| This thesis consists of two parts. The first part develops a new method of estimating multi-parameter term-structure models using panel data. This technique involves recursively estimating some parameters along the cross-sectional dimension and the rest of the parameters along the time-series dimension until convergence is achieved. By breaking down the parameter estimation problem into two simpler problems along the time-series and cross-sectional dimensions, we are able to isolate and solve common problems plaguing other common methods such as quasi maximum likelihood estimation via the Kalman filter. We apply this technique successfully to the 1- and 2-factor Vasicek and Cox, Ingersoll, Ross models using both Eurodollar futures and Fama-Bliss Treasury data. Simulation results indicate that our technique yields reasonable parameter estimates for these models. The second part of this thesis presents and estimates a dynamic arbitrage-free model for the forward curve. The model combines features of the Preferred Habitat model, the Expectations Hypothesis model and Affine Yield Curve models. It decomposes the instantaneous forward rate curve for any particular day into the sum of three components: the unconditional or steady state forward curve, maturity-specific deviation from the unconditional curve and date-specific deviation from the unconditional curve. We propose a class of arbitrage-free models that parameterizes each of these component curves as a sum of several exponential functions. We then fit two of these specifications to Fama-Bliss Treasury data, as well as generate forecasts from the fits. For forecast horizons of 12-months or longer, the forecasts of these two models significantly outperform that of the benchmark models: the Random-Walk model, the Expectations Hypothesis model and the Expectations Hypothesis with Term-Premium model. |
