The focus of this thesis is on bond return predictability and providing an empirical and economic understanding of bond risk premia. The thesis consists of an empirical analysis of time-varying bond risk premia along three major branches of the current term structure literature, namely yields-only, macro-finance and multi-currency term structure models. All these models belong to the well-known class of affine models introduced by Ang and Piazzesi (2003), whereas the latter two embed unspanned factors. Unspanned factors are state variables that have an effect on bond risk premia but do not span the cross-section of yields, as recently introduced by Duffee (2011), Joslin, Priebsch and Singleton (2011) and Boos (2011). The section concerning yields-only models contributes by providing evidence of three priced risk premia of bonds in the US market, extending the analysis of Cochrane and Piazzesi (2005) and Boos (2011). The section concerning macrofinance models adds to the new branch of models with unspanned macro factors and extends existing research by analyzing the effects of unspanned macro factors on risk premia beyond the level risk premium and extending into a broader and longer data set of macroeconomic variables. The section concerning multi-currency models firstly introduces unspanned factors into international models by taking mutually unspanned latent yield curve factors of domestic and foreign countries as state variables. The information in foreign yield curves is found to be partly unspanned by the domestic yield curve and improves bond return predictability beyond local models.