This thesis takes the two step procedure implemented by (Diebold & Li (2006)) in which yields at different maturities are anticipated in time; however it proposes alternatives for the dynamic factor approach in the second stage depending in the structure presented by the data, for example, due to poor forecast using the VAR-1 of factors in levels (Rodriguez Revilla et al. (2016)) a stationarity and co-integration analysis are added to correct any miss specification, besides, in order to have more control in shocks coming from outside the model, some macro-economic variables correlated with the term structure are incorporated as (Ang & Piazzesi (2003), Ang et al. (2006)). Results show that having more information placed in exogenous variables does not improve any scenario in which these same observations are ignored, instead, it can diminish the quality of forecast; On the other hand, when any miss specification coming from non-stationary factors are corrected, some improvements can be achieved in particular for medium and long term bonds these corrections show equal performance of a random walk.
