We introduce the Generalized Multi-Asset Model (GMAM), a novel hierarchical Bayesian state-space framework designed to estimate latent (de-smoothed) economic returns and factor exposures for illiquid alternative investments, including private equity, private credit, real estate, and hedge funds. GMAM addresses key challenges such as stale NAVs, return smoothing, and sparse data by incorporating a Bayesian moving-average process and stochastic search variable selection (SSVS). Using a sample of 57 funds on our platform, GMAM outperforms ordinary least squares (OLS) in matching predicted returns, achieving a mean out-of-sample R² of 21.96% versus 17.02% for OLS. In de-smoothing diagnostics, GMAM largely recovers the dynamics of an artificially smoothed S&P 500 series and reverses NAV-induced volatility compression across funds, with a median reported-to-latent volatility ratio of 0.56. GMAM contributes a scalable, probabilistic approach to modeling alternative assets, bridging the gap between academic theory and practitioner needs.
Sheth et al. (Fri,) studied this question.