Uncovering Bidder Behaviour in the German PV Auction Pilot: Insights from Agent-based Modeling.

AuthorWelisch, Marijke

    This paper enables a deeper understanding of the ground-mounted solar PV auctions in Germany. The German PV pilot took place over six rounds in 2015 and 2016 in which both pay-as-bid (PAB) and uniform pricing (UP) schemes were implemented. We contribute to the understanding of the extreme price reduction in the six rounds of the pilot. Furthermore we investigate, how the different legislatory and auction design changes, namely the exemption for arable land bidders and the change between UP and PAB influenced bidding behaviour. Granted the opportunity to make use of detailed data on the pilot provided by the German Federal Ministry of Economic Affairs and Energy (BMWi), we statistically analysed empirical outcomes of these auctions to define input parameters for our agent based model. The model is further endorsed by decision theory. We thus benefit from empirical experience to improve our model and at the same time learn from the modeling results how varying design parameters changes auction results. This two-sided learning offers new insights regarding the bidder behaviour in auctions for renewable energy support and the combination of methods provides decision support for policy makers from a novel scientific perspective.

    The underlying research adds on to a strand of literature on auctions for renewable energy that is relatively recent. In particular, the paper uses agent-based modeling, a well established methodology to simulate auctions and electricity markets but which has only recently been applied to assess RES auctions in particular (see Anatolitis and Welisch, 2017; Welisch, 2018; and Welisch, 2019). Earlier studies using an agent-based model (ABM) to model auctions in a non-renewables context are e.g. Mizuta et al. (2000) or Hailu et al. (2011). Other recent papers dealing with auction-based renewables support are e.g. Haufe and Ehrhart (2018) which gives an overview of relevant auction design elements Winkler et al. (2018) which evaluates the performance of auctions for renewable energy support or del Rio (2017) and Mora et al. (2017) which give a qualitative overview on European experiences with renewables auctions. Other work looks into the theoretical background and implications of certain design elements, see e.g. Kreiss et al. (2017, 2019). Specific country cases also exist, as for example Kylili and Fokaides (2015) who evaluate the functioning of an auction-based support scheme in Cyprus and Wigand et al. (2016) who summarize case studies across Europe.

    By remodeling the auctions and looking into the detailed agent-based modeling outcomes, i.e., costs, distribution of bidder types and dropout rates, we enable a better understanding of the decision processes underlying the auctions and motivating the participants. Agent-based modeling allows to simulate different forms of behaviour and makes the underlying procedures visible. We furthermore have some of the advantages of econometric analysis in our model, as we make use of empirical data and analyse the short time series of auctions that already took place before modeling the auction participants. By comparing model results to empirical outcomes, this paper aims to provide an explanation for the steep drop in bid prices. The findings are also relevant in the eye of a current legislatory change: the Bundeslanderklausel/Freiflachen-Offnungsverordnung. This new law allows the German federal states (Bundeslander) to come up with their own restrictions or open their disadvantaged arable land for tendering of ground-mounted solar P V. This change in legislation led to an opening of these formerly restricted areas for upcoming auctions. (1) We show through our modeling how this could influence future auction outcomes and discuss the resulting policy implications.

    The structure of our paper is as follows: First, some background information is given on the German solar PV pilot to introduce the topic. Next, an auction-theoretic background of bidding behaviour in auctions is given--and the limitations for theoretical analysis of repeated multi-unit auctions are explained. We then describe our agent based model, which incorporates implications of the theoretical analysis wherever feasible, but also insights from the empirical auction outcomes. This model then simulates the auction pilot with the given parameters on design and our knowledge on agent distribution in the German electricity market as well as on the price development of PV modules and generation of electricity from large-scale solar PV. Empirical auction outcomes are used to improve our modeling, however without pre-empting our model results. They instead allow for an optimal depiction of the distribution of participants in terms of e.g. costs and project sizes in the German large-scale PV sector. In the results section, we explain the bid prices and bidder distribution and evaluate how bidding evolved over the respective rounds. Specifically, we show the price development as compared to the actual prices, the distribution of bids over the six rounds and insights into the behaviour of those bidders who submit bids for the restricted arable land areas.


    In the German ground-mounted solar PV auctions, the auctioneer is the German federal network agency (Bundesnetzagentur). A sliding feed-in-premium to support large-scale solar PV installations for a support period of 20 years is tendered (Bundestag, 2017). The auctioneer publishes the successful capacity amounts in detail. The lowest and highest accepted bids together with the weighted average winning bid are also made public. The actual bid prices and project costs remain private information.

    In the auctions, participants submit their (sealed) bid in each round. Specifically, the bid contains a price in ct/kWh and a corresponding capacity in kW (kilowatt) of their individual projects. The location of the project is also submitted (Bundestag (2017), [section] 30), such that the auctioneer is immediately able to differentiate between disadvantaged arable land which is per definition not suitable for farming in its current state (in the following just referred to as arable land for simplification purposes) and other areas, namely the area adjacent to a highway or railway or a converted area which was previously used for military, business purposes, infrastructure or housing (named converted areas in the following). The difference between these two areas is a crucial feature of the German PV auction scheme, as the former is restricted due to reservations by the German farmer's association (Bauernverband), see e.g. Deutscher Landwirtschaftsverlag GmbH (2015). The described procedure holds for PAB and UP. Bids are chosen while the cumulative amount of capacity is lower than the demand. Immediately after the procured quantity is reached or surpassed for the first time, the auction round is closed. This procedure is implemented into our model in all its specifications (see also: Anatolitis and Welisch, 2017).

    The German PV pilot consisted of six rounds, three in 2015 and 2016 respectively. In each round quantities between 125 and 200 megawatt (MW) were tendered. The ceiling price started at 11.29 [euro] ct/kilowatt hour (kWh) (2) then decreased to 11.19 ct/kWh and then 11.09 ct/kWh for the remaining four rounds (Bundestag, 2017). Two of the pilot rounds (rounds 2 and 3) were held as UP auctions and the rest were PAB. Their results, which will also be discussed in the following were a sharp decrease in support costs, (3) which were previously admnistratively set with a feed-in-tariff system.


    Since 2015, the support payments for ground-mounted solar PV plants in Germany, are determined by repeated, static multi-unit auctions. This section will elucidate this auction design as well as the characteristics of the participating bidders auction-theoretically. It will explain how we transferred this framework into an agent-based modeling approach and where there are limitations of transferability between theory and practice. We will start with the auction-theoretic basics of the revenue equivalence principle and then add complexity by including repeated games, asymmetric bidders and common values.

    The multi-unit characteristic is common to most auctions for renewable energy support, meaning that more than one project is awarded to supply the auction demand. In the analysed case of large-scale ground-mounted solar PV auctions in Germany, the auction volume in the first round was 150 MW of installed capacity, the maximum bid volume was 10 MW and thus at least 15 projects had to be awarded. As there were also smaller bid volumes, in total 25 projects were awarded in the first auction. The demand of installed capacity is considered to be homogeneous. (4)

    The two most common formats are the so called PAB or discriminatory auction and the UP auction. The latter can be further divided into the highest accepted or the lowest rejected bidder setting the price. In a simplified setting, PAB and UP auctions have the same expected revenue, given only bidders with single unit supply participate (Engelbrecht-Wiggans, 1988). Nevertheless, bidding behaviour is quite different between the two pricing rules (Weber, 1983). Especially, it can be shown that a UP auction where the lowest rejected bid determines the uniform price (which is not the case in the German ground-mounted solar PV auctions) is incentive compatible: it is the optimal strategy for a participating bidder to submit her true costs independent of the bidding strategy of every other bidder. The bidder cannot improve her expected profit by deviating from this strategy. A participant's bid does not determine the price she receives in a UP auction, which is different in case of winning in a PAB auction. Thus, in a PAB auction a bidder has an incentive to exaggerate the costs in the bid in order to gain a...

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT