Develop learning agents to optimize auction design by predicting customer bidding behaviour

Combining empirical evidence such as historical transaction data with intuition and experience allows auctioneers make better decisions in auction markets. A research study with Floraholland reveals how we can leverage the power of data to improve the decision-making in dynamic, complex auction markets. Decision-making plays a big part in auction markets and traditionally, these decisions are largely based on the buyers’ and sellers’ intuition and previous experience. Auctions are popular mechanisms for price discovery and resource allocation. They play an important role in the modern economy.

Auctioneers need to decide when to sell what products, in which quantities, and what information to disclose to the buyers. Due to the cognitive limitations, auctioneers cannot process all the market information fast enough to make informed decisions. Instead, they mainly rely on the intuition and past experience.

We examined the promises of data-driven decision-making in these complex auctions by studying the interplay of different informational and strategic factors. We found that theoretically guided analytical tools have great potential for facilitating the real-time decision-making in complex auction markets. Therefore, the strength of empirical data should be combined with with knowledge about the market environment.  If auctioneers want to make the best decisions, they must integrate the domain knowledge and the rich data from different sources. We also investigated the effect of information revelation policy on price dynamics and market performance. One of the findings is that bidders tend to pay higher prices when the identities of winners are concealed from public view. Such positive effect holds for both online and offline bidders, suggesting that the weak signals or the increased market state information from the offline channel cannot compensate for the loss of the additional information conveyed via winners' identities. In addition, our analysis show that anonymising the winning bids also helps to mitigate the price declining trend in sequential rounds.

PhD Thesis

http://repub.eur.nl/pub/51543

Integrating Data and Knowledge Leads to Better Decisions in Auction Markets
Integrating Data and Knowledge Leads to Better Decisions in Auction Markets  
Play video