We explore statistical commonalities among granular measures of market liquidity with the goal of illuminating system-level patterns in aggregate liquidity. We calculate daily ``invariant'' price impacts \cite{KyleObizhaeva2014theory} to assemble a granular panel of liquidity measures for equity, corporate bond, and futures markets. We estimate Bayesian models of hidden Markov chains and use Markov chain Monte Carlo analysis to measure the latent structure governing liquidity at the system-wide level. Three latent liquidity regimes--- high, medium, and low price-impact---are adequate to describe each of the markets. Focusing on the equities subpanel, we test whether a collection of macroeconomic time series can recover the estimated liquidity dynamics. This allows an economically meaningful attribution of the latent liquidity states and yields meaningful predictions of liquidity disruptions as far as 15 trading days in advance of the 2008 financial crisis.
Copyright. All Rights Reserved.