FINTECH THRUST SEMINAR |  Time-varying Factor Selection: A Sparse Fused GMM Approach

Time-varying Factor Selection: A Sparse Fused GMM Approach

Abstract:

This paper proposes a new approach for estimating a time-varying coefficient model under the GMM framework. Our sparse fused GMM (SFGMM) method provides simultaneous specification and estimation for time-varying parameters, heterogeneous structural breaks, and time-varying sparsity of a potentially high dimension of covariates. We derive large sample properties for our estimator with and without prior knowledge of structural changes and test the conditional stochastic discount factor (SDF) model. Our method addresses the "factor zoo" challenge by providing a new perspective for time-varying factor selection.

First, our asymptotic theory on the time-varying specified model suggests re-jecting the fixed model hypothesis, indicating the significant factors and their iden-tities change over time. Second, we find the collective explanatory power of risk factors is high during periods of high interest rates or high inflation but declines when market liquidity is low. Third, the SFGMM strategy achieves the best risk-adjusted investment performance in the past four decades for out-of-sample per-formance comparison. Finally, we evaluate the unsynchronized factor discovery to accommodate real-time academic publication timings and find many factors are no longer selected or significant after publication.