In Chapter 1, I analyze firms' misallocation through the output distortions channel, using a production-based asset pricing model as a framework. In the model, alpha measures the firm's ability to choose technologies to adapt to exogenous shocks. I find in the cross-section of the test portfolios the estimated curvature parameter alpha is more than two times the original value obtained in Belo (2010). This implies misallocations reduce the firm's ability to respond to the different states of nature. I calibrate and solve the model in the special case of a single representative firm. I find that the impact of misallocation on firm value, production, capital, investment, and investment return is larger when firms' ability to adapt to exogenous shocks is reduced. This indicates that firms may be less agile to adapt across states of nature and provides more evidence of the detrimental effect of misallocations.

In Chapter 2 (with Denis Mokanov and Danyu Zhang), we document several facts about equity analysts' earnings expectations: (1) consensus earnings expectations underreact to news unconditionally, (2) the degree of underreaction declines during high-volatility periods, and (3) the degree of underreaction declines over our sample. To account for these findings, we develop a simple model featuring time-varying inattention. We show that our model is able to account for the unconditional profitability of momentum, momentum crashes, and the diminishing profitability of momentum over our sample. We propose a trading strategy that mixes short-run and long-run momentum signals and show that the mixed momentum strategy outperforms the conventional momentum strategies. Finally, we use a machine learning algorithm to estimate the predictable component of earnings surprises and construct a portfolio that is long (short) on stocks with excessively pessimistic (optimistic) earnings expectations. The resultant trading strategy generates an annualized Sharpe ratio of about 1.16 and its returns are not explained by popular factor models.