For the Utilities sector, the loading on the default spread factor is negative for most of the estimation period, as can be seen in Figure 2c. The sensitivity reversed signs for a brief period leading up to the end of the bull market. Since then the sensitivity has turned negative, and the absolute level has increased substantially. The behavior is consistent with the sector’s under- performance. Utility firms had large exposure to the credit market following the 1990’s. After 2000, default spreads widened with adverse effects for the sector. Although Bayes factor tests prefer the constant sensitivity specification to the dynamic one, the time-variation in the loading on changes in the yield curve, show an interesting evolution when considering the forecast of expected returns in the Healthcare sector. A shift from a negative sensitivity regime to a positive one is evident in early 1999. Since then, a positive change in the slope of the curve is associated with higher conditional expected returns.
Time variation of the intercept terms is displayed in figures 2e and 2f for the Technology and Telecommunications sectors respectively. The path of the intercept terms in both sectors, and especially in Technology, mirrors the fate of the sector, pre and post the market collapse in mid- 2000. The variation of the intercept term could be explained by either the omission from our model of additional factors that help forecast expected returns, or the existence of momentum effects in the market. This issue is addressed in more detail in a later section.
Cross Sectional Regression Results
As a general test of how well the time varying parameter model describes return behavior across portfolios, we propose a series of cross-sectional tests against the benchmark Sharpe-Lintner- Black Capital Asset Pricing Model (CAPM). Despite the many shortcomings of the CAPM, some discussed earlier in Section 1, in its basic form, it should be expected to explain a significant component of the variation in returns across sector portfolios. In contrast to a model estimated using lagged information, one might entertain the prior that a model with only lagged information would be at a disadvantage to a model such as the CAPM that incorporates contemporaneous information.
We consider the cross-sectional regression