Real-Time Pricing: Ready for the Meter
Real-time pricing: Ready for the meter?
Public Utilities Fortnightly; Arlington; Nov 1, 1998; Nainish K GuptaAlbert L
Danielsen
Abstract:
An empirical study using 1995 data that calibrates own-price elasticity for demand under real-time pricing for a group of 4 anonymous industrial customers is presented. Overall, the study confirms that customers can and do shift their usage patterns in response to real-time rates. The study breaks the data down into 2 sets, corresponding to prices below and above 6 cents per kilowatt-hour. This breakout shows that the coefficient of own-price elasticity was more likely to be negative for price levels above 6 cents than for prices below that level.
Full Text:
An Empirical Study of Customer Response
Evidence suggests a decision point at 6 cents per kWh, indicating that self-generation becomes a highly viable option at that price.
WHAT ROLE SHOULD REAL-TIME PRICING play in a deregulated electricity market? Can it serve as an incentive to induce customers to remain loyal to their power supplier? How do customers respond to price changes carried out under RTP tariffs? Real-time pricing programs are now being used as a proxy for market-based pricing. The recently passed deregulation bill in Illinois directs utilities to develop and offer real-time pricing to all customer classes as a proxy for market based pricing. Utilities are offering "virtual" products that operate very much like RTP and are aimed at providing customers access to market priced power, while still retaining the customer. Even the newly instituted locational marginal pricing scheme in the PJM pool represents an application of RTP. So far, however, the typical customer billed under realtime pricing has been a large industrial firm with a monthly individual peak demand of at least 1 megawatt and a 250-kilowatt demand peak coincident with the utility system peak. These customers are large and flexible enough to shift a major amount of usage from peak to off-peak periods when provided with appropriate financial incentives. RTP programs have not yet made it to the small residential customers. According to a newsletter published earlier last year by the Electric Power Research Institute, somewhere between two-thirds to three-quarters of all electric customers respond to real-time pricing by shifting load when RTP is incorporated into utility rate structures. According to EPRI, typical values for load shifts range from 0.02, for slightly responsive customers, to around 0.50, for the very responsive. Those customers that do respond show a load-weighted average of about 0.14 for within-day flexibility-a load response on the same day of the price change.Values for between-day flexibility typically run about the same. Thus, a doubling of the hourly price from a given reference level (i.e., 100-percent increase) will typically lead to a 14 percent shift in load from that hour to other hours of the day. As a general matter, based on evidence from a variety of sources, one can say that certain customer attributes appear relevant in producing different levels of response to RTP across customer groups. One such factor is on-site generation. Nearly all RTP customers with on-site generation show a significant load response; parameters that measure customer flexibility tend to average anywhere from two to 10 times the values seen for other responding customers that cannot generate their own power.
Another important factor is the previous willingness of a particular customer to nominate a portion of load as interruptible. Among those customers without on-site generation, those formerly designated as interruptible show a greater propensity (84 percent) to shift load than that exhibited by other customer categories (only 60 percent by comparison). Flexibility parameters for these previously interruptible customers average about twice the magnitude (.13) of comparable variables exhibited by the noninterruptible responders (0.6). Moreover, industrial customers in certain SIC groups (Standard Industrial Classification) tend to show a greater response to real-time pricing than do customers in other groups. Those seven SIC groups are reserved for: Mining (SIC code 14); paper (26); stone, clay, and glass (32); primary metals (33); machinery and equipment (35); water and sanitary services (49); and hospitals (80). For customers without on-site generation, 86 percent of those in the seven listed SIC groups respond significantly to hourly prices, as opposed to 65 percent of all other customers. Their flexbility parameters average about three times the magnitude of customers in other SIC groups (.12 compared to .04). Nevertheless, despite these general observations, few empirical studies have been published on customer response to RTP. To add to the available pool of evidence, in this article we present an empirical study using 1995 data that calibrates own-price elasticity of demand under real-time pricing for a group of four anonymous industrial customers. Overall, our study confirms that customers can and do shift their usage patterns in response to real-time rates. Our study breaks out the data into two sets, corresponding to prices below and above 6 cents per kilowatt-hour. This breakout shows that the coefficient of own-price elasticity was more likely to be negative (showing a drop in purchases of electricity as price increased) for price levels above 6 cents than for prices below that threshold level.
Is temperature relevant to the price response?
In our study the data were taken from the summer months. A large portion of the electricity could have been consumed by the air-conditioning and ventilation systems. To measure this, we obtained hourly weather readings and included temperature as an explanatory variable. Surprisingly, however, the effect of temperature on consumption of electricity appeared mixed and inconclusive in our study. One explanation might be that some industrial processes are sensitive to temperature and are shut down as the mercury rises, thus reducing overall power demand. Another possible explanation is that the presence of onsite generation may alter demand for electricity. Some Factors That Motivate Customers Average tatistics for customer response to real-time pricing can mask considerable variability among customers with different characteristics. Responsiveness depends on a number of key factors. These factors can be divided into the following three main categories: incentive, ability, and willingness. That is, the greater the financial incentive that a customer has to shift load, the greater is the likely response. The size of the incentive depends on the amount of price variation and the relative importance of electricity costs as a share of overall operating costs. Similarly, a customer's ability to shift load depends in part on factors such as the presence of onsite generation that can be used to replace purchased power, and the flexibility of the customer's production process. Examples of the latter include the sensitivity of the production process and the ability to store product on-site. Finally, even with an incentive and the ability to respond, a customer may have the willingness to take the initiative to review the transmitted hourly prices and change its operations. Various business needs or operating styles of both management and operating personnel are likely to affect how customers respond. Moreover, customers are dynamic optimizers. They don't aim to maximize efficiency or minimize costs at just a single moment. A customer decides on power purchases in part by comparing the current hourly price to anticipated future hourly prices. Thus a customer presently facing a high hourly price might decide to postpone consumption until a later hour, when lower hourly prices are expected, provided that the inconvenience does not outweigh the savings. Hence, past consumption also will affect present and future consumption. Also, high RTP prices usually last for four to six hours or more. The typical industrial process usually takes at least a couple of hours to come back on line once it has been interrupted. In such cases, most of the lost production is usually rescheduled for the next day, giving further illustration that changes in customer demand may lag price changes to a significant degree.
Study and Method
The data used in measuring these demand elasticities are taken from Pareto Electric Corporation (name changed to protect confidentiality of the utility). Pareto has 360 industrial customers, eight of which take service under RTP tariffs. Pareto introduced its one-part RTP program in June 1994. The data set consists of hourly price information and hourly consumption of electricity as metered by Pareto for a representative sample of four out of the eight industrial customers for the period June 1, 1995, to September 30,1995. Table 1 hows the characteristics of the customers included in the study. (Customers are identified only by SIC number and product line.) "Hours of operation" denotes the number of hours a day and number of days a week the customer consumes electricity. "Price threshold" marks the price at which the customer switches to self-generation or shuts down. "Peak load" is the maximum demand placed by the customer on Pareto Electric's generating system.
The model estimated is: Y=f (P,Y^sub t-1^, T), where Y represents the quantity of electricity, P equals price, and T equals temperature. The independent variable Y^sub t-1^ is included because the measurement of the consumption of electricity is lagged by 24 hours (minus one day) to account for the fact, noted above, that customers tailor consumption decisions in part to reflect not only the current hourly price but the anticipated future hourly price.'
The econometric method consists of sets of regressions using the Ordinary Least Squares (OLS) method. In the first set of regressions the data for all the months is pooled together for each customer and the demand equation is estimated for each customer. However, the customers are not all pooled together, since some information on differences between the customer responses to RTP might be lost in aggregation. (See Table 2.) In the second and third sets, the demand equation is applied to estimate prices both below and above 6 cents per kWh, the price threshold at which self-generation becomes a truly viable option, and at which point one might expect to see some variation in customer response. (See Tables 3 and 4.)
Analyzing the Results
Table 2 shows that when the data are not broken down by price, the estimated negative coefficients of elasticity2 indicate that customers can shift their usage patterns in response to real time rates. Moreover, there is a positive relationship between quantity of electricity consumed in the present period and the quantity of electricity consumed at the same period the day before, as indicated by the fact that the coefficients for lagged quantity (Y^sub t-l^) are all positive. Only one out of four customers (that being Customer 2) shows a figure significantly different from zero at the 5percent confidence leveL
The effect of temperature in this case appears confusing. All other things being equal, an increase in temperature should increase electricity consumption, and therefore the estimated coefficient should be positive. That is the case here only for Customer 3, for whom the elasticity is not significandy different from zero at the 5-percent confidence level. Table 3 shows the demand equation estimates for prices below 6 cents. The estimated own-price elasticity of demand coefficients are positive for three out of four customers. Also, for three out of the four customers, these own-price elasticity estimates are significantly different from zero at the 5-percent confidence level, as seen by the t values. This implies that customers do not significantly change their consumption for change in prices under 6 cents, that 6 cents is indeed a critical price mark. Table 4 shows the estimated demand equation for prices above 6 cents per kWh. The estimated own-price elasticity of demand coefficients are negative and are significantly different from zero at the 5-percent confidence level for three out of the four customers. The analysis once again supports our hypothesis that customers do shift their usage patterns in response to real-time rates. The effect of temperature on consumption of electricity seems to be mixed. The coefficient is positive for three out of four customers. All the temperature elasticities are significantly different from zero at the 5-percent level for Customers 1 and 2 and are significantly different from zero at the 1-percent level for Customers 3 and 4.
Do Prices Always Clear the Market?3
In our study we assumed that the supply curve of electricity facing a typical customer is perfectly elastic-that at a given price, the customer can consume unlimited quantities of electricity. That has been a standard assumption in all the demand analysis studies performed on time-of-use and time-of-day pricing, but remains valid if customer demand doesn't exceed available supply at the real-time price. To validate that assumption, we put together a forecast of customer loads and of the availability and cost of generation and transmission resources owned or otherwise available to Pareto. We did this to measure the possible effects on Pareto's prices caused by the change in consumption, to determine whether our findings might have been affected by customer demand.4 In our case, Pareto reserves the right during a system emergency to pre-empt its hourly confirmed prices and impose a price of $3.00 per kWh. Pareto may exercise this right upon a one-hour notice and no more than 12 times a year. Pareto defines a "system emergency" as (1) the unexpected loss of one or more generation facilities during a critical period, leading to system generation reserves falling below desired levels, or (2) the unexpected loss of one or more major transmission facilities, which affects Pareto's ability to deliver energy.
Nevertheless, we believe our assumption of a perfectly elastic supply curve for electricity remains valid. Here is one way of confirming that assumption. Large utilities have about 30,000 MW of generating capacity available. For the price of electricity supplied to increase by at least 50 cents per kWh, either one or both of the following events must occur: (a) A drop of 10 percent (3,000 MW) or more in generating capacity or generating reserves, or (b) an increase of 10 percent or more in demand. To put these numbers in perspective, consider that a large industrial customers will demand about 3,0004,000 kW per hour. Hence, for prices to increase sharply because of a change in customer demand, at least 1,000 customers have to change their demand by at least 1,000 kW (one megawatt) simultaneously. According to utility sources, this never happens.
Footnote
Information on the quantity of electricity self-generated by the customers was not available. A translog cost function could not be estimated, nor is there any calculation of cross-price elasticities between self-generation and Pareto's generation. Since these customers take electricity under Pareto's day-ahead RTP program, the dependent variable in the data set was corrected for first-order auto-correlation to account for any correlations between present and past consumptions of electric power.
2 A log-linear model is estimated to measure the customers own-price elasticity of electricity consumption. The advantage of using a log-linear function is that the coefficients of the variables are the estimated elasticities and that it is
[Footnote]
easy to compute. The disadvantage of estimating a log-linear function is a loss of information, as the relationship etween price (P) and quantity of electricity (Y) for these customers (as seen in a data plot) does not appear to be log-linear. However, it has been shown that alternative, complex, non-linear functional forms provide similar results. 3. This section is based on extensive discussions with Mike O'Sheasy, manager for rate design at Georgia Power Corp., and Raymond Vice, manager for operations and engineering at Southern Company Services.
[Footnote]
4. Fuel costs for electricity and alternative fuels such as natural gas and oil, together with Operations and Maintenance costs, the cost of new generating units, the capitalization structure and an allowed return on equity are used to develop a forecast of retail electricity prices by class. Customer forecasts, completed using regression models, feed into the various energy models. The short-term (one to three years) energy models use an advanced time series regression method (ARIMA, or "Auto Regressive Integrated Moving Averages", an econometric method used for time series analysis), which relies on the most recent observations to produce the forecast. Patterns are deciphered from the data, and the deviations from these patterns influence the forecast. Variables that drive energy sales for each customer class are then selected for their significance to that sector to develop the long-term (four to 20 years) energy models, which use an end-use methodology. These models represent the elements of energy use in fine detail. Each major energy-using activity-refrigeration, space heating, and the like-is identified and the corresponding energy consumption is specified.