A Local Series Estimation Method with its Application in Solving Asset Pricing Models with Time-Varying State Variables

Project: Research

View graph of relations

Description

We propose a nonparametric local series estimation method for unknown policyfunctions in an exchange economy (Epstein and Zin, 1989), which will allow the truedynamics of time-varying state variables to determine asset prices. Unlike currentnumerical solution methods, this new method does not require the imposition of tightauxiliary parametric assumptions on the conditional distributions or matching momentsof state variables and thus avoids spurious model conclusions due to misspecificationerrors. Blanchard (2016) advocates for efforts to make general equilibrium models lessinsular to the real economy by allowing empirical data to determine the true dynamics ofthe structural model. However, the asset pricing literature still does not have a clearunderstanding of how to connect policy functions, such as price-dividend and wealth-consumptionratio functions, with empirical data which exhibit time-varying features.This proposal will allow for true evolutions of both stationary and nonstationary statevariables, such as consumption and dividend growth rates and income innovationsinstead of assuming some time- and event-invariant pseudo data generating processesfor computational convenience. This will allow researchers to carry DSGE models fromthe laboratory to the real world.The method proposed provides a particularly powerful and reliable way to reveal the roletime-varying state variables, such as income risks, play in determining the time-seriesproperties of the risk-free rate, return on wealth and market returns. We also intend toexamine to what extent the declining predictability ability of dividend-price ratios onexcess returns after 2002 can be explained by the time-varying features of price-dividendratios and state dynamics.This proposal is also among the first to combine local and global estimation methods,which have been employed as substitutes in most cases. For the first time in theliterature, we will establish the consistency and asymptotic normality of unknown policyfunctions, which offers another salient contribution to econometric studies.In addition, this method will become a pivotal approach for obtaining a consistentestimation of unknown policy functions in the presence of time-varying state variableswith unknown dynamics. Our proposal can be extended in many directions and appliedin many contexts. First, the methodology can be applied to DSGE models of theproduction economy that has latent state variables. In addition, this method can befurther extended from the time series framework to a panel data structure to analyzeheterogeneous agent models with both stationary and non-stationary state variables.

Detail(s)

Project number9048129
Grant typeECS
StatusActive
Effective start/end date1/01/19 → …