When Power Companies Break Their Promises
How strategic outages cost Alberta electricity consumers $600 million
Imagine you own several power plants. Some of them are locked into long-term contracts that require you to deliver electricity at a fixed price. Others sell freely on the spot market, where the price fluctuates with supply and demand. Now imagine demand spikes and spot prices soar. Your contracted plants keep earning the same modest rate, but your uncontracted plants are making a fortune. What if you could make the spot price go even higher? All you would need to do is take one of your contracted plants offline, declare an “emergency” outage, and let the market tighten. You lose a bit on the contract side, but every other megawatt you sell just became much more valuable.
That is exactly what happened in Alberta.
The story
In the fall of 2010, a major electricity supplier in Alberta developed what internal documents later revealed to be a Portfolio Bidding Strategy. The idea was to coordinate forced outages of coal-fired power plants under long-term contracts, tightening the market so that the firm’s other plants, the ones selling on the spot market, would earn much higher prices. The contracted plants were the sacrifice; the rest of the portfolio was the payoff.
The outages were disguised as emergency maintenance. But the timing told a different story: they consistently coincided with periods of high demand and low wind output, precisely when removing capacity would have the largest price impact. Internal emails between traders and plant operators confirmed the coordination. One trader wrote to a manager after an event that it was “a great example of the ongoing coordination we have to optimize outages.” The manager replied that prices had jumped to over $400 during peak hours and called it “great value.”
The scheme unraveled when two buyers of long-term contracts filed complaints. The Alberta Market Surveillance Administrator investigated and the Alberta Utilities Commission ultimately concluded that the firm had “unfairly exercised its outage timing discretion for its own advantage.” A $56 million settlement followed.
Using machine learning to detect manipulation
How do you prove that an outage was strategic rather than a genuine emergency? In a paper published in The RAND Journal of Economics with Étienne Billette de Villemeur, we propose a machine learning approach to answer exactly this question. Using hourly bid data and plant-level production records, we train models to predict what supply and prices would have looked like had the outages not occurred. By comparing these counterfactual predictions with what actually happened, we can detect anomalous bidding patterns, estimate the price impact of each event, and quantify the damages.
The numbers are staggering. We estimate that the strategy delivered up to $67 million in extra revenues to the firm, but the cost to Alberta’s electricity consumers was far greater: up to $600 million in additional procurement costs over the period, a 17% increase in the province’s annual electricity bill. The gap between private gain and public cost reflects the equilibrium effects of market manipulation: when a large player withdraws capacity, all prices rise, not just the manipulator’s.
A blueprint for detecting market manipulation
The approach we develop is not specific to electricity. Any market where participants make sequential commitments and where bid or transaction data is available could benefit from similar methods. Financial markets, commodity exchanges, procurement auctions: wherever a firm can act on private information to shift prices in its favor, machine learning can help regulators build the counterfactual scenario needed to detect and measure misconduct.
In the Alberta case, we show that changes in bidding strategy during the outages carried a clear statistical signature. The firm’s bids revealed its intent even before the price impact materialized. This kind of early warning signal, detectable in real time from publicly available bid data, could allow regulators to intervene sooner rather than investigate years after the fact.
More broadly, our paper shows the other side of a common policy assumption. Long-term contracts are widely seen as a tool to limit market power: if a firm has already sold most of its output forward, it has less incentive to manipulate spot prices. But when contracts include imperfect commitment mechanisms (penalties that are too low, maintenance discretion that is too broad), they can actually create new opportunities for manipulation.
The lesson extends well beyond electricity. Anywhere contracts are sequential, penalties are imperfect, and one party has the power to walk away at a strategic moment, there is room for this kind of conduct. Machine learning gives us new tools to find it.
Benatia, D. and Billette de Villemeur, É. (2025). “Strategic Reneging and Market Power in Sequential Markets.” The RAND Journal of Economics, 56:3-34. DOI.
