Forecasting the yield curve: A statistical model with market survey data André Luís Leite a , Romeu Braz Pereira Gomes Filho a , José Valentim Machado Vicente b, ⁎ a Central Bank of Brazil, Brazil b Faculdades Ibmec-RJ and Central Bank of Brazil, Av. Presidente Vargas 730 7th Floor, Rio de Janeiro, Brazil abstract article info Article history: Received 12 June 2009 Received in revised form 1 December 2009 Accepted 1 February 2010 Available online 10 February 2010 JEL classification: G1 E4 C5 Keywords: Yield curve forecasting Risk premium Market surveys In this paper we propose a statistical model to forecast the yield curve, using two major sources of information: data from a market survey and the forward rate risk premium. We apply the model to forecast the Brazilian yield curve six months ahead and compare the results with the well-known model of Diebold and Li (2006), a random walk process and the predictions based on the forward rate. The proposed model produces accurate forecasts and outperforms all the competitor models in terms of root mean square error (RMSE). © 2010 Elsevier Inc. All rights reserved. 1. Introduction A tool to predict the yield curve is undoubtedly of great worth to financial market analysts and investors. However, despite of the relevance of this issue, few practical improvements have been made in recent years. In this study we propose a parsimonious technique to model the term structure of interest rates that relies on two components: market survey data and the forward rate risk premium. Then we use the proposed model to forecast the Brazilian domestic yield curve using public data available on a daily basis. To the best of our knowledge, there is no other article in the finance literature that works with this approach applied to an emerging country. 1 Research into term structure of interest rates (TSIR) basically rests on two classes of models, usually known as statistical models and equilibrium models. In the first group the TSIR is constructed through an interpolation process and forecasts are done using time series models. In the second group, the models incorporate equilibrium arguments, such as no-arbitrage, to analyze the TSIR, and forecasts are produced by the dynamics implied in the model. Despite the lack of economic theory grounds, statistical models are preferred in practical problems due to their lesser estimation complexity. 2 To attain computational simplicity, we use a statistical model in this paper. 3 Macroeconomic variables have been frequently used to analyze the dynamics of interest rates. The Fischer equation (Fisher, 1930) and the Taylor rule (Taylor, 1993) help make price indexes the main macroeco- nomic variable used to model the TSIR, since they specify a direct relation between inflation and interest rates. 4 On the other hand, some authors find that market surveys are powerful predictors of future inflation (see, for instance, Ang, Bekaert, & Wei, 2007; Mehra, 2002). To combine these two features in single variable we use the inflation expectation calculated by the Central Bank of Brazil (CBB) from a survey among professional forecasters as the explanatory variable for future interest rates. Campbell and Shiller (1991), Cochrane and Piazzesi (2005), Dai and Singleton (2002), and Fama and Bliss (1987) analyze the failure of the expectation hypothesis and the importance of time-varying risk premia. Ludvigson and Ng (2007) find evidence that the interest rate International Review of Financial Analysis 19 (2010) 108–112 ⁎ Corresponding author. Tel.: + 55 21 2189 5762. E-mail addresses: andreluis.leite@bcb.gov.br (A.L. Leite), romeu.gomes@bcb.gov.br (R.B.P.G. Filho), jose.valentim@bcb.gov.br (J.V.M. Vicente). 1 Some recent papers deal with the prediction of the Brazilian yield curve. They all use different approaches from ours. Among others, we can cite Almeida, Gomes, Leite, and Vicente (2009), Lima, Luduvice, and Tabak (2006), and Vicente and Tabak (2008). 2 Although Almeida and Vicente (2008) and Christensen, Diebold, and Rudebusch (2008) present evidence in favor of the inclusion of no-arbitrage conditions when the goal is to forecast interest rates, this issue is not without controversy, as shown by Duffee (2008). 3 Although they are simpler to estimate, pure econometric models based on VAR or ARMA processes have low predictive power of future interest rates, as shown by Diebold and Li (2006) and Lima et al. (2006). 4 Among other works that use price indexes to model the term structure, we can cite Ang and Piazzesi (2003), Diebold, Piazzesi, and Rudebusch (2005), Diebold, Rudebusch, and Aruoba (2006), Hördahl, Tristani, and Vestin (2006) and Huse (2007). 1057-5219/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.irfa.2010.02.001 Contents lists available at ScienceDirect International Review of Financial Analysis