Regime Learning and Asset Prices in a Long-Run Model : Theory

This paper tries to draw on the relative merits of both the jump risk models and the long-run risk models with a linkage established by Bayesian learning, in an attempt to improve both asset pricing approaches in producing a better mechanism for understanding asset prices regularities.Rather than treating event risk as direct jumps in the level of aggregate income, we model it as changes in the underlying state of the world, the economic regimes, which affect aggregate consumption and dividend flows through their growth and volatility's dependence on the state.Realistically, information about the state transition is imperfect in this representative agent endowment economy and agents with recursive utility perform Bayesian learning to form and update beliefs about the conditional state arrival in order to make optimal long-run consumption investment decisions. This new learning component to the consumption-based paradigm will generate novel pricing implications through inducing extra covariance to be priced. Specifically, besides the aggregate uncertainty stemming from jump risk exposure, the presence of imperfect learning behavior also generates individual ambiguity. We shall see that such dual channels can help better explain some asset pricing regularities observed, e.g. the dual puzzles, predictability issues, time-varying conditional moments, etc., and shed some new light on the long-run cash flow news approach in asset pricing