Bower and Bunn (2000) present an agent-based simulation model in which generation companies are represented as autonomous adaptive agents that participate in a repetitive daily market and search for strategies that maximise their profit based on the results obtained in the previous session. Each company expresses its strategic decisions by means of the prices at which it offers the output of its plants. Every day, companies are assumed to pursue 2 main objectives: a minimum rate of utilisation for their generation portfolio and a higher profit than that of the previous day. The only information available to each generation company consists of its own profits and the hourly output of its generating units. As usual in these models, the demand side is simply represented by a linear demand curve.

Such a setting allowed the authors to test a number of potential market designs relevant for the changes that have recently occurred in England and Wales wholesale electricity market. In particular, they compared the market outcome that results under the pay-as-bid rule to that obtained when uniform pricing is assumed. Additionally, they evaluated the influence of allowing companies to submit different offers for each hour, instead of keeping them unchanged for the whole day. The conclusion is that daily bidding together with uniform pricing yields the lowest prices, whereas hourly bidding under the pay-as-bid rule leads to the highest prices.

The introduction of NETA in the UK represented a good opportunity to test the usefulness of large-scale agent-based simulation to provide some insights on market design. The agent-based platform enables a detailed description of the market, taking into account discrete supply functions, different marginal costs for each technology, and the interactions between different generators. Bunn and Oliveira (2001) followed that work by developing a simulation platform that represents, with much more detail, the way that market clearing in NETA was designed to function. This platform models the interactions between the Power Exchange and Balancing Mechanism; considers that generators may own different types of technologies; considers an active demand side, including suppliers; and takes into account the learning dynamics underlying these markets as a process by which a player selects the policy to use in the game by interacting with its opponents. In later work, they adapt and extend this simulation platform to analyse if the 2 particular generators in the Competition Commission Inquiry had gained enough market power to operate against the public interest (Bunn and Oliveira 2003).

Researchers at Argonne National Laboratory in Chicago have developed the Electricity Market Complex Adaptive System (EMCAS) model (North et al. 2002; Veselka et al. 2002). Like the above-mentioned simulation models developed at the London Business School, the EMCAS model is an electronic laboratory that probes the possible effects of market rules by simulating the strategic behaviour of participants. EMCAS agents learn from their previous experiences and modify their behaviour based on the success or failure of their previous strategies. Genetic algorithms are used to drive the adaptive learning of some agents, and pool, bilateral contract and ancillary services markets are included. The EMCAS model is arguably the most sophisticated agent-based electricity model to date, embodying more development hours than other simulation models of its type.

At Iowa State University, Leigh Tesfatsion and her colleagues have examined market power experimentally in an agent-based computational wholesale electricity market operating under different concentration and capacity conditions (Nicolaisen et al. 2001). Pricing is determined by a double auction with discriminatory midpoint pricing. Buyers and sellers use a modified Roth-Erev individual reinforcement learning algorithm to determine their price and quantity offers in each auction round. High market efficiency is generally attained, but the aggregate measures used are too crude to reflect the opportunities for exercising market power that buyers and sellers face. Their results suggest that the precise form of learning behaviour assumed may be largely irrelevant in a double auction system.

Taylor et al. (2003) developed an agent-based model to simulate the complexity of the large-scale Victorian gas market in south-eastern Australia. The model can be used to elicit possible emergent behaviour that could not be elicited otherwise under an uncertain future of deregulation and restructuring. Like an electricity market, the complexity in the gas market derives from the uncertain effects of a multiplicity of possible participant interactions in numerous segments, such as production, storage, transmission, distribution, retailing, service differentiation, wholesale trading, power generation and risk management. The agent-oriented programming platform devised for this work has the potential to overcome the limitations of traditional approaches (discussed earlier) when attempting to operate in changing environments. A similar platform has been adopted for our National Electricity Market Simulator.

Among the 90 registered participants in the NEM, most fall into categories based on the role they perform in the market. ^[3] These categories are generators, Transmission Network Service Providers (TNSPs), distribution network service providers (DNSPs), market network service providers (MNSPs), customers (retailers and end-users), and traders. NEMMCO, key regulators (such as NECA and the Australian Competition and Consumer Commission), and a number of other organisations (like the Council of Australian Governments and the Australian Greenhouse Office) are among the additional actors. For convenience, in the remainder of this chapter we shall refer to all of these relatively autonomous actors as agents.

Agents are intelligent and adaptive, meaning that they make operational and strategic decisions on the basis of the information available to them and the market’s rules, and they modify their own strategies on the basis of new information that comes their way. In many decision situations, however, some agents can do no better than exhibit purposive but contingent behaviour. Although they have specific goals of their own (for example, to maximise profits, market share, utilisation factors, etc.), their ability to attain these goals is largely beyond their own control. Furthermore, earlier research on the NEM has shown that demand side participants have very little ability to influence outcomes compared with those on the supply side.

In an adaptive market such as the NEM, no single agent can control what all the other agents are doing. Many of the collective outcomes are not obvious, because simple summation to linear aggregates is impossible. The outcomes are governed by and dependent on a system of nonlinear interactions between individual agents and the market environment, i.e., between agents and groups of other agents or between agents and the whole market. In such limited-information situations, some agents can and do exert more influence than others. Groups of agents may benefit from informal partnerships or tacit collusion. Bidding and rebidding is unlikely to be competitive under these circumstances, since individual agents or coalitions may find it more profitable to search for opportunities to exert market power.

In electricity markets, we often find examples of locally interacting agents producing unexpected, large-scale outcomes. Problems experienced by the Californian electricity market provide a vivid illustration. During the summer of 2000, wholesale electricity prices in California were nearly 500 per cent higher than they were during the same months in 1998 or 1999 (see Figure 11.4). This explosion of prices was not only unexpected but sustained. Unlike previous price spikes observed in the US wholesale markets, the California experience proved to be more than a transient phenomenon of a few days’ duration. It persisted until roughly mid-June 2001. Increases in gas prices and consumer demand, reduced availability of power imports, and higher prices for emissions permits are known to have contributed to price rises. But these market fundamentals are insufficient to explain the extraordinary gap between realised prices and competitive benchmark prices. Evidence of capacity withholding by suppliers (generators or traders) has been found, and this, together with other factors, is thought to have led to such remarkable price increases and market manipulations during 2000/2001.

Although some observers have argued that the NEM spot market is more robust to gaming than its Californian equivalent, the NEM is still vulnerable to market power.^[4] For example, generator companies can profit more than customers by exploiting the bidding rules set by the National Electricity Code Administrator (NECA). Because no single body has control over market outcomes, agents exploiting anomalies can reap rich rewards. The rest of the market suffers accordingly. Recent spot prices have been observed to fluctuate from a low of around $5/MWh to a high of $10,000/MWh. Some of this volatility is caused by the weather, industrial action or diurnal/seasonal peaks and troughs in demand, but a significant proportion is not. Rather, it is an intrinsic by-product of the way the trading system and its rules operate.