Soumen Banerjee

We study full implementation with evidence in an environment with bounded utilities. We show that a social choice function is Nash implementable in a direct revelation mechanism if and only if it satisfies the measurability condition proposed by Ben-Porath and Lipman (2012). Building on a novel classification of lies according to their refutability with evidence, the mechanism requires only two agents, accounts for mixed-strategy equilibria and accommodates evidentiary costs. While monetary transfers are used, they are off the equilibrium and can be balanced with three or more agents. In a richer model of evidence due to Kartik and Terceiux (2012), we establish pure-strategy implementation with two or more agents in a direct revelation mechanism. We also obtain a necessary and sufficient condition on the evidence structure for renegotiation-proof bilateral contracts, based on the classification of lies.

How should mechanisms be designed and evaluated when agents learn from feedback rather than solve for equilibrium? This paper answers that question through two lenses: challengeability for design and learnability for evaluation. Challengeability classifies profitable misreports by who can expose them: another agent or only the deviator herself. Using this distinction, we construct a finite mechanism that fully implements any Maskin-monotonic social choice function in correlated equilibrium in finite complete-information quasilinear environments with lotteries and transfers. The mechanism induces a unique correlated-equilibrium outcome at every state, which makes it learnable: under standard no-internal-regret dynamics, the long-run empirical distribution of play converges to the target outcome. Learnability also gives a way to compare mechanisms once agents adapt. We illustrate both lenses in bilateral trade and public-goods environments and compare our mechanism to the Abreu–Matsushima virtual benchmark. In simulations, the row-scaled Hart–Mas-Colell internal-regret rule used in the paper converges quickly to the desired outcome and generates lower burnt-transfer costs and smaller surplus-division distortions than the virtual benchmark in both applications.

Flexibility in electricity demand can be leveraged for demand side management (DSM) to enable aspects such as “demand following generation” and provide ancillary services to support grid integration of renewable and distributed energy resources. The upcoming connected devices in demand centers such as commercial buildings and households may be leveraged for this. However, the coordination among distributed demand centers in a scalable decentralized manner to achieve DSM objectives is still a challenge.

In this article, a generalized hierarchical transactive energy based multiagent framework is proposed. This framework includes energy management demand agents (EMDAs) at the building level, which coordinate the operation of different appliances within the buildings. EMDAs also actively participate on day-ahead Walrasian market at the cluster of buildings level. In this market, the effect of wide-adoption of DSM on generation dispatch is also studied.

A particular instance of this framework is also implemented in a cyber-test system consisting of standard industrial microcontroller platforms to emulate practical implementations via smart meters. Experimental results of the proposed system are included to demonstrate the possibility of avoiding dispatch from expensive generation units by reduction in peak demand and providing demand response inherently.

Algorithmic pricing raises questions of both interpretation and intervention: when autonomous deep-learning systems sustain supracompetitive prices, what strategy have they learned, and how can market institutions alter it? This paper develops an interpretable framework for learned collusion in repeated pricing. Strategic deep-learning networks are embedded in a differentiated-products Bertrand market, with recent price histories compressed into finite states that record price levels, rival price movements, and movement persistence. This representation improves learning efficiency and preserves the dynamic information relevant for reward and punishment while keeping learned behavior economically interpretable: in the baseline environment, agents learn supracompetitive prices and a coherent collusive asymmetry, punishing rival price cuts and accommodating rival price increases. The framework is then used to study an order-book mechanism that assembles temporary buyer commitments and allocates them to sellers willing to make sufficiently deep undercuts, partially insulating those undercuts from retaliatory punishment. Trained under the mechanism, agents set lower realized prices, closing about 47% of the baseline collusion gap. Diagnostics show the reduction works through the intended channel, as protected undercuts become less exposed to punishment, weakening the threat that sustains high prices. The results show how interpretable learning frameworks can connect algorithmic pricing outcomes to economic mechanisms, and how market design can target the enforcement channel behind learned collusion.