Regularized GMM for Time-Varying Models with Application to Asset Pricing
43 Pages Posted: 8 Apr 2021 Last revised: 18 Jul 2022
Date Written: July 18, 2022
We develop a novel method to estimate time-varying GMM models via a ridge fusion regularization
scheme, which allows for a high dimension of instrumental variables. Our method
relaxes restrictions on the types of time variation (abrupt or smooth) and their sources and can
be implemented by a one-step procedure. Under regularizations, we have established consistency
and derived the limiting distribution for independent and dependent observations. This
regularizedGMMmethod provides an alternative solution for estimating the dynamic stochastic
discount factor (SDF) model by utilizing a large cross section and many conditioning variables.
The simulation study shows its robust performance for various data generating processes and
sample sizes. We apply our method to U.S. equities from 1972 to 2021. Our time-varying estimates
for factor risk price (SDF loadings) respond to changes in performance for multiple risk
factors and summarize potential regime-switching scenarios. By outperforming multiple benchmark
models, we demonstrate the gains in asset pricing and investment performance for our
regularized GMM model for in-sample and out-of-sample analysis.
Keywords: GMM, regularization, ridge fusion penalty, stochastic discount factor, time-varying model.
JEL Classification: C13, C14, C55, C58, C61, G12
Suggested Citation: Suggested Citation