Abstracts Category : Other

Fixed-income portfolio optimization based on dynamic Nelson-Siegel models with macroeconomic factors for the Brazilian yield curve

by Richard Schnorrenberger

Abstract : The study investigates the statistical and economic value of forecasted yields generated by dynamic yield curve models which incorporate a large macroeconomic dataset. The analysis starts of by modeling and forecasting the term structure of the Brazilian nominal interest rates using several specifications for the dynamic Nelson-Siegel (DNS) framework, suggested by Diebold & Li (2006). The first exercise concerns the incorporation of macro factors extracted from a large macroeconomic dataset, including forward-looking variables, to compare the forecast performance between some macroeconomic representations of the DNS model and itself. The results for forecast horizons above three months support the evidence for the incorporation of one macro factor that summarizes broad macroeconomic information regarding mainly inflation expectations. The conclusion that macroeconomic information tends to improvement in yield curve forecasting extend results found in previous literature. In order to assess the economic value of those forecasted yields, a fixed-income portfolio optimization using the mean-variance approach of Markowitz (1952) is performed. The analysis indicate that good yield curve predictions are important to achieve economic gains from forecasted yields in terms of portfolio performance. Preferred forecasted yields for short forecast horizons perform quite well for optimal mean-variance portfolios with one-step-ahead estimates for fixed-income returns, while forecasted yields generated by a macroeconomic DNS specification outperforms in terms of portfolio performance with twelve-step-ahead estimates. Therefore, there is an economic and statistical gain from considering a large macroeconomic dataset to forecast the Brazilian yield curve dynamics, specially for longer forecast horizons and for medium- and long-term maturities. ; O estudo investiga o valor estatstico e econmico dos rendimentos previstos por modelos dinmicos da curva de juros que incorporam um grande conjunto de dados macroeconmicos. A anlise parte da modelagem e previso da estrutura a termo das taxas de juros nominais brasileiras, usando diversas especificaes para o modelo dinmico de Nelson-Siegel (DNS), sugerido por Diebold & Li (2006). O primeiro exerccio diz respeito incorporao de macro-fatores extrados de um grande conjunto de dados macroeconmicos, incluindo variveis de expectativas, para comparar o desempenho de previso entre algumas representaes macroeconmicas do modelo DNS e ele mesmo. Os resultados para horizontes de previso acima de trs meses apoiam a evidncia para a incorporao de um fator macro que resume principalmente informaes gerais sobre expectativas de inflao. A concluso de que informao macroeconmica tende a aprimorar a previso da curva de juros estende os resultados encontrados na literatura recente. Para avaliar o valor econmico dos rendimentos previstos, realizada uma otimizao de carteira de renda fixa usando a abordagem de mdia-varincia de Markowitz (1952). A anlise indicaAdvisors/Committee Members: Moura, Guilherme Valle (advisor).