Multiattribute Call Markets.

Abstract

Multiattribute auctions support automated negotiation in settings where buyers and sellers have valuations for alternate configurations of a good, as defined by configuration attributes. Bidders express offers to buy or sell alternate configurations by specifying configuration-dependent reserve prices, and the auction determines both the traded goods and transaction prices based on these offers.

While multiattribute auctions have been deployed in single-buyer procurement settings, the development of double-sided multiattribute auctions-allowing the free participation of both buyers and sellers-has received little attention from academia or industry. In this work I develop a multiattribute call market, a specific type of double auction in which bids accumulate over an extended period of time, before the auction determines trades based on the aggregate collection of bids. Building on a polynomial-time clearing algorithm, I contribute an efficient algorithm for information feedback. Supporting the implementation of market-based algorithms, information feedback support extends the range of settings for which multiattribute call markets achieve efficiency.

Multiattribute auctions are only one of many auction variants introduced in recent years. The rapidly growing space of alternative auctions and trading scenarios calls for both a standardized language with which to specify auctions, as well as a computational test environment in which to evaluate alternate designs. I present a novel auction description language and deployment environment that supports the specification of a broad class of auctions, improving on prior approaches through a scripting language that employs both static parameter settings and rule-based behavior invocation. The market game platform, AB3D, can execute these auction scripts to implement multi-auction and multi-agent trading scenarios.

The efficiency of multiattribute call markets depends crucially on the underlying valuations of participants. I analyze the expected performance limitations of multiattribute call markets, using existing analytical results where applicable. Addressing a lack of theoretical guidance in many natural settings, I introduce a computational metric on bidder valuations, and show a correlation between this metric and the expected efficiency of multiattribute call markets. As further validation, I integrate multiattribute markets into an existing supply chain simulation, demonstrating efficiency gains over a more conventional negotiation procedure.