This paper analyses bidder behaviour in the German photovoltaic (PV) auction pilot. It uses a novel approach combining insights from data analysis and decision theory to optimise an agent-based simulation model. We model both a uniform pricing (UP) scheme, a pay-as-bid (PAB) scheme, and a benchmark case, where agents adapt their bidding strategy. The findings are contrasted with empirical auction outcomes. The comparison shows that, especially in the early rounds, bid prices were above the costs-possibly due to uncertainties and false expectations concerning competition. This is particularly visible in the first round. Adapting their expectations to a higher competition level, bidders in the pay-as-bid simulation subsequently decrease their bids. From simulating a separate auction for amble land bidders, we see that this bidder type reduces support costs substantially and that an implicitly discriminatory auction yields more aggressive bids and can induce further cost reductions.