Characterizing investor expectations for assets with varying risk

Abstract

How do financial market investors form expectations about assets with different risk characteristics? We examine this question using Euro-area yield curves for AAA-rated and AAA-with-other bonds. Investors’ conditional forecasts about the yield curves for different assets, at various forecasting horizons, are modeled using a VAR model with time-varying parameters. Two processes are assumed for the evolution of these parameters: a constant-gain learning model and a new endogenous learning technique proposed here. Both these algorithms allow investors to account for structural changes in the data. The endogenous learning mechanism also allows investors to compensate for large deviations in observed coefficients used for forecasting, relative to past data. Daily data is used to estimate the gain parameters for the learning algorithms, and we find that these gains vary across asset types, implying investors form conditional expectations differently for assets with differential risks. For 2005–2015, the investors’ conditional forecasts for the AAA-rated bonds are better described using the endogenous learning mechanism, implying that investors with lower risk preferences are more sensitive to large deviations in the data.