Deep Learning Based Auction Design for Selling Agricultural Produce through Farmer Collectives to Maximize Nash Social Welfare (2512.22039v1)

Abstract: This paper is motivated by the need to design a robust market mechanism to benefit farmers (producers of agricultural produce) as well as buyers of agricultural produce (consumers). Our proposal is a volume discount auction with a Farmer Collective (FC) as the selling agent and high volume or retail consumers as buying agents. An FC is a cooperative of farmers coming together to harness the power of aggregation and economies of scale. Our auction mechanism seeks to satisfy fundamental properties such as incentive compatibility and individual rationality, and an extremely relevant property for the agriculture setting, namely, Nash social welfare maximization. Besides satisfying these properties, our proposed auction mechanism also ensures that certain practical business constraints are met. Since an auction satisfying all of these properties exactly is a theoretical impossibility, we invoke the idea of designing deep learning networks that learn such an auction with minimal violation of the desired properties. The proposed auction, which we call VDA-SAP (Volume Discount Auction for Selling Agricultural Produce), is superior in many ways to the classical VCG (Vickrey-Clarke-Groves) mechanism in terms of richness of properties satisfied and further outperforms other baseline auctions as well. We demonstrate our results for a realistic setting of an FC selling perishable vegetables to potential buyers.