We study fair and efficient dynamic allocation under stochastic supply, using public housing as the application. Ex-post evaluation faults mechanisms for unforeseeable supply realizations, whereas ex-ante evaluation discards assignment-relevant information. We address both problems by developing a robust interim benchmark that evaluates each assignment using exactly the information available at the time it is made. Using this benchmark, we characterize when fairness and efficiency are jointly feasible. We construct the Choice-Based Waiting List (CBWL) mechanism and show that it is interim Pareto efficient in every problem, and eliminates interim justified envy and is robustly strategy-proof in acyclic problems. We then prove that acyclicity is the exact domain restriction for joint feasibility of interim fairness and efficiency. Stylized centralized waiting lists and site-based waiting lists are neither interim fair nor interim efficient, even in acyclic domains. Illustrative simulations calibrated to U.S. public housing suggest that the CBWL can generate economically meaningful welfare gains.

[Online Appendix]

Networked industries often rely on long-lived infrastructure whose operational availability can be controlled within the trading window. We study a two-sided transferable-utility matching model in which coalitions can withhold access to connections, thereby changing the active network on which trade is settled. This creates a network disruption externality: by altering outsiders' feasible options, a deviating coalition changes the settlement environment itself, so fixed-network core stability need not imply decentralized viability. We define network stability relative to a settlement rule and show that it coincides with the core of a derived network disruption game. Existence is characterized by Bondareva--Shapley balancedness, and a decomposition isolates when instability is driven by disruption rents rather than within-network rematching. Institutionally, a punishment benchmark yields a conservative impossibility test and highlights simple network primitives that guarantee nonexistence for all settlement rules. A design benchmark—disruption internalization—restores stability.

[Google Docs]

The recommendation principle asserts that a sender can, without loss of generality, restrict attention to experiments that directly recommend actions to the receiver. We study when this principle is valid in communication problems with commitment and provide a characterization of its applicability in terms of the convexity of the set of posteriors for which each action is optimal. Our characterization yields a simple sufficiency test for the applicability of the recommendation principle. The test implies existing applications of the recommendation principle in the literature as immediate corollaries and extends its validity beyond standard environments, including cases in which the receiver’s utility is not linear in the posterior or depends on the distribution over posteriors generated by the experiment.

[Online Appendix]

I study the relationship between a matching platform's design and users' welfare. Increasing users' number of prospects has a positive choice effect (users are more likely to find a desirable partner) and a negative competition effect (users are less likely to match). The interaction of choice and competition effects has three significant consequences. First, welfare is not strictly increasing but is single-peaked in users' number of prospects, leading to ambiguous platform network effects. Second, market sides even out in size through agents' optimal enrollment decisions. Finally, a designer generally fails to maximize users' welfare and platform enrollment simultaneously. 

[slides]