Exploring online auction markets for remanufactured electronics using explainable artificial intelligence: evidence from eBay

Park, Yeun-Soo ORCID: 0009-0007-5140-950X (2025). Exploring online auction markets for remanufactured electronics using explainable artificial intelligence: evidence from eBay. University of Birmingham. Ph.D.

Park2025PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (5MB)

Abstract

Despite the growing relevance of remanufactured goods in circular economy and closed-loop supply chains, existing literature has paid limited attention to online auction-based transactions, focusing predominantly on fixed-price markets, treating remanufactured products as a homogeneous category, and largely overlooking bidder heterogeneity and the nuanced role of product condition levels in signal-based pricing mechanisms. This thesis addresses this gap by examining online auctions for remanufactured electronics, with a focus on how seller-, buyer-, and platform-level signals influence auction outcomes, how these effects differ across product condition levels, and how bidder heterogeneity informs distinct bidding strategies and behaviours.

While market signalling has been widely studied in fixed-price contexts, its role in online auctions for remanufactured products remains underexplored. This study addresses this gap by comparing traditional linear regression (LR) with machine learning (ML) approaches to model final auction prices. Using a dataset from eBay, the analysis reveals that Random Forest outperforms LR models in predictive accuracy, confirming the non-linear and interaction-driven nature of pricing dynamics in this domain. Explainable Artificial Intelligence (XAI) technique, specifically, SHapley Additive exPlanations (SHAP), is used to enhance model explainability, identifying key predictors (e.g., feedback score, bid count, starting price and product images) and revealing their complex interactions. Notably, the study uncovers an asymmetric feedback effect and provides feature importance alongside insights into the joint influence of market signals. These results contribute to a better understanding of how auction prices are shaped by the interplay of seller, buyer, and platform-level signals in remanufactured product markets.
The second empirical study investigates how auction outcomes and the effectiveness of market signals vary across remanufactured product condition tiers: Excellent, Very Good, Good. Based on eBay transaction data, the findings reveal distinct behavioural and pricing patterns by condition level. Higher-condition items tend to be listed by less experienced sellers employing affective descriptions, low starting prices, and extensive visual content—strategies associated with higher final prices. Conversely, lower-condition items are offered by more experienced sellers who rely on functional signals but achieve lower price outcomes. This study is the first to systematically analyse condition-specific heterogeneity in auction prices for remanufactured products, offering new empirical evidence on the differential role of market signals by product condition.

The third empirical study explores bidder heterogeneity in remanufactured product auctions by extending the behavioural framework of Bapna et al. (2004) with additional variables such as feedback scores and average bid increments. Through K-means clustering followed by SHAP-enhanced XGBoost classification, the study identifies a new hybrid bidder type, Shooters, positioned between traditional Snipers and Participators. The approach provides interpretable, data-driven insights into the behavioural attributes that differentiate bidder segments. Findings indicate that Snipers exhibit the highest win rates and bidding efficiency, while Shooters also perform competitively, demonstrating the viability of adaptive hybrid strategies. This is the first study to empirically classify bidder behaviour in remanufactured product auctions, offering novel contributions to auction theory and bidder segmentation.

Collectively, by analysing the price effects of market signals, the influence of product condition, and bidder heterogeneity in online auctions, while integrating advanced ML and XAI techniques, this research addresses key gaps in the literature on remanufactured product markets. These contributions enhance our understanding of consumer behaviour and pricing dynamics in auction environments marked by high information asymmetry, while also advancing theoretical discourse in remanufacturing, closed-loop supply chains, and data-driven operations research.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):