Learning and Asset Pricing
- Author(s): Shin, Michael
- Advisor(s): Branch, William
- et al.
The rational expectations (RE) hypothesis although elegant and useful requires demanding assumptions on part of the agent. A key outcome of the RE hypothesis is that beliefs disappear as an independent force in the model. A branch of the literature focuses on relaxations of the RE hypothesis to allow agents to instead learn the data-generating process (DGP) over time. With adaptive learning, beliefs re-emerge as a key element that influences the DGP. I argue that this relaxation is important for asset pricing settings, which are complex environments where individual beliefs play a key role in decision making. My dissertation will explore instances where learning can improve our understanding of asset pricing.
Chapter 1 presents a simple asset pricing model with endogenous participation that can match key volatility moments when agents adaptively learn about both the risk and the return of stocks. With learning about risk, excess volatility of prices is driven by fluctuations in the participation rate that arise because agents' risk estimates vary with prices. I find that learning about risk is quantitatively more important than learning about returns. A calibrated model can jointly match the mean participation rate, the volatility of participation rates, and explain 25% of the excess volatility of stock prices observed in U.S. data.
Chapter 2 presents a simplified version of the model in Chapter 1 and tests the model in a laboratory setting. Recent evidence suggests subjective returns play a key role in stock market participation. Furthermore, there is strong evidence that stock market experiences, i.e. realized returns, impact subjective returns. I bring a model into the laboratory and find that learning-driven subjective returns can explain limited participation. Stock market participation is increasing in both subjective returns and past realized returns. I find direct evidence that "learning from experience" generates heterogeneity in subjective returns, where subjects who experience low returns have lower subjective returns than subjects who experience high returns. In particular, subjects over-weigh price trends when they experience high returns and under-weigh it when they experience low returns.
Chapter 3 presents an asset pricing model where agents test the specification of their models, while adaptively updating the parameters and find that restricted-perceptions equilibria (RPE) naturally arise. I extend upon recently developed model specification techniques to a multi-agent framework. Multiple agents are endowed with different models which they update and test the specification in real time. When a model is rejected, agents draw a new model from a distribution. I find that the rational expectations equilibrium (REE) is not locally stable with respect to hypothesis testing under reasonable parameterization. With constant-gain learning, the model spends most of its time in a subset of the RPE and in particular, the dominant model used is not the fully-specified model, but a misspecified one.