Aplicação de redes neurais na precificação de debêntures

Resumo

Previous studies on pricing of Corporate Bonds have shown that prices for these securities in Brazil cannot be explained only by credit risk, but also by other factors, such as liquidity risk. On the other hand, other studies also have shown that neural networks models have been more successful than traditional models in explaining issues related to corporate bonds, such as modeling default probabilities and ratings from agencies such as Standard & Poors and Moodys. The purpose of this study is to test neural networks technique in pricing corporate bonds in Brasil and compare the results obtained with the ones obtained through linear regressions. To accomplish this, accounting variables and specific features of a bond such as time to maturity and calllable features were used as independent variables. Regarding dependent variables, ANDIMA’s daily rates were used as a reference for market value for corporate bonds. The variables described above were tested in several models through ordinary least squares and the model which presented the best result was also tested in neural networks with two hidden layers. The neural networks with six and eight neurons presented better results than models estimated through pooling and ordinary least squares both in the training stage as in the testing one. Nonetheless, there’s still much room for improvement in the models considering the size of the database available is still small and the rates published by ANDIMA are averages of a small group of financial institutions and may not reflect the true market value of a corporate bond.

Ficha do documento

Tipo

Dissertação

Ano

2008

Instituição

Fundação Getulio Vargas

Fonte

Repositório da FGV

Idioma

Português

Acesso

Acesso aberto

Identificador

oai:repositorio.fgv.br:10438/2046

Temas

Economia

Palavras-chave

FinançasDebênturesCréditosRedes neurais (Computação)Avaliação de riscosEconomiaCréditos - Avaliação de riscos