Macro factors and the yield curve

Term structure models in finance are dominated by latent factors that lack economic content. This thesis presents a combined macro-finance approach to study the yield curve. It answers three main questions. First, how are the yield curve and bond risk premiums related to the macroeconomy? Second, does the level of macroeconomic volatility affect investors' risk compensations? Third, are term structure models useful for forecasting?; Chapters 2 and 3 present estimation results of the macro-finance model for the periods 1988-2002 and 1965-1987 respectively. They show that a lot of the variations in bond yields can be explained by macroeconomic fundamentals. The macro-factors explain the bulk of variations in bond yields and account for all the time variation in the market prices of risk. Variance decompositions show that real economic activity account for most yield curve movements between 1988 and 2002 whereas inflation is found to be much more important for the earlier period. In Chapter 3, I also show that the market prices of macroeconomic risks have shifted over time. The more volatile macroeconomic environment between 1965 and 1987, marked by high and volatile inflation, has greater risk compensations than the later period of macroeconomic quiescence. For estimation, I develop a likelihood-based method, implemented with Markov Chain Monte Carlo, to estimate the factors and model parameters simultaneously.; Chapter 4 studies bond yields forecasting. Forecasts computed from a Gaussian term structure model with factors identified from the time series outperform those based on factors identified from the cross-section by a wide-margin. Forecasting gains increase with the forecast horizons. The term structure model forecasts better than a whole host of time series models. The superiority stems from cross-equation restrictions that limit cross-sectional variations in bond yields over time. Finally, a combined macro-finance model of the yield curve improves long-horizon forecasts even further. Out-of-sample results show that the pricing error assumption made in many empirical studies of the term structure helps deliver great in-sample fit but misses important variations in bond yields. Such variations can only be identified in the time series and are important for forecasting yield curve movements.