil, and others derived from these three, are reported daily at the central bank’s website (https://www3.bcb.gov.br/expectativas/publico/en/serieestatisticas). The survey currently encompasses over 100 registered participants.4 We used the complete database of the survey, from January 2002 to September 2012, to build five core inflation expectations series in addition to a series of modes. The methodology is detailed in what follows. For all series, the data refer to forecasts surveyed at each day of the month corresponding to the day previous to that used to produce the Top-5 rank published by the Investor’s Relations Office of the Central Bank of Brazil. The forecast horizon considered was 12-months, accumulated from the month following the survey date onwards. 1More recently, in the Inflation Report of March 2011, the Central Bank of Brazil started to publish the median of selected segments of survey participants. 2For instance, the Inflation Perspectives chapter of Bank of England’s Inflation Report reports the mean expectations of a group of surveyed professionals. The mean is also the representative measure chosen to report the Macro Markets Home Price Expectations Survey, as well as Consensus Economics forecasts, which, in turn, also reports the individual forecasts. US Michigan Survey of Expectations reports the median response as its representative measure. 3Kohlscheen (2010), Guillén (2008) and Carvalho and Bugarin (2006), for instance. 4For a complete description of the survey’s database, please refer to Marques et al. (2003). F.A. de Carvalho / EconomiA 14 (2013) 11–26 13 The first core measure built for this study was the symmetric trimmed mean (Fig. 1 and Appendix A). Its computation involved ordering all projections according to their magnitude at each sampling day, and excluding those placed in the outer 10% ranges. The remaining 80% of the individual forecasts were used to calculate the mean. Fig. 1. Symmetric trimmed mean core of 12-month-ahead inflation expectations. Second, we built an asymmetric median and mean core of inflation expectations. To this end, we carried out two asymmetry tests at each surveyed date: one based on Pearson’s asymmetry coefficient5and another based on the third moment of the sampling distribution.6In both tests, distributions are classified as asymmetric when the absolute value of the resulting asymmetry coefficient is larger than 0.3. The results of this initial identification test of asymmetry in the expectations series are reported in Fig. 2. The direction of the asymmetry does not always coincide in both tests. In fact, there was contradiction in 40% of the sample. -2.55 -1.55 -0.55 0.45 1.45 2.45 3.45 Jan-02 Jul-02 Jan-03 Jul-03 Jan-04 Jul-04 Jan-05 Jul-05 Jan-06 Jul-06 Jan-07 Jul-07 Jan-08 Jul-08 Jan-09 Jul-09 Jan-10 Jul-10 Jan-11 Jul-11 Jan-12 Jul-12 Survey date Pearson Asymme try Coef .Standard Asymme try Test Fig. 2. Asymmetry tests of the 12-month-ahead inflation expectations distributions. 5Pearson’s Asymmetry coefficient = ((Mode − Mean)/(Standard Deviation)). The mode was computed according to the methodology described in this paper. 6Asymmetry(x) =E(x−Mean(x))3 (Std(x))3. 14 F.A. de Carvalho / EconomiA 14 (2013) 11–26 After determining whether the distribution of expectations at each survey date is symmetric or not according to each type of asymmetry test, we removed the outliers as follows: • If the distribution was found to be asymmetric, we removed the 2.5% smallest and the 2.5% highest values of the sample at each survey date; • If the distribution was asymmetric to the left (i.e., mode < mean), we removed the 5% highest values of the sample at each survey date; • If the distribution was asymmetric to the right (i.e., mode > mean), we removed the 5% lowest values of the sample at each survey date. This asymmetric trimming methodology is used by the Central Bank of Brazil to calculate the daily average base rate (Selic), aiming at eliminating observations that are less representative of the aggregate forecast and which might bias the sample mean. There is an important degree of arbitrariness in the construction of asymmetric core measures. First, the size of the trim (5%) in the distribution tails, regardless of the degree of asymmetry found, does not necessarily imply that the remaining distribution will be void of asymmetry. Second, the methodology requires computation of the sample mode, which also bears an important degree of arbitrariness. The resulting asymmetric core series are presented in Fig. 3 and in Appendix A. Fig. 3. Asymmetric core measures of 12-month-ahead inflation expectations. F.A. de Carvalho / EconomiA 14 (2013) 11–26 15 Finally, we built a series of modes for each survey date (Fig. 4 and Appendix A). The mode is more representative of a “consensus” measure than the median. However, its computation is not straightforward. To compute the mode, we first built distribution histograms of the forecasts at each survey date. Next, we identified the mean point of the interval with the highest concentration of forecasts. This calculation, however, is sensitive to the size of the bin chosen to slice the sample. The choice of the size of the bin for each survey date was arbitrary, and had the purpose of obtaining only one modal bin. Fig. 4. Mode of 12-month-ahead inflation expectations. 3. Comparing the alternative measures to the median and testing its predictive power We carried out statistical tests to investigate whether the alternative measures representative of inflation forecasts were statistically different from the median. These tests are inspired in the unbiasedness tests traditionally used in the literature (e.g., Marimon and Sunder, 1993; Zarnowitz, 1985; Keane and Runkle, 1990). The tests consist of assessing the joint null H0: c(1) = 0 and c(2) = 1 in the equation: Alternative measure = c(1) + c(2) × Median + White noise (1) Rejecting H0 implies that the alternative measure under evaluation is statistically different from the median. The results are reported in Tables A1 and A3 in Appendix A. The series of symmetric trimmed means is statistically indistinguishable from the series of medians. With respect to the asymmetric trimmed mean and median cores, the statistical tests point to important differences between the core series and the median. The only exception is one of the core series obtained from the Pearson coefficient. With respect to the mode, when we built the histograms, we noticed that the shape of the forecast distribution is highly variable over the sampled period, presenting great asymmetry at certain moments. The tests indicate a significant difference between the mode and the median. Comparing the alternative measures with the realized value of consumer inflation (IPCA), we tested the null H0: c(1) = 0 at the equation Forecast bias of the alternative measure = c(1) + noise(MA(12)) (2) where the bias present in the alternative measure corresponds to the difference between the headline consumer inflation and the considered alternative measure of inflation forecasts. Rejecting H0 implies that there is evidence of bias in the forecasts. In addition, we used a Newey–West covariance matrix with MA(12)7errors, as suggested by Keane and Runkle (1990), since the forecast errors for a 12-month-ahead horizon accumulate along these months in face of unexpected shocks. 7For further details on the reasons for using the correction in the covariance matrix for these tests, please refer to Carvalho and Minella (2012). 16 F.A. de Carvalho / EconomiA 14 (2013) 11–26 The predictive power of inflation expectations can be measured by the p-value obtained in the unbiasedness tests. The lower the p-value, the stronger the evidence of systematic forecast bias. Table 1 presents the p-values of the unbiasedness tests for 12-month-ahead inflation forecasts compared with the realized headline consumer inflation. The tests use data up to December 2011 since market expectations surveyed at that date refer to realized inflation 12 months ahead, i.e., accumulated until December 2012, which corresponds to the last date for which actual inflation data was available when this paper was prepared. Table 1 P-value of the unbiasedness tests for 12-month-ahead inflation expectations. 2002:1 to 2011:12 2004:1 to 2011:12 2003:1 to 2011:12 MedianComplete Panel 0.015 0.016 0.120 Mean 0.008 0.018 0.167 Mode 0.012 0.014 0.081 Trimmed MeanSymmetriccore 0.008 0.016 0.143 trimming 10% of each tail) Asymmetric core Trimmed MeanPearson Coef. 0.009 0.017 0.138 Trimmed Median 0.017 0.017 0.115 Trimmed Mean3rdmomentcriterium 0.008 0.023 0.197 Trimmed Median 0.012 0.019 0.138 The results of the unbiasedness tests show that, for the complete sample, all investigated measures present systematic bias. When we select sub-samples that exclude one or another crisis period, the results change. In Carvalho and Minella (2012), a chosen sub-sample started in January 2004, which excludes the effects of a crisis of confidence in the future conduct of monetary policy after a leftist presidential candidate was elected back in 2002. Since the data sample used in that work went until 2007, the authors could not find any indication of bias in the median inflation expectation for that subperiod. However, in 2007 and 2010, there were important forecast errors, and the tests considered in this paper still point to a systematic bias in all investigated statistics, even if we exclude the confidence crisis. If we restrict the sample to begin in January 2003, when the forecast errors were strongly negative, in average these errors cancel out with the positive errors of the following year, and the tests reject the null of a systematic bias. 4. Are inflation expectations better indicators of the headline consumer inflation or of the core inflation? Ranchhod (2003) carries out exercises to verify the predictive power of inflation expectations surveyed in New Zealand. One of the results obtained is that, even when survey participants forecast headline inflation, their forecasts are a more adequate representation of smoothed measures of inflation, such as exclusion core indices. The reason seems to be that inflation of more volatile items in the consumer price index is more difficult to be anticipated. Inspired by that work, we compared inflation expectations for the headline inflation in Brazil with actual values of the smoothed and trimmed mean core index for consumer price inflation. The results are presented at Table A4 in Appendix A. In the complete sample,8the unbiasedness tests do not indicate systematic bias in forecasts when compared with the core inflation. The only exception to that was the mode. However, this result is strongly affected by counterbalance of the sizable positive forecast errors at the beginning of the series with the sizable negative forecast errors observed in 2006 and 2008. If we begin the tests in January 2004, all investigated statistics show an important forecast bias. 5. Concluding remarks This brief paper shows that the mode of inflation expectations for a 12-month-ahead horizon and a great number of asymmetric core measures present important differences with respect to the median inflation expectation. 8We did not carry out tests for the trimmed median or mean core. F.A. de Carvalho / EconomiA 14 (2013) 11–26 17 In the complete sample, which includes years in which the crisis of confidence in the future conduct of economic policy in Brazil affected most noticeably the predictive power of inflation expectations from professional forecasters, all analyzed measures (median, symmetric trimmed cored, asymmetric core and mode) present systematic forecast bias for the headline consumer inflation. However, in the sub-sample that begins in January 2003, the analyzed inflation forecasts cease to present bias, likely due not to an improvement in predictive capacity, but to statiscal cancelling out of positive through negative errors. In fact, the choice of subsample influences the result. Inspired in Ranchhod (2003), we carry out tests to check whether the forecasts made for the headline inflation are more appropriate representations of a less volatile measure of inflation, such as the smoothed trimmed core consumer inflation. Contrary to the unbiasedness tests for the headline inflation, there is no indication of systematic bias in the analyzed measures (with the exception of the mode using a 95% confidence interval) when we compare the inflation expectations with the core inflation. However, for subsamples beginning in January 2003 or January 2004, the tests indicate forecast bias. Appendix A. See Tables A1–A5 Table A1 Test of statistical difference between the symmetric trimmed mean and the median 12-month-ahead inflation expectations. Dependent variable (Y): symmetric trimmed mean core of inflation Expectations Sample: January 2002 to September 2012 Number of observations: 129 Equation: Y = c(1) + c(2) × Median of infl. expectations Coefficient Std t-ratio P-value c(1) −0.037 0.020 −1.840 0.068 c(2) 1.007 0.004 273.266 0.000 R20.998302 Mean dependent var 5.216861 Adjusted R20.998289 SD dependent var 1.357727 Regression Std 0.056165 Akaike −2.90568 SSR 0.400616 Schwarz −2.861342 Log likelihood 189.4163 Hannan–Quinn −2.887664 F-statistics 74674.53 Durbin–Watson 2.187917 Prob (F-statistics) 0 Wald test: c(1) = 0, c(2) = 1 Test statistic Value dF Probability F-statistics 1.727744 (2, 127) 0.1818 Chi-square 3.455488 2 0.1777 18 F.A. de Carvalho / EconomiA 14 (2013) 11–26 Table A2 Tests of Statistical difference between the asymmetric trimmed mean and median core and the median inflation expectations. Dependent variable (Y): asymmetric trimmed mean core of inflation Expectations (Pearson coeff.) Sample: January 2002 to September 2012 Number of observations: 129 Equation: Y = c(1) + c(2) × Median expectations Coefficient Std t-ratio P-value c(1) 0.000 0.022 −0.002 0.999 c(2) 1.000 0.004 245.890 0.000 R20.997904 Mean dependent var 5.216725 Adjusted R20.997887 SD dependent var 1.348531 Regression Std 0.061982 Akaike −2.708548 SSR 0.487912 Schwarz −2.66421 Log likelihood 176.7013 Hannan–Quinn −2.690532 F-statistics 60462.02 Durbin–Watson 1.80517 Prob (F-statistics) 0 Wald test: c(1) = 0, c(2) = 1 Test statistic Value dF Probability F-statistic 0.022819 (2, 127) 0.9774 Chi-square 0.045638 2 0.9774 Dependent variable (Y): asymmetric trimmed median core of inflation Expectations (Pearson coeff.) Sample: January 2002 to September 2012 Number of obs: 129 Equation: Y = c(1) + c(2) × Median expectations Coefficient Std t-ratio P-value c(1) 0.041 0.010 3.952 0.000 c(2) 0.992 0.002 517.150 0.000 R20.999525 Mean dependent var 5.215422 Adjusted R20.999522 SD dependent var 1.336546 Regression Std 0.029233 Akaike −4.211672 SSR 0.108528 Schwarz −4.167334 Log likelihood 273.6529 Hannan–Quinn −4.193657 F-statistics 267443.7 Durbin–Watson 1.867673 Prob (F-Statistics) 0 Wald test: c(1) = 0, c(2) = 1 Test Statistic Value dF Probability F-statistic 9.79399 (2, 127) 0.0001 Chi-square 19.58798 2 0.0001 Dependent variable (Y): asymmetric trimmed mean core of inflation Expectations (asymmetry measured by the 3rd moment) Sample: January 2002 to September 2012 Number of obs: 108 Equation: Y = c(1) + c(2) × Median Expectations Coefficient Std t-ratio P-value c(1) −0.189 0.036 −5.234 0.000 c(2) 1.041 0.007 155.116 0.000 . F.A. de Carvalho / EconomiA 14 (2013) 11–26 25 Table A5 (Continued ) Sep-02 6.3 Feb-08 4.5 Oct-02 6.1 Mar-08 4.2 Nov-02 6.0 Apr-08 4.1 Dec-02 5.9 May-08 4.0 Jan-03 5.7 Jun-08 4.0 Feb-03 5.4 Jul-08 4.1 Mar-03 5.4 Aug-08 4.1 Apr-03 5.5 Sep-08 4.2 May-03 5.8 Oct-08 4.3 Jun-03 6.1 Nov-08 4.4 Jul-03 6.1 Dec-08 4.5 Aug-03 6.2 Jan-09 4.5 Sep-03 6.2 Feb-09 4.5 Oct-03 6.2 Mar-09 4.5 Nov-03 6.1 Apr-09 4.7 Dec-03 5.9 May-09 4.9 Jan-04 5.7 Jun-09 4.9 Feb-04 5.5 Jul-09 5.1 Mar-04 5.4 Aug-09 5.2 Apr-04 5.5 Sep-09 5.2 May-04 5.2 Oct-09 5.2 Jun-04 5.1 Nov-09 5.3 Jul-04 5.1 Dec-09 5.4 Aug-04 4.9 Jan-10 5.5 Sep-04 4.8 Feb-10 5.4 Oct-04 4.6 Mar-10 5.4 Nov-04 4.5 Apr-10 5.2 Dec-04 4.5 May-10 5.0 Jan-05 4.5 Jun-10 5.3 Feb-05 4.4 Jul-10 5.5 Mar-05 4.2 Aug-10 5.5 Apr-05 4.1 Sep-10 5.6 May-05 4.1 Oct-10 5.6 Jun-05 4.3 Nov-10 5.5 Jul-05 4.5 Dec-10 5.3 Aug-05 4.5 Jan-11 5.3 Sep-05 4.2 Feb-11 5.4 Oct-05 4.0 Mar-11 5.5 Nov-05 4.1 Apr-11 5.5 Dec-05 4.0 May-11 5.5 Asymmetric trimmed mean core of inflation expectations (Pearson coeff.) Jan-01 4.9 Jun-06 3.5 Feb-01 4.7 Jul-06 3.6 Mar-01 4.7 Aug-06 3.7 Apr-01 4.4 Sep-06 3.8 May-01 4.3 Oct-06 3.9 Jun-01 4.5 Nov-06 3.9 Jul-Ol 4.6 Dec-06 4.2 Aug-01 5.2 Jan-07 4.3 Sep-01 5.6 Feb-07 4.2 Oct-01 7.9 Mar-07 4.3 Nov-01 10.1 Apr-07 4.4 Dec-01 11.2 May-07 4.8 Jan-02 10.6 Jun-07 5.1 Feb-02 9.9 Jul-07 5.3 Mar-02 9.3 Aug-07 5.3 Apr-02 8.5 Sep-07 5.2 May-02 8.2 Oct-07 5.2 Jun-02 7.5 Nov-07 5.4 26 F.A. de Carvalho / EconomiA 14 (2013) 11–26 Table A5 (Continued ) Jul-02 6.8 Dec-07 5.1 Aug-02 6.4 Jan-08 4.8 Sep-02 6.2 Feb-08 4.5 Oct-02 6.1 Mar-08 4.3 Nov-02 6.0 Apr-08 4.2 Dec-02 5.9 May-08 4.0 Jan-03 5.8 Jun-08 4.1 Feb-03 5.4 Jul-08 4.1 Mar-03 5.4 Aug-08 4.1 Apr-03 5.5 Sep-08 4.2 May-03 5.7 Oct-08 4.3 Jun-03 5.9 Nov-08 4.4 Jul-03 6.0 Dec-08 4.5 Aug-03 6.1 Jan-09 4.5 Sep-03 6.1 Feb-09 4.4 Oct-03 6.1 Mar-09 4.6 Nov-03 6.0 Apr-09 4.7 Dec-03 5.8 May-09 4.9 Jan-04 5.6 Jun-09 4.9 Feb-04 5.4 Jul-09 5.0 Mar-04 5.4 Aug-09 5.1 Apr-04 5.4 Sep-09 5.2 May-04 5.2 Oct-09 5.1 Jun-04 5.1 Nov-09 5.2 Jul-04 5.1 Dec-09 5.4 Aug-04 5.0 Jan-10 5.5 Sep-04 4.8 Feb-10 5.3 Oct-04 4.5 Mar-10 5.3 Nov-04 4.6 Apr-10 5.3 Dec-04 4.5 May-10 5.0 Jan-05 4.5 Jun-10 5.2 Feb-05 4.4 Jul-10 5.5 Mar-05 4.2 Aug-10 5.5 Apr-05 4.1 Sep-10 5.7 May-05 4.2 Oct-10 5.7 Jun-05 4.4 Nov-10 5.5 Jul-05 4.5 Dec-10 5.3 Aug-05 4.5 Jan-11 5.3 Sep-05 4.3 Feb-11 5.3 Oct-05 4.0 Mar-11 5.4 Nov-05 4.2 Apr-11 5.5 Dec-05 4.1 May-11 5.5 References Carvalho, F., Minella, A., 2012. Survey forecasts in Brazil: a prismatic assessment of epidemiology, performance, and determinants. J. Int. Money Finance 31 (October (6)), 1371–1391. Carvalho, F., Bugarin, M., 2006. Inflation expectations in Latin America. Economía (Washington) 2006, 101–145. Guillén, D., 2008. Ensaios sobre Expectativas de Inflac¸ão no Brasil, Tese de Mestrado. PUC-Rio. Keane, M., Runkle, D., 1990. Testing the rationality of price forecasts: new evidence from panel data. Am. Econ. Rev. 80, 714–735. Kohlscheen, 2010. Uma Nota sobre Erros de Previsão da Inflac¸ão de Curto Prazo, Trabalho para Discussão No. 227, Banco Central do Brasil, Novembro. Marques, A., Fachada, P., Cavalcanti, D., 2003. Sistema Banco Central de Expectativas de Mercado, Nota Técnica No. 36. Banco Central do Brasil, Maio. Marimon, R., Sunder, S., 1993. Indeterminacy of equilibria in a hyperinflationary world: experimental evidence. Econometrica 61 (5), 1073–1107. Ranchhod, S., 2003. The relationship between inflation expectations survey data and inflation”. Reserve Bank N.Z. Bull. 66 (4), 50–65. Zarnowitz, V., 1985. Rational expectations and macroeconomic forecasts. J. Bus. Stat. 3, 293–311.