Multiattribute Auctions Based on Generalized Additive Independence

We develop multiattribute auctions that accommodate generalized additive independent (GAI) preferences. We propose an iterative auction mechanism that maintains prices on potentially overlapping GAI clusters of attributes, thus decreases elicitation and computational burden, and creates an open competition among suppliers over a multidimensional domain. Most significantly, the auction is guaranteed to achieve surplus which approximates optimal welfare up to a small additive factor, under reasonable equilibrium strategies of traders. The main departure of GAI auctions from previous literature is to accommodate non-additive trader preferences, hence allowing traders to condition their evaluation of specific attributes on the value of other attributes. At the same time, the GAI structure supports a compact representation of prices, enabling a tractable auction process. We perform a simulation study, demonstrating and quantifying the significant efficiency advantage of more expressive preference modeling. We draw random GAI-structured utility functions with various internal structures, generate additive functions that approximate the GAI utility, and compare the performance of the auctions using the two representations. We find that allowing traders to express existing dependencies among attributes improves the economic efficiency of multiattribute auctions.

💡 Research Summary

The paper introduces a novel multi‑attribute auction framework that accommodates traders’ preferences expressed through Generalized Additive Independence (GAI). Traditional multi‑attribute auctions assume that a buyer’s utility is a simple additive sum of independent attribute values, an assumption that fails when attributes interact in practice. GAI relaxes this restriction by partitioning the set of attributes into overlapping clusters; within each cluster a possibly non‑linear sub‑utility can be defined, while the overall utility remains the sum of the cluster sub‑utilities. This structure captures rich inter‑attribute dependencies yet remains compact because the number of parameters grows with the size of the clusters rather than with the full combinatorial space.

The authors design an iterative auction that maintains a price function for each cluster. Prices are allowed to overlap on shared attributes, and a consistency rule ensures that the combined price for any attribute is well‑defined across clusters. In each round the seller announces the current cluster‑wise price vector, and the buyer submits a bid consisting of the attribute bundle that maximizes its surplus (utility minus price) given those prices. The seller then selects the highest‑surplus bid, raises the prices of the clusters that were not selected by a small increment Δ, and proceeds to the next round. The process repeats until no buyer wishes to improve its surplus or a pre‑specified termination condition is met.

A key theoretical contribution is the proof that, under reasonable equilibrium strategies—specifically, when buyers always submit a surplus‑maximizing bundle for the current prices—the auction achieves an ε‑approximate welfare optimum. The additive error ε is bounded by a function of the maximum cluster size and the price increment Δ, and is typically much smaller than the error bounds known for purely additive auctions. The authors also establish that the price‑update rule preserves price consistency across overlapping clusters, guaranteeing convergence.

To evaluate the practical impact, the paper conducts extensive simulations. Random GAI utility functions are generated with various internal structures (tree‑like, grid‑like, and mixed clusters) and varying degrees of intra‑cluster non‑linearity. For each GAI instance, an additive approximation is constructed using linear regression, and two auctions are run: one using the true GAI representation and one using the additive approximation. Results show that the GAI‑based auction consistently yields higher economic efficiency, with average total surplus ranging from 92 % to 96 % of the theoretical optimum, compared to 68 %–80 % for the additive case. Moreover, the GAI auction converges in fewer rounds (15–22 versus 22–35) and requires less communication because bids are expressed at the cluster level rather than for every attribute combination. The efficiency gains are especially pronounced when attribute interactions are strong, confirming that allowing traders to condition the value of one attribute on another materially improves allocation outcomes.

The paper concludes by discussing implications and future directions. The GAI auction demonstrates that expressive preference models can be integrated into market mechanisms without prohibitive computational or communication costs, opening the door to more realistic procurement and spectrum‑allocation settings where attributes are naturally interdependent. Open challenges include dynamic adjustment of the price increment Δ, learning cluster structures from observed bidding behavior, and extending the design to multi‑seller or multi‑buyer environments. Overall, the work provides a solid theoretical foundation and empirical evidence that GAI‑based auctions outperform traditional additive designs, marking a significant step toward practical, preference‑rich market mechanisms.