A multi-agent simulation approach to farmland auction markets : repeated games with agents that learn

The focus of this thesis is to better explore and understand the effects of agent interactions, information feedback, and adaptive learning in a repeated game of bidding in farmland auction markets. This thesis will develop a multi-agent model of farm-land auction markets based on data from the Saskatchewan Dark Brown Soil Zone of the Canadian Prairies. Several auction types will be modeled and data will be gathered on land transactions between farm agents to ascertain which auction type (if any) is best suited for farmland markets. Specifically, the model gathers information for 3 types of sealed-bid auctions, and 1 English auction and compares them on the basis of efficiency, price information revelation, stability, and with respect to repeated bidding and agent learning. The effects of auction choice on macro-level indicators, such as farm exits, retirement, financial stability, average productivity, farm size, and participation were unknown at the outset of this thesis because of the complex dynamic nature of the environment. I find that the chosen learning mechanism employed here affects both price and variance of prices in all auctions. I also find that the second-price-sealed-bid auction generates the most perceived surplus, most equitable share of surplus, and also decreases uncertainty in the common-value element of prices. A priori it was believed that auction choice would have an impact on pricing efficiency, price levels, and shares of surplus generated from auctions as predicted by theoretical works. Surprisingly, auction choice does not influence market structure or evolution