Can Learning Reduce the Dark Matter in the Long-Run Risk Model?
DescriptionThe long-run risk (LRR) model helps explain a number of stylized facts in asset pricing. However, the model contains too much “dark matter” (Chen et al., 2019) in the sense that it relies on features that are diﬃcult to directly test and measure such as LRR in consumption growth. The excessive dark matter poses a serious threat to the viability of the LRR model because it means that the model may be overﬁtting returns data and may fall apart when there is a structural break.In this project, I intend to investigate whether incorporating learning can sig-niﬁcantly reduce the dark matter in the LRR model. Standard LRR models assume that the representative agent observes the time-varying expected growth rate, and the large dark-matter measure is also calculated under the assumption. However, given the abundance of evidence that learning improves the performance of the LRR model on various fronts (Andrei and Hasler, 2015; Croce et al., 2015; Shaliastovich, 2015; Johannes et al., 2016; Hasler et al., 2019; Choi, 2019), it seems natural to investigate whether learning can address the dark-matter issue as well.Speciﬁcally, I will consider three speciﬁcations in the order of less knowledge on the part of the agent and examine the associated changes in the level of dark matter along the way: (i) The agent observes the expected growth rate; (ii) the agent does not observe the expected growth rate and learns about it through Bayesian ﬁltering; and (iii) not only does the agent not observe the expected growth rate but she does not know its dynamics, either, and she learns about both as in Choi (2019).
|Effective start/end date||1/01/22 → …|