Turing off the lights: Consumer-allowed service interruptions could control market power and decrease prices. (The California Crisis).

Rassenti, Stephen J. ; Smith, Vernon L. ; Wilson, Bart J. 等

EARLY LAST DECADE, CONGRESS PASSED legislation that allows the deregulation of wholesale electricity production and prices in the United States. Under the legislation, states or regions that implement deregulation must develop restructuring plans for the power industry. The plans, in part, must specify the auction market rules for determining the hourly wholesale price of energy that wholesale producers sell to retail distributors, who in turn sell to their customers.

In areas that have implemented deregulation, the process has resulted in market designs hammered out by regulators, consultants, industry representatives, and various power-marketing intermediaries. The resulting plans employ supply-side bidding mechanisms for the hourly spot market, coupled with varying degrees of freedom for producers and distributors to arrange bilateral long-term contracts.

In California, producers submitted price-conditional offers of energy to the spot market, known as the California Power Exchange (California PX). That way, producers supplied the instantaneous demand at the price that balanced production with consumption.

However, the widely fluctuating rates paid by retail distributors to producers were not reflected in the rates that retailers charged most of their customers. Until the recent demise of the PX, most of California's wholesale power was resold to consumers at fixed rates per kilowatt-hour, after payment of a fixed monthly access charge. That meant that any consumer would be guaranteed that his instantaneous demand would always be satisfied at the regulated delivery price. Thus, deregulation at the wholesale level was not coupled with any significant change in the retail technology for delivering, metering, and charging the end use consumer.

THE MUST-SERVE MARKET

In order to appreciate the implications of such a system, imagine what would happen if, say, the airline industry operated under similar rules. Airlines would be required to charge all passengers an identical, regulated monthly access fee and a fixed price per mile traveled, regardless of the flyer's destination, time and day of flight, and even his willingness to pay. Also, as part of the service, airline seats would have to be available to any passenger whenever he wants to fly, even if it is on the busiest day of the year. That would mean that, in order to operate in accordance with regulation, the airlines would have to purchase and maintain enough airplanes to satisfy peak demand, yet always charge the same fare regardless of how many planes are in use at a given time.

Fortunately, that is not the way the airline market works. For airlines, efficient pricing is very sensitive to peak versus off-peak demand; tickets are less expensive at times when fewer planes are needed to satisfy demand and there is plenty of capacity at the airports involved, and more expensive when more planes are needed to satisfy demand or airport capacity has reached its limit. Competitive markets will tend to yield higher prices for peak users because it is their travel demands that require the airlines to buy more airplanes and restrict the number of passengers that they can service at an airport. (The airline market did not always work that way; among the first casualties of airline deregulation were the "fair" pricing policies under regulation that tended to be insensitive to time and destination of travel.)

However, the electricity industry operates in a "mustserve" market similar to our hypothetical airline market. That must-serve structure was inherited from America's traditional, rigidly regulated power system that, because of political pressures, placed reliability of electricity flows above all other goals, regardless of cost. Under regulation, the cost was collectivized by averaging it across all users, regardless of an individual consumer's willingness to pay to keep the lights on. The local utility was expected to maintain service, or restore it quickly (even in inclement weather), and spread the cost of super-reliability over all customers. That cost included the maintenance of substantial reserves in generation and transmission capacity, with regulators establishing standard electricity rates high enough to ensure that utilities received a return on the capital so invested.

Thus, system reliability and the capacity to satisfy all retail demand were exclusively a supply-side adjustment problem. The consequence of that mindset was uncontrolled cost creep that increased to a gallop as utilities invested large sums of money in expensive plants in an effort to satisfy peak demand, and then were granted rate adjustments to cover the expenses. Reaction to that pricing became part of the political outcry for deregulation.

DEREGULATION BEGINS

Implicitly, the process of deregulation assumed that the built-in supply-side bias did not require fundamental rethinking when it came time to design spot markets for the new world of competition. In the formation of market institutions, however, the devil is always in the details.

Beginning three years ago in Midwestern and Eastern markets, peak prices hit short-run levels of 10- to 100-times the normal price level of $20-$30 per megawatt hour. That was the predictable direct consequence of a market design in which completely unresponsive spot demand impinges on a responsive discretionary supply.

California's crisis Last fall and winter, the California PX was plagued by exorbitant increases in spot prices. Investor-owned local distribution utilities, which were required to obtain all their power through the PX, were pushed to the edge of bankruptcy because they were forced to buy all of the electricity their customers demanded -- even if the demand resulted in the radical bidding up of prices -- yet the distribution utilities then had to resell the power to their customers at low fixed retail rates (with a temporary exception in San Diego).

The crisis led to political action that imposed price caps on the market. But, of course, the caps only reduced power producers' response to the shortages. Political action also forced unwilling out-of-state vendors to continue selling energy to the financially troubled distributors for fear that, if they cut off the flow of electricity, the distributors would go bankrupt and the suppliers would never see the money already owed them.

Changing demand But what would have happened if, instead of following a must-serve philosophy, distributors and their customers had cut back -- or even cut out -- electricity use during peak hours? Would such a decrease in peak demand lead to lower peak wholesale prices and more supply? To answer those questions, we carried out a series of economic experiments that examined consumer response in such a market. We found that, if as little as 16 percent of buyers during peak demand voluntarily accepted an interruption of power (in exchange for financial compensation), the high level of prices and the tendency for upward price spikes to occur when electricity supplies are tight would have been completely avoided.

In such a system, the distributors would submit bids to customers, offering to purchase the ability to interrupt service. The interruptions would be voluntary, perhaps implemented by radio-control signals sent by the distributor to selectively turn off some of the customers' appliances. Or perhaps the distributor could adopt real-time prices that would provide customers with financial incentive to "self-interrupt" their service by cutting use during peak hours.

In California, such voluntary acceptance of interruption certainly seems preferable to last winter's involuntary area-wide blackouts that affected all consumers alike, including those operating elevators, energy-dependent production lines, and computers. What is more, in such a decentralized market, no price controls would need to be imposed, no involuntary penalties would need to be inflicted for high consumption, and no action would be taken against sellers for "unjust enrichment."

OUR EXPERIMENTS

In testing our models to see if a voluntary interruption scheme would be effective, we sought to answer two questions:

* What is the effect of supply-side market power on competition and prices? Specifically, what would be the difference between a market in which there is concentrated ownership of certain generator types vis-a-vis a market in which there is no such concentration of ownership?

* What is the effect on competition and prices when there is demand-side bidding by the distributor, who profits from buying in the spot exchange and reselling to retail customers at a given price schedule?

In summary, we compared data on prices for four different treatment conditions: supplier market power or not, coupled with demand-side bidding or not. Our experiments used profit-motivated subjects who were paid their earnings in cash at the end of each experiment. Buyers profited by buying low, sellers by selling high, with all participants using their own devices to work out profitable actions in light of experience.

We compared data from experiments that implement our theoretical demand-side bidding in a both-sides active, sealed bid-ask auction, with data from a one-sided auction in which only generator owners are submitting price conditional offers of power to the market. The latter type of auction is the one that characterizes most closely the exchanges found in California, the Midwest, and on the East Coast.

Representing supply The supply schedule underlying all the experiments is represented by arraying generator marginal costs from lowest to highest. That arrangement is appropriate because power companies typically run their more efficient units for longer periods of time and only operate their least-efficient units at peak times when they are straining to keep up with demand. In our representation, each generator type has a constant marginal cost up to its capacity --a common approximation in power systems.

The resulting supply schedule is shown in Figure 1 as the line marked by the coming-on-line of various generators operated by suppliers S1, S2, S3, S4, and S5. In that representation, we distinguish three standard types of generators:

* Low-cost, base-load generators that are normally running and supplying power continuously at all times of the day. In Figure 1, generators with a running cost of 20 per unit of output (e.g. $20 per megawatt-hour) are shown on the lowest supply step. The generators' total supply capacity is 12 units (e.g. 1,200 megawatts) of power. At any price above 20, all of those generators are profitable.

* Intermediate cost ("load follower") generators that are normally running at all times except during the low-consumption off-peak hours. Those are shown in Figure 1 as incurring a cost of 76 per unit of output, with a total capacity of 10. The generators are profitable at any price above 76, as are the base-load generators. So, at prices above 76 but below when the high-cost units come on, the total energy supplied is 12 + 10 = 22.

* High per-unit cost generators that are normally used only during peak demand. Four of those generators are shown in Figure 1 with per-unit cost of 166, and three more with unit costs of 186. Hence, the supply moves up to 26 units when the going price is above 166, and up to 29 units when the price is above 186.

Representing demand For all of our experiments, demand was represented by a three-step resale value schedule. As shown in Figure 1, the first step -- a large spike of demand at price 226 -- is the so-called "must serve" demand that depicts the fixed retail price for all customers requiring uninterruptible service. Here, we represent the U.S. industry in its current state in which most of the demand is not responsive to price. (Of course, under full deregulation we would expect that condition to change overtime, depending on prices and incentives to conserve usage.) With full retail pass-through of wholesale energy cost-based prices, we would expect a rapid adoption of both the culture and the technology for implementing price responsiveness.

The first block of inelastic demand varies over a daily cycle that begins with the "shoulder" demand (16 units), increases to the "peak" demand (21 units), returns to the shoulder demand, then decreases to "off-peak" demand (6 units). The cycle is repeated each "day" in our experiments, for a total of 14 cycles. Thus, the relatively steady peak and off-peak hourly price sequences for typical wholesale demand are each combined into one simplified block (representing several hourly markets) for the peak and another for the off-peak periods of the day.

We also simplified the transition demand sequence from off-peak to peak and vice versa by representing the transitions with a single steady block of "shoulder" demand. Our small simplifications enabled us to focus on the key issues we wanted to study while still capturing the essence of the daily natural cycle in demand in all electrical delivery systems.

The market at work Electricity distributors profit by reselling wholesale power purchased in each auction period during the day at their fixed three-step resale prices. Of course, in our theoretical market that includes voluntary interruptions, distributors and their customers will not arrange to cut service during the must-serve first step. However, the second and third steps can be interrupted, but the customer has agreed to continue accepting service if the price does not eclipse the step-two price (206) and step-three price (96). The lower resale prices for power represent discounts by the wholesaler to various retail customers in return for their willingness to have part or all of their deliveries interrupted at the discretion of the wholesaler. Those on the second step, however, never pay more than 206, and those on the third more than 96.

The resale values remain identical for two treatments: No Demand-side Bidding and Demand-side Bidding. In the absence of demand-side bidding, all of the listed resale values and associated quantities demanded are completely revealed as bids to the market by a robot bidder. That eliminates strategic bidding behavior by buyers, while strategic behavior on the seller side is fully active.

With demand-side bidding, four profit-motivated human subjects function as wholesale buyers, and are free to bid their respective demand each period at their discretion. They can reveal (bid their true demand as in Figure 1), under-reveal, or withhold demand strategically (by bidding at prices below their demand) in the same way that the five sellers are free to reveal, under-reveal, or withhold supply from the market by submitting asking prices higher than the true marginal cost.

The spot market pricing mechanism Where there is no active demand-side bidding in each spot pricing period, sellers privately submit a schedule of asking prices for their generation capacity. The aggregate of all generator offer schedules is obtained by arraying all of the individual offer price steps from lowest to highest. The offer schedule is then "crossed" with the actual demand (resale value) schedule to determine a single uniform market-clearing price where the two schedules intersect. All generators receive, and all wholesale buyers pay, the same price. That treatment parallels the energy markets in most regions of the United States that have instituted hourly spot markets, except that in our markets we make no provision for bilateral contracts secretly priced outside the exchange; all energy transfers pass through the spot market.

In the second experimental price mechanism treatment, there is active demand-side bidding by wholesale buyers. The buyers, in addition to the sellers, must privately submit schedules of bid prices for the purchase of electricity in each spot pricing period. The aggregate of the bid array, ordered from highest to lowest, is crossed with the offer schedule of sellers. Where the bid-ask (reported demand and supply) schedules intersect determines a single uniform price applicable to all buyers and sellers. The two-sided auction market allows buyers potentially to neutralize the expression of seller market power by under-revealing their resale values or withholding some of their demand for interruptible electricity.

UNILATERAL MARKET POWER

How do we create market power in generator ownership? To answer that question, look at the second step of the supply schedule, and the shoulder demand that intersects that step. Above each generator segment in the step there is a list of the generator companies that own capacity on the step. Seller S1 owns two generators that operate profitably at prices set at the intermediate level of demand, and S2 similarly owns two units.

Gaming Given that ownership pattern, either S1 or S2 can increase its profits unilaterally by withholding some of its capacity on the intermediate cost step. During the shoulder periods, the competitive price is equal to the marginal cost (76) of the intermediate generators. However, either S1 or S2 can unilaterally withdraw (not submit offers for) four units of production entirely so that the price rises to the third step of the supply curve (166), where four units of peaking generation capacity contest any further attempted increase in price.

Alternatively, either S1 or S2 can increase the offer price for its intermediate cost generator capacity so that the offer sets the market price. It is important to note that it requires only one of the two sellers, S1 or S2, to undertake that profitable action that reduces its volume sold but also benefits all other sellers. Either one of the producers will be even better off by not having reduced its sales volume if the other seller withholds supply to raise the price. Unless they tacitly coordinate their offers, each has an incentive to free-ride on the increased offer price of the other.

At the competitive price of 76,S1 and S2 both earn a profit of 224 [(76-20) x 4 units]. If either S1 or S2 raises the offer on its intermediate units to 166, the price-setter's profit rises to 584 [(166-20) x 4 units]. That unilateral deviation is even profitable at a price of 96 -- the third shoulder demand step -- where S1 and S2's profit would be 384.

Changing the system How can we change the ownership distribution of generators to eliminate market power during the shoulder demand periods? We can eliminate the market power incentives simply by transferring ownership of one of S1's and one of S2's intermediate cost generators to S4 and S5 respectively. We will call this the No Power treatment.

With that ostensibly minor reallocation of capacity at Nodes 1 and 3, not a single supplier can increase profit unilaterally by offering units at supra-marginal price levels in the shoulder period. If a single supplier raises its offer above 96, that supplier surely will not sell its intermediate units of capacity and, furthermore, will not increase the price received for its base load units. In that case, it is not profitable for any supplier to deviate unilaterally from revealing its marginal cost. Only if two suppliers -- through tacit coordination -- decided to raise their offers on the intermediate-cost (second-step) capacity could a supra-competitive price emerge.

Notice that, in both the Power and No Power treatments, no supplier can exercise market power during peak demands; all unilateral deviations are unprofitable. Even in the Power treatment, unilateral increases in offers by S1 and S2 to raise the price from the competitive level of 166 to the peak production costs of 186 result in a loss of profit of 360 [(166-76) x 4 units] from the intermediate units of production and yield a gain of only 80 [(186-166) x 4 units] on the base load units. That design, with generation competitive at peak demand levels but not at intermediate levels, illustrates the important principle that market power need not be associated only with peak demand conditions. Market power is about the ownership distribution of different generators classed by marginal cost, given a fixed and unresponsive demand.

S1 and S2 can exert some market power during offpeak demands by raising the offer prices on two units of base-load capacity, regardless of the allocation of intermediate capacity generators. The theoretical upper bound on the price during off-peak demand is 76, where price is contested by the marginal cost of intermediate generating capacity. We included some market power incentives in the off-peak demand as a common control providing such incentives across sessions in all treatments.

DESIGN AND HYPOTHESES

To test how active demand-side bidding and the ownership pattern of generators in electricity markets contribute singly and in tandem to the exercise of market power, we conducted 16 market experiments using students at the University of Arizona. Each session lasted approximately 90 minutes. We conducted four sessions in each of the four combinations of treatments: {No Power, Power} x (No Demand-side Bidding, Demand-side Bidding}. Each session was comprised of 14 market days that sequence through the different levels of demand as follows: shoulder, peak, shoulder, and off-peak.

Table 1 presents our hypotheses concerning the treatments effects of Demand-side Bidding and Power relative to baselines of No Demand-side Bidding and No Power. We hypothesized that active demand-side bidding will lower prices and decrease volatility for every level of demand when there is market power. We also hypothesized that the Power treatment will increase prices and volatility in the shoulder demand but not affect peak and off-peak prices when there is no market power.

Results Using the last 10 days of trading in each session, Figure 2 displays the average price and variances for each of the four combinations of treatments. Notice in the two upper panels that demand-side bidding reduces prices in the shoulder and peak periods for both the Power and No Power designs, and also reduces prices in the off-peak periods in the No Power treatment. The effect of demand-side bidding is particularly striking in the shoulder periods in the Power design. Demand-side Bidding completely neutralizes the exercise of market power during shoulder periods by reducing the average price from 137 to 86. Without demand-side bidding, the No Power suppliers push the price up to the value of the interruptible unit of demand, 96.

As hypothesized, we also find that shoulder prices are highly volatile in the Power design without demand-side bidding. The variance of price changes is 532 in the Power with No Demand-side Bidding treatment, but only 36 in the No Power with No Demand-side Bidding treatment. However, when Demand-side Bidding interacts with the Power design, the variance in shoulder periods drops from 532 to 30. In the Power design, demand-side bidding consistently and dramatically reduces the volatility of prices. In the No Power design, demand-side bidding reduces volatility in the off-peak and shoulder periods. Demand-side bidding effectively limits price volatility whether or not generators have market power.

PRICES WITH AND WITHOUT DEMAND-SIDE BIDDING

Figure 3 illustrates two time series paths of five experimental days for two sessions in the Power treatment, one with and the other without demand-side bidding. It is obvious that the session with demand-side bidding leads to lower prices in shoulder and peak periods. Furthermore, for the same time of day, prices are very stable across the days with demand-side bidding. Without demand-side bidding, the shoulder and peak prices vary rather noticeably from day to day.

Figure 4 illustrates why prices are lower and less volatile with demand-side bidding. The left panel illustrates the revealed bids and offers from both sides of the market in one Power session. The right panel illustrates the sellers' offers and the robotic bids at full resale value from another Power session. Notice that, in the No Demand-side Bidding, session, the offers are not very competitive because the sellers freely push the price up against the unresponsive buyers. In striking contrast, with Demand-side Bidding the lower bids by the buyers force the sellers to submit competitive offers resulting in lower prices in a 100percent efficient allocation.

CONCLUSIONS

In this article, we have reported on laboratory experiments that used profit-motivated subjects to examine the effect of market power on the level and volatility of prices in a supply-side auction market with fully revealed demand. Using Power vs. No Power baseline experiments, we extended the study to include measuring the effect of demand-side bidding, which was our primary treatment of interest.

While holding constant the cost and structure of supply, and the resale value and structure of wholesale demand, we inquired as to how effective is the introduction of profit-motivated demand-side bidders in restraining supply-side market power. Specifically, we measured the effect of demand-side bidding on both the level and volatility of wholesale prices. We did the comparisons in a technologically conservative environment in which most -- 84 percent of peak -- demand was what the industry and regulators call "must serve." In that sense, we studied a transition environment still influenced by regulatory rigidities likely to change and adapt over time to decentralized management by markets.

We found that, in the absence of demand-side bidding, suppliers who have market power because of the ownership distribution of generators are able to push up the general level of prices in all phases of the daily demand cycle -- peak, shoulder, and off-peak. The high average level of prices peppered with frequent upward price spikes parallels the recent experience in regional electricity spot markets such as California and the Midwestern and Eastern sectors. The effect of demand-side bidding is to reduce both the level and volatility of prices that emerge from the exercise of market power by generators. When there is no generator market power, the effect of demand-side bidding is to reduce the shoulder and peak levels of prices and to reduce price volatility at off-peak and shoulder demand periods.

Interruption's payoff The public policy implications of our experiments are evident: Wholesale spot markets need to be institutionally restructured to make explicit provision for demand-side bidding. Distributors need to incentivizei more of their customers to accept contracts for voluntary power interruptions. Distributors stand to gain by interrupting demand sufficiently to avoid paying higher peak and shoulder spot prices, and the savings can be used to pass on incentive discounts to customers whose demand, or portions of it, can be reduced or delayed to off-peak periods when supply capacity is ample. The technology and capacity for implementing such a policy have long existed and can be expanded, but incentives for introducing it have been inadequate.

Our policy proposal recognizes that adjustment to the daily, weekly, and seasonal variation in demand, and to the need to provide adequate security reserves, is as much a demand-side problem as it is a supply-side problem. The history of regulation has created an institutional environment that sees such adjustment as exclusively a supply responsibility. The result is an inefficient, costly, and inflexible system that has produced the recent price shocks and involuntary disruption of energy flows in California.

Demand-side bidding, coupled with the supporting interruptible-service incentive contracts, can eliminate price spikes and price increases, and reduce the need for reserve supplies of generator capacity and transmission capacity.

[Figure 3 omitted]

[Figure 4 omitted]

Table 1

Hypotheses of Treatment Effects by Level of Demand

 Effect of Effect of
Level of no demand side bidding demand side bidding
Demand and market power and market power

Shoulder Higher prices Lower prices
 and volatility and volatility
Peak ? Lower prices
 and volatility
Off-peak ? Lower prices
 and volatility

 Effect of
Level of demand side bidding
Demand and no market power

Shoulder ?

Peak ?

Off-peak ?
Figure 2

Market Power and Price

Experiment results for average prices and variances by treatment.

 Competitive price Demand side bidding

AVERAGE PRICES
NO POWER DESIGN

Off-peak 20 47
Shoulder 76 85
Peak 166 139

AVERAGE PRICES
POWER DESIGN

Off-peak 20 46
Shoulder 76 86
Peak 166 163

 No demand side bidding

AVERAGE PRICES
NO POWER DESIGN

Off-peak 40
Shoulder 95
Peak 176

AVERAGE PRICES
POWER DESIGN

Off-peak 64
Shoulder 137
Peak 179

Note: Table made from bar graph
 Demand side bidding

VARIANCE OF CHANGES IN PRICE
FROM DAY TO DAY
NO POWER DESIGN

Off-Peak 53
Shoulder 18
Peak 113

VARIANCE OF CHANGES IN PRICE
FROM DAY TO DAY
POWER DESIGN

Off-Peak 50
Shoulder 30
Peak 22

 No demand side bidding

VARIANCE OF CHANGES IN PRICE
FROM DAY TO DAY
NO POWER DESIGN

Off-Peak 245
Shoulder 36
Peak 108

VARIANCE OF CHANGES IN PRICE
FROM DAY TO DAY
POWER DESIGN

Off-Peak 315
Shoulder 532
Peak 83

Note: Table made from bar graph

READINGS

* "Controlling Market Power and Price Spikes in Electricity Networks: Demand-side Bidding," by S. Rassenti V. Smith, and B. Wilson. Interdisciplinary Center for Economic Science Working Paper, George Mason University, 2001. Available online at http://gunston.gmu.edu/bwilson3/working--papers.htm.

* "Efficiency and Income Shares in High Demand Energy Network: Who Receives the Congestion Rends When a Line is Constrained?" by S. Backerman, S. Rassenti, and V. Smith. Pacific Economic Review, 1997.

* "The Exercise of Market Power in Laboratory Experiments," by C. Holt. Journal of Law and Economics, Vol. 32 (1989).

*"Market Design and Motivated Human Trading Behavior in Electricity Markets," by M. Olson, S. Rassenti, V. Smith, and M. Rigdon. Published in the Proceedings of the 32nd Hawaii International Conference on Systems Sciences, 1999.

Stephen J. Rassenti, Vernon L. Smith, and Bart J. Wilson are affiliated with the Interdisciplinary Center for Economic Science at George Mason University. They can be contacted by E-mail at [email protected] [email protected] and [email protected].