We characterize the dynamic fragmentation of U.S. equity markets using a unique data set that disaggregates dark transactions by venue types. The “pecking order” hypothesis of trading venues states that investors “sort” various venue types, putting low-cost-low-immediacy venues on top and high-cost-high-immediacy venues at the bottom. Hence, midpoint dark pools on top, non-midpoint dark pools in the middle, and lit markets at the bottom. As predicted, following VIX shocks, macroeconomic news, and firms’ earnings surprises, changes in venue market shares become progressively more positive (or less negative) down the pecking order. We further document heterogeneity across dark venue types and stock size groups.

We study liquidity supply in fragmented markets. Market makers intermediate heterogeneous order flows, trading off spread revenue against inventory costs. Applying our model to payment for order flow (PFOF), we demonstrate that portfolio-based considerations of inventory management incentivize market makers to segment retail orders by siphoning them off-exchange. Banning order flow segmentation reduces total welfare, can make trading more costly for all investors, and can resolve a prisoner's dilemma among market makers. These results differentiate our inventory-based model from the existing information-based theories of PFOF.