Abstract
Sparse models, though long preferred and pursued by economists, ap- pear ineffective/unstable relative to large models (Giannone et al., 2021). To achieve sparsity for interpretation while effectively exploiting big data for superior empirical performance, we introduce a general framework jointly clustering observations and selecting variables for modeling panel data. We derive analytical marginal likelihoods in our Bayesian Clustering Model (BCM), to incorporate economic guidance, address parameter/model un- certainties, and prevent overfitting. We apply BCM to estimating uncommon- factor-asset-pricing models for data-driven asset clusters and macroeco- nomic regimes. We find (i) cross-sectional heterogeneity linked to (inter- actions of) return volatility, size, and value, (ii) structural changes in fac- tor relevance predicted by market volatility and valuation, and (iii) MK- TRF and SMB as common factors and multiple uncommon factors across characteristics-managed-market-timed clusters. BCM helps explain volatility- or size-related anomalies, validate within-group tests, and mitigates the “factor zoo” problem, while outperforming benchmark common-factor mod- els in investments and pricing equities.

This publication has 61 references indexed in Scilit: