In most equilibrium asset pricing models, investors are assumed to know the dynamics of firm cash flows and economic fundamentals for sure. I develop a new regime-switching framework that relaxes this assumption without losing mathematical rigor or empirical tractability. In the model, investors know true cash flow dynamics conditional on macroeconomic or asset-level expansions and contractions, but do not know transition dynamics between the regimes. Using a new solution concept, I obtain asset prices in closed form for given subjective beliefs about future regime likelihood and show how beliefs can be jointly backed out from the term structures of long-term bond yields and individual and S&P500 VIX measures and their futures.

The new framework qualitatively explains a wedge between investor risk premium and observed average excess returns, a cyclicality gap between subjective risk premia and realized excess returns, a pro-cyclical gap between VIX and realized volatility, yield curve inversion/reversions without a regime shift, return predictability, narrative momentum, market crashes without macroeconomic contractions, and fast price recoveries during economic downturns.

I propose a new model of rare economic disasters. Existing models of disaster risk assume disasters are like "nuclear strikes" from which economies can only gradually recover. As a result, model-implied default-free real yield curves are downward sloping, call option prices are low, and fast recoveries-unanticipated. Existing models also face the "dark matter" problem of calibration: their quantitative implications depend on the statistical properties of time-varying disaster probability, which are quantitatively intangible due to the rarity of disasters.

I propose a regime-switching model with subjective regime beliefs to address these issues. Regime-switching captures the idea that disasters are outcomes of persistent but potentially reversible bad regimes; that while disaster risk matters, so does the possibility of a boom. Subjective regime beliefs introduce a degree of freedom in the model that allows beliefs to be calibrated based on VIX. The model establishes a formal theoretical link between VIX and S&P500, can explain sudden recoveries, yield curve inversions/reversions, and time-varying discrepancy between VIX and realized volatility. The model also generates an extension of Gordon growth formula with time-varying discount rates and growth expectations, is likely to explain high put and call prices, and can offer additional quantitative insights into the puzzles already addressed by disaster risk literature.

The paper introduces an asset pricing model in which disaster-specific news constitute a major source of variation in asset prices independently from contemporaneous realizations of macroeconomic variables. The disaster-related variation in prices comes from changes in regime beliefs of rationally inattentive agents. The main contribution of the paper is incorporating rationally inattentive learning in the SDF framework and deriving closed-form expressions for endogenously formed regime beliefs. It aims to formalize the notion that investors have limited capacity for attention and learning, cannot possibly know true cash flow dynamics and that non-cash-flow news about future prospects can cause dramatic shifts in regime beliefs and asset prices.

I revisit a two-period, two-state model of Chavchanidze (2021), in which I analyze the portfolio selection problem of a rationally inattentive investor with uniform priors and fixed prices. In this paper, I relax the two restrictions and describe optimal information strategy for arbitrary prior beliefs, obtain unconditional signal probabilities and posterior beliefs in closed form and discuss asset pricing implications in a stochastic discount factor framework. The results mirror findings in Chavchanidze (2021): even with arbitrary priors and flexible prices, incomplete information induces risk-aversion in the good state and risk-seeking in the bad state relative to the complete information case. I also comment briefly on how this framework can yield insights about risk premium and low risk-free rate puzzles.

I incorporate behavioral and bounded rationality elements into a single asset-pricing framework by setting up a two-period consumption-based portfolio selection problem in which a representative agent has biased priors, does not observe the current state and thus has incomplete information about future state probabilities. He forms posterior beliefs using signals that he selects according to the rational inattention discrete choice framework of Matějka and McKay (2015), where the precision of the beliefs depend intuitively on the priors and the cost of information $\lambda$. In the case of log-utility, the optimal portfolio is a convex combination of the N portfolios the investor would have selected in each of the N states if they were fully observable, where the weights reflect the subjective posterior likelihood of time-zero states. The posterior beliefs are induced by parsimonious reweighing of priors, where the weights depend on $λ$, discount factor $\beta$ and the relative entropies of the future state distributions induced by different time-zero states.

Using a two-state example, I demonstrate how the cost of information and biases can be jointly analyzed in this framework and discuss implied deviations from fully rational behavior. The major advantage of the proposed model is its flexibility. When the cost of information $\lambda$ is zero and the agent has correct priors, the model reduces to the standard neoclassical framework. When $\lambda$ is non-zero and the agent has correct priors, it is a model of bounded rationality with endogenous signals and form of information, where the cost of information reflects the mental capacity of the agent. When $\lambda$ is zero and the agent has biased priors, the model reduces to the behavioral framework with standard preferences. The proposed framework could lay the foundations for multi-periods heterogeneous-agents models in which the effects of biases and costly information can be jointly analyzed and its consumption-based formulation might render it useful well beyond the asset pricing context.

The regulatory model for capital requirements under the first pillar of the Basel Accords assumes that portfolios are perfectly fine-grained and there exists no concentration risk from large exposures. Granularity Adjustment (GA) methodologies provide a way of assessing the amount of additional capital, which would be needed to account for undiversified concentration risk. Gordy and Lütkebohmert (2013) propose a closed-form expression that is derived within an Extended Single-Factor CreditRisk+ model and is calibrated to representative data from 38 German banks. We use a simulation framework to examine the behavior of their GA formulation against an actuarial CreditRisk+ model, in which obligors carry default risk only. Our findings show that the Gordy and Lütkebohmert (2013) specification approximates relatively well GA for large, diversified portfolios. In smaller and more granular portfolios that also carry the largest amount of non-accounted for concentration risk, however, the formula leads to a significant underestimation of GA, which calls for special regulatory attention under the second pillar of the Basel framework.

Probability distributions of many random variables defined on a sample of size n are known to be getting concentrated around a particular value exponentially fast with increasing sample size. The Large Deviation Principle (LDP) is an abstraction and a rigorous description of such behavior. In this paper we explore several statements of the LDP and give a friendly yet rigorous exposition of its essence using the notion of exponential decay and little-o notation. Further, we prove some sufficient conditions under which one could infer that the LDP is satisfied by a random variable just by looking at its density/mass function. We also comment about the relationship between the LDP and the Central LImit Theorem, and prove some sufficient conditions under which the satisfaction of the LDP implies that a Generalized Law of Large Numbers holds.