An investigation of linear and non-linear forecasting models for financial data

This thesis focuses on the statistical properties of linear and nonlinear forecasting models that can be applied to financial data. These properties were examined through analysis of approaches that bond investors use to form expectations of inflation measured on a daily basis using continuously compounded percent returns on the Lehman Treasury bond market index and on the Center of Research Bureau spot and futures commodity price index for the period from January 3, 1983 to June 30, 2008. The strength of the relationships was additionally investigated using constant maturity Treasury bonds of different maturities ranging from 3 months to 30 years.The present thesis argues that bond holders watch closely spot commodities markets for early indications of price changes in the aggregate terms. The thesis also argues that the relationships have been stable even after considering the effect of external economic environment on bond market returns expressed as shocks volatility. Significant non-linear effects were linked to unusual daily market moves beyond 4 standard deviations rather than implicit non-linear nature of the relationships.Partial evidence was found to support the argument that similar unidirectional relationships from spot commodities to bonds would hold for U.S. Treasury bonds across all maturities. Daily movements in the commodity spot market lead prices of 10- and 30-year government bonds. Treasury securities do not influence spot commodity markets. Traders of commodity futures use bond prices of the previous ten trading days in forming inflation expectations. These unidirectional relationships were found to be strong for government bonds across all maturities. Commodity futures contain less information about inflationary expectations than government bonds.