Using a controlled trading experiment spanning 150,000 trades, we study how retail brokers route orders and respond to execution quality. Within each broker, execution quality varies substantially and persistently across wholesalers. Yet broker responses to these dispersions are strikingly heterogeneous: while some route more orders to lower-cost wholesalers, others consistently send more to higher-cost ones. We develop a switching cost model that rationalizes these contrasting behaviors. A natural exper- iment involving wholesaler entry demonstrates that competition improves execution quality even for the most responsive broker. This heterogeneity in how brokers implement best execution requirements shapes the competitive pressure on wholesalers.

Although still relatively new, nocturnal trading in U.S. equities, defined as trading between 8:00 p.m. and 4:00 a.m., has grown rapidly. It is largely retail-driven, concentrated in a small set of securities, and marked by substantial order imbalances. Using unique transaction-level data, we show that nocturnal execution costs exceed regular-hours benchmarks but are broadly consistent with elevated adverse selection faced by liquidity suppliers. Nocturnal returns generally do not reverse during the subsequent regular-hours session, except in a small subset of high-sentiment stocks. Overall, the nocturnal session appears to be an important source of price discovery and may create profit opportunities for retail liquidity demanders despite higher transaction costs.

A long-standing puzzle in U.S. equity markets is why hundreds of off-exchange venues coexist. Under the conventional best-execution metrics of execution cost and speed, many venues appear both slower and more expensive than their competitors. We propose that execution-speed volatility helps explain this inconsistency. This paper constructs the first empirical execution-speed volatility measure and shows that higher execution costs are associated with lower execution-speed volatility. The compensation is strongest in the right tail of the speed distribution, where orders execute with extreme delay or fail to execute at all. Among securities likely to be involved in arbitrage, where trading success depends on coordinating multiple execution legs, speed volatility but not speed is compensated. We find that younger venues tend to offer lower speed volatility and higher cost. Finally, a random-coefficients venue-choice model shows that incorporating execution-speed volatility as a third dimension of execution quality substantially improves model fit and helps rationalize market fragmentation as an equilibrium outcome.