Trend-cycle decompositions allowing structural changes and outliers: Econometric analyses and macroeconomic applications

This dissertation pertains to issues related to trend-cycle decompositions allowing structural breaks and outliers.;The first chapter is concerned with this decomposition problem for the postwar quarterly US real GDP series. Recent work shows that common methods such as the Unobserved Components (UC) and Beveridge-Nelson decompositions for US real GDP yield very different cycles with little resemblance to the NBER chronology, and some imply a negative correlation between the cycle and the trend. We argue that these features are due to the neglect of a change in the slope of the trend in 1973. Allowing this change, all methods yield a non-stochastic trend with a cycle that accords well with the NBER chronology. We also propose a generalized UC model, which permits endogenous infrequent changes using a mixtures of Normals distribution for the shocks.;The second chapter generalizes the generalized UC model with mixtures of Normals to handle sudden level shifts and outliers as well. We investigate the differences in the implied trend and cycle compared to the popular decomposition obtained with the Hodrick and Prescott filter. Our results show important qualitative and quantitative differences in the implied cycles for both real GDP and consumption series for the G7 countries. Using our decomposition, the cross-country consumption correlations are generally higher than the output correlations, a result which provides a partial solution to the consumption correlation puzzle in the international finance literature.;The third chapter considers another important issue in Macroeconomics. A puzzling feature in the literature is that it is believed that a positive technology shock decreases hours. This result arises because of structural breaks in the trend of labor productivity and the low frequency component of hours. We revisit the empirical evidence by allowing non-linearities in the trend function of both the productivity and hours worked series. Along with statistical tests for unit root and structural breaks, using our generalized unobserved components model with flexible trends allows us to conclude that productivity shocks unambiguously increase hours worked, consistent with the prediction of standard real business cycle models.