THE POTENTIAL FOR REAL TIME ELECTRICITY PRICING IN SWEDEN Mattias Vesterberg∗ , Chandra Kiran Krishnamurthy†

and

Oben Bayrak‡

April 13, 2014

Abstract This paper explores the practicability of dynamic pricing, and in particular the incentives and possibilities for consumers to respond to such pricing. Using a unique and highly detailed data set from Sweden, price elasticities at the monthly level are estimated; the estimated price elasticities are insignificant. In addition, using a SUR framework, end-use specific load curves are estimated, with a view to analyzing how these correlate to possible restrictions on load shifting (e.g. the office hours schedule). Based on these results, it is not evident that in the short-run, households have the possibility of shifting heating and lighting consumption to off-peak hours. Further, the cost reduction from shifting load from “expensive” to “cheap” hours is computed to be very small ; roughly 2-5% daily cost reduction from shifting load up to 7 hours ahead. These results have important implications for Swedish energy policy, in particular for the Swedish government’s stated goal of real-time pricing. The success of real time pricing depends heavily on demand response which, the results here indicate, are unlikely to be large without modest investments in technology and a substantial focus on it from the retailers, who appear to have little to gain from this switch in the short run.

1

Introduction

With the restructuring of electricity markets there has been an increasing focus on efficient pricing of electricity. Sweden is not an exception, and introduced hourly retail pricing in october 2012 .1 Compared to average retail pricing, common as of today, a price that better reflects the true cost of producing electricity ∗ Corresponding

Author: Center for Environmental and Resource Economics (CERE), Umeå School of Business and

Economics and Industrial Doctoral School, Umeå University, Umeå 90187, Sweden. Email: [email protected]. Tel: +46 73 9906515 † Center for Environmental and Resource Economics, Umeå School of Business, Umeå University ‡ Center for Environmental and Resource Economics and Dept. of Forest Economics, SLU Umeå 1 Every household has the right to have hourly prices, without having to pay any extra for neccesary metering equipment. The hourly prices are set one day ahead on Nord Pool spot market, see section on Swedish energy market for a brief decription of Nord Pool

1

on a more dynamic basis (e.g. hourly pricing) will – at least in theory – give rise to substantial efficiency gains. These efficiency gains arise mainly from a more efficient allocation of consumption, leading to a reduction in the need for costly peak capacity. However, evidence for practicability of such a pricing scheme, and in particular the possibilities for consumers to respond to such pricing, is rather scarce. Specifically, the success of dynamic pricing requires substitutability of electricity within a given day2 . One way of empirically exploring the latter possibility would be to compare the timing of current within-day electricity consumption with possible restrictions on substitability, such as working hours and temperature variation. Unfortunately, the data required for such an analysis are rarely available and only a few studies (Bartels and Fiebig (1990, 2000); Larsen and Nesbakken (2004)) in the very large literature on electricity demand have been able to illustrate how residential electricity consumption varies within a day and how to use this information to understand possibilities for substitutability. The current paper adds to the sparse literature on this issue. This paper reports the results of a study, commissioned by the Swedish Energy Agency, conducted between 2005 and 2008 in Sweden, metering household electricity consumption at the appliance level for about 400 households. Each household had its consumption metered at 10-minute intervals for a period of between 15 and 90 days, for virtually all appliances (up to 46 different appliances for each household). In addition to the load data, both indoor and outdoor temperature were recorded for all households, as well as household characteristics. Data at this level of detail have rarely been available for most countries. Using this data set, we estimate end-use-specific price elasticities at the monthly level. Preliminary results suggest no significant elasticity to monthly price changes. Secondly, using a seemingly unrelated regression framework, we estimate end-use- specific load curves (conditional on household characteristics) and analyze how these correlate to possible restrictions (e.g. the office hours schedule) on substitutability of load within the day. That is, we do not explicitly explore substitutability of electricity, but rather analyze possible restrictions to substitutability. We find that household total load has two peaks corresponding, roughly, to the morning pre-office hours (6-8 AM) and evening post-return-to-home hours (6-9 PM) and this is the period when the (Nord Pool spot-) prices are at their highest. Unsurprisingly, we find that the largest part of the load by far tends to be heating, lighting and cooking. At end-use level, this analysis also sheds light on relatively intuitive facts; households use heating when it’s cold, lighting when it’s dark and cooking before they leave for work and when they get home. Based on these results, it is not evident that in the short-run, households have the possibility of shifting heating and lighting consumption to off-peak hours. Similarly, it would not make sense to cook food in the middle of the night even if that would be cheaper than cooking in the afternoon. 2 In other words, while price elasticity–when measued using daily data– implies a certain amount of within-day substitutability by definition, an important component of “flexible” pricing schemes for the Swedish context is substituability of load, rather than reduction of total load (which is neither envisaged nor needed, in the Swedish context). Therefore, price elasticity, in itself, is not sufficient since the elasticity might arise from a reduction of total load, as indeed was found in the case of Chicago in Allcott (2011).

2

Finally, to examine the incentives for load shifting, we calculate cost savings from shifting load under a real time pricing scheme. These calculations show suprisingly small cost savings; roughly 2-5 percent reduction in daily cost from shifting load up to 7 hours ahead, implying that the incentives for both households and retailers appear to be small, at least in the short run. In addition, using the recorded indoor temperature data, we find two important and rather unexplored aspects of electricity consumption which bear on short-run household-level consumption reduction: 1. households, either due to preferences or inefficient technology, have a relatively high indoor temperature, averaging above 20 ◦ C, much higher than anticipated3 2. average indoor temperature does not substantially fall after the peak (i.e. morning office-going) hours. These two facts illustrate a rather important point for the Swedish setting: that there exist substantial potential to reduce usage (with moderate technological investments, e.g. heat thermostat) without reducing utility. Our estimation results have important implications for Swedish energy policy, and in particular for the government’s stated goal of real time pricing (RTP). The success of this pricing scheme depends heavily on demand response which, our results indicate, are unlikely to be large without substantial investments in new technology and a focus on it from the retailers, who appear to have little to gain from this switch, at least in the short run, based upon our simple cost shifting experiments (see Section X). We turn now to a brief review of the literature, focusing mostly on studies using data at the daily or higher frequency, as well as on studies focused on Sweden, where available. As noted above, literature on sub-annual appliance-level electricity demand is sparse, and hourly data is especially rare. However, the conditional demand approach pioneered in Parti and Parti (1980) and refined in Bartels and Fiebig (1990, 2000) and Fiebig et al. (1991)4 is a way to overcome the lack of appliance-level data. The idea is to combine data on total load (metered at hourly or half-hourly intervals) with survey information on appliance holdings, and exploit the appliance holding heterogeneity to estimate contribution from each appliance to total load. Households with a particular appliance are compared to those without the appliance and the differences in total load are attributable to the appliance. The estimated coefficients, interpretable as the the mean contribution of each appliance to the total load are then used to produce daily load curves for selected appliances. Evidently, this method has an obvious disadvantage in an ability to estimate the load of appliances with high penetration rates such as TV, washing machine and lighting, i.e. end-uses likely possessed by all households. B&F partly solve this issue by combining survey data with real-time metering data. By optimally choosing what appliances to be directly metered, they are able to provide more precise 3 Recommended minimum indoor temperature of 18 degrees is listed on The National Board of Health and Welfare’s (Socialstyrelsens) website http://www.socialstyrelsen.se/publikationer2013/2013-11-29 4 Hereinafter in general referred to as B&F

3

estimates. In particular, they find that quite substantial gains may be achieved with only a relatively small number of meters. A random coefficient model is used to allow for variation in both intensity of appliance utilization between households but also for size or capacity for particular appliances. The mean response associated with each appliance is then estimated using the data from both households that were and were not directly metered. See also Hsiao et al. (1995) for a bayesian approach on combining metering data with conditional demand analysis. Larsen and Nesbakken (2004) compare conditional demand approach with an engineering model, ERAD, whose inputs include engineering knowledge regarding technical and other features of housing stock, enabling estimation of energy demand for space heating. They compare the numerical results from the two models and provide a few recommendations regarding choice of end-use approach and what questions to implement in household surveys designed to disaggregate electricity consumption. In particular, the most important drawback of the engineering approach is the high need of detailed information about household behaviour and technical features of appliances at average household level. This is also, as far as we are aware of, the only study for the Nordic countries. Kavousian et al. (2013) examine structural and behavioral determinants of residential electricity consumption using data on households’ total electricity consumption metered at 10-minute intervals, together with survey data on household characteristics including appliance stock.5 Their results show that weather, location and floor area are among the most important determinants of residential electricity consumption.6 In addition to these variables, number of refrigerators and entertainment devices (e.g., VCRs) are among the most important determinants of daily minimum consumption, while number of occupants and highconsumption appliances such as electric water heaters are the most significant determinants of daily maximum consumption. To our knowledge, only one paper has so far used the same data set as we do. Widén et. al. (2009) develop a model for computation of daily electricity and hot-water demand profiles from time-use data using simple conversion schemes together with data on daylight and temperature. The model is shown to make realistic reproductions of electricity demand for individual households and to generate well-corresponding load distributions when compared to available metering data. Turning to measures of price responsiveness, to our knowledge all estimates of price elasticity for Sweden are on annual data and hence no estimates of short run (i.e. sub-annual) price responsiveness exist. Further, very few papers provide estimates of end-use specific price elasticities7 . Brännlund et al. (2007) estimate 5 Contrary to B&F and Larsen and Nesbakken (Larsen and Nesbakken (2004)), they do not estimate any appliance level measures but focus instead on differences between peak and off-peak consumption. 6 Interestingly, they observed no significant correlation between electricity consumption and income level, home ownership, or building age. 7 Except for Parti and Parti (1980), none of the above mentioned papers on appliance load estimate price elasticities. The reason is lack of price variation in their data sets.

4

price elasticity for space heating and find estimates of −0.13. They then estimate separate elasticities for electric heating, oil heating and distric heating ranging from −0.71 to −0.24 depending on specification. Damsgaard (2003) finds price elasticities ranging from −1.35 to −0.37 depending on heating system (using micro data). Andersson (1997) estimate the long-run elasticity for electric heating households to −1.36. Krishnamurthy and Kriström (2013) estimate price elasticity for Sweden to be between −0.68 and −1 using OECD survey data. All of these estimates are from annual data and hence should be interpreted as long term responsiveness, at least in a RTP context. There is also some experimental literature on how consumers respond to dynamic pricing schemes. For example, the real-world but small-scale experiment by Swedish Elforsk Market Design8 and local utilities in southern Sweden9 analyze the short-run household response to staged price spikes in the interval of 3 to 10 SEK/kWh (Elforsk Report, Consumer response to peak pricing (2006)). The results from this experiment showed that the load was reduced by up to 50 percent at hours with price spikes, with households with mixed heating (i.e. electric heating combined with other sources of heating, for example wood stove) succededing in reducing total load the most10 . However, it is important to note that such price spikes are unusual in Sweden. As emphasized earlier, a clear understanding of both price responsiveness and baseline consumption patterns are natural inputs to any policy analysis concerning dynamic pricing, such as RTP. In particular, the success of RTP require that consumers are able to respond to price variation by re-allocating consumption within a given day11 . Before we elaborate more on this and how our results inform us on these important issues, we briefly review some of the relevant literature on efficiency gains from RTP. Note first that the benefits of RTP are typically obtained from a more efficient allocation of consumption, where consumption is shifted from “expensive to “cheap” hours. This then translates into re-allocations in capacity, with reductions in costly peak and mid-merit capacity12 . Borenstein and Holland (Borenstein and Holland (2003)) and Borenstein (Borenstein (2005)), in the context of the US, simulate the long-run effects of RTP and find significant increase in consumer surplus of 3 - 11 percent. The efficiency gains arise from a reduction in the need for peak capacity (and not just reduction in generation, as in the short run) but also from some reduction in mid-merit capacity in favour of more baseload capacity. This is accomplished even with very low price elasticities, of only -0.025. Savolainen and Svento (Savolainen and Svento (2012)) reproduce the simulations in Borenstein and Holland (2003) and Borenstein (2005) for a Nordic market setting. Specifically, they impose 8 Elforsk

(Electricity Research in English) is owned by the Swedish electricity industry. Energi and Vallentuna Energi 10 The load reduction from increasing prices to 10 sek/kWh was not significantly greater than for prices of 3 SEK/kWh. 11 Energy conservation is not the policy goal of RTP and is also not something the retailers are likely to be supportive about; see Section X for a breif overview of the market structure of electricity in Sweden. 12 Since benefits are obtained from changes in capacity, the magnitude of gains are likely sensitive to the actual generation technologies. (see next section on Swedish capacity) 9 Skånska

5

capacity restriction on nuclear and hydro power, which makes sense in a Nordic setting where the possibilities for expanding hydro and nuclear capacity is limited. Further, they also include tradable emission permits in their simulation, and find that RTP reduces the need for peak capacity as well as mid-merit capacity. As a further illustration, Holland and Mansur (2006) simulate the short-run effects of RTP in the US context, and find significant reduction in peak generation, but an overall increase in total generation. Further, the simulated welfare gains from this re-allocation are a very modest 0.24 percent. All the simulations above assume a constant-elasticity hourly demand function,implying at least some substitutability between hours13 . The finding of some slight increase in total demand , indicating that consumers shift load from peak hours to off-peak hours, might then be explained by this assumption. However, one of the few recent empirical studies evaluating an actual RTP scheme in Chicago by Allcott (Allcott (2011)), finds that even if consumers on RTP are fairly price elastic the gains are small; the estimated increase in consumer surplus amounts to roughly $10 per year or roughly 2 percent of annual household electricity expenditure (In fact, Allcott goes as far as suggesting that even if residential RTP might be theoretically sound, it might “provide an important real-world example of situation where this is not currently welfare-enhancing” (pp839)). Further, in his empirical analysis Allcott notes that RTP on net cause no load-shifting. Rather, the response to the pricing scheme is by energy conservation through reduction in load during peak hours, but with no increase of consumption during off-peak hours. To some extent this stands in contrast with the constant elasticity assumption made in Borenstein and Holland (2003), Borenstein (2005) and Savolainen and Svento (2012), since a constant elasticity demand should imply some load shifting. The very small benefits of, and therefore incentives to, RTP for any individual household might imply that consumers are less likely to respond to RTP in the hypothesized way – by shifting load from peak to off-peak. There might also be reason to belive that there are short-run restrictions on the households possibilitiy to respond, further enhancing this result. For example, working hours, outside temperature etc might impose restrictions on the shiftability of load within a given day14 . Small monetary incentives together with restrictions on substitutability might then explain Allcotts result of no net load shifting. Finally, in the context of Sweden, the Swedish Energy Market Inspectorate (SEMI), in their CBA of introduction of RTP in Sweden, estimate the benefits to society to be significant

15

and are in favour of

introducing RTP in Sweden, either on a volontary or mandatory basis (Lundgren et al., 2010)16 . The main benefit is from replacing peak capacity with increased demand flexibility. Specifically, the alternative cost 13 If

demand is completely inelastic, households will not shift any load at all. (2005)briefly discuss these issues in his sensitivity analysis, where he allow elasticity to vary with the demand level, first with elasticity increasing in demand levels and then the opposite with elasticity decreasing in demand levels. For the latter case, he finds that the efficiency gains are “much smaller than in the case in which demand is more elastic at peak times” and also smaller than with constant elasticity (Borenstein 2005 pp14). 15 1541 to 1989 million SEK depending on share of households on RTP. 16 For all households with more than 8000 kWh of annual consumption. This consumption corresponds to an electricly heated detached house 14 Borenstein

6

of the peak capacity (defined as capacity used less than 40 hours per year) is estimated to be roughly 5 SEK/kWh (roughly 0.56 €/kWh). In the SEMI report it is assumed that a substantial share – 40 percent – of all households will be on real time pricing by 2030, or roughly 60000 new RTP contracts per year.17 However, SEMI reports that during the five first months of the RTP program in Sweden only about 6300 households had adopted this new pricing scheme (SEMI (2013))18 . In this paper we explore the above mentioned restrictions by analyzing demand behavior in detail prior to the Swedish implementation of RTP. That is, our data do not cover households on RTP. However, we argue that by understanding current behavior we are able to empirically explore the possibilities and limitations of consumer short-run response to future RTP (or similiar dynamic pricing schemes). Specifically, using a unique and highly detailed data set on appliance-level electricity consumption in Sweden we estimate enduse specific load curves and discuss how these relate to possible restrictions such as working hours, outside temperature and daylight (or in the case of lighting, the lack of daylight). We argue that these conditions might impose significant restrictions on any short-run attempt to shift load from “expensive” to “cheap” hours. Further, we analyze the above mentioned issues about limited monetary incentives by calculating cost savings for both the household and the utility when switching from monthly prices to real time prices and report suprisingly small cost savings.19 Finally, we estimate price elasticities for both total consumption and end-use specific consumption (albeit to monthly prices). We find no significant monthly price elasticity for any end-use. Our results indicate that there are short-run issues to be solved before reaping the long-run efficiency gains from real time pricing. The rest of this paper is structured as follows. A brief description of the Swedish energy market is provided in section §2, followed, in section §3, by a description of the data used in this paper, together with a few summary statistics. section 4.1 describes the estimation of load curves, along with the results of this estimation, while computations regarding the cost of servicing different end-uses and cost changes as load is shifted are described in section 4.2. section 4.3 details the estimation of price elasticity for both total load and end-use specific consumption, and section §5 concludes. 17 This estimate is partly based on the idea that Swedish residential consumers in general are rather active on the electricity market. For example, between 2005 and 2010 the number of variable-price contracts (contracts with prices varying over months, as compared to longer-term contracts) has increased from zero to 30 percent of all contracts, indicating that many consumers are interested in flexible types of contracts. 18 One explanation of this rather low number is probably that as of spring 2013, only 41 (out of roughly 120) utilities provided RTP contracts (according to a survey by SEMI), and of these only a few had hourly prices on display. Most contracts required the household to call the supplier for the current price information. This indicate that the incentives for the utilities to promote RTP are limited. 19 Further, as Holland and Mansur (2005) points of, the incentives will be decreasing as more consumers adopt to RTP as this would reduce variability in prices and increase real time prices relative to the flat rate, and hence the potential monetary gains.

7

2

Swedish energy market

The deregulation of the hitherto highly regulated Swedish electricity market in 1996, following the example of other europeean countries, introduced competition between electricity supplying companies.20 This period also marked the beginning of a market integration with the other Nordic countries via a common spot market, the “Nord Pool”. Following this deregulation, the market price today is determined by demand and supply on the Nord Pool power exchange situated in Oslo, Norway. The day-ahead market, Elspot, is the main arena for trading power in the Nordic region.21 Here, contracts are made between the 360 sellers and buyers for the delivery of power the following day, and the price is simply set by supply and demand. Further, there exist a intraday maket, Elbas, to cover potential imbalances occuring between closing of Elspot at noon and delivery the next day. The average system price in 2012 was 0.27 SEK per kWh. At its peak in February, the daily average for the system price was 0.84 SEK per kWh, while at its lowest, at the end of July, it was 0.067 SEK per kWh. The bulk of electricity is produced by hydro and nuclear, constituting 45 and 39 percent of total production, respectively (figures for 2011, from Statistics Sweden). The remaining production is from thermal (co-generation) plants and windpower, together with some smaller sources of peak capacity22 . This peak capacity, loosely defined (but in line with previous literature, for example Savolainen and Svento (2012)) to be the technology with highest marginal cost (and hence least utilisation), consist of gas turbines and oil fired condensing power plants.23 In total, this peak capacity consist of roughly 3000 MW of installed capacity24 . According to SEMI (Lundgren et al. (2010)) the reserve capacity needed is likely to increase to 4000 MW as Sweden expands the share of wind power production. Turning to the demand side, Sweden has a high electricity intensity per capita – roughly 14000 kWh per capita – placing the country at the top ten in the world25 . This is explained both by cold winters and an energy intensive industry. The residential sector accounts for roughly 23 percent of total consumption. Disaggregated information about residential demand is sparse, but Statistics Sweden assumes that a “typical” household with electric heating consume roughly 20000 kWh per year26 . As of 2013, about 40 percent of all households are on flat-rate contracts with yearly or longer contract durations, about 30 percent are 20 Contrary to the production and supplying of electricity, the distribution is a monopoly due to the natural monopoly characteristics of electricity transmission 21 Almost 75 percent of all electricity used in the Nordic countries are traded at Nord Pool spot market. 22 Shortages and blackouts are not a big problem in Sweden. Because of this, the peak capacity is also limited. 23 Svenska Kraftnät, a government owned entity (agency_ or company?), is responsible for managing peak capacity, and can procure up to 2000 MW in peak capacity (from electricity producers in Sweden). Further, the Swedish Parliament has decided that the capacity reserve will be gradually phased out by the 15 March 2020 and replaced by market-driven capacity and/or demand flexibility. 24 The generation from these technologies is quite small, and during the last ten years production has varied between 0.2 - 1.5 percent (300−2000 GWh) of total annual production. Figures from http://www.scb.se/Pages/TableAndChart____24270.aspx 25 http://www.svenskenergi.se/Elfakta/Elanvandning/ 26 However, this approximation is very rough. See Statistics Sweden.

8

on monthly-price contracts, and the remaining have a so-called default contract27 . The general trend is of households switching to monthly-price contracts, instead of fixed or default contracts . For the last seven years (2005-2012) the average flat-rate price was 50.4 SEK/kWh, and for monthly-price contracts 45.7 SEK/kWh. Sweden has set out on rather ambitious climite policy targets following the EU Climate and Energy Package (the 20-20-20 target), stipulating, among other things, a 20 percent increase in energy efficiency and 50 percent of energy consumption from renewable sources in 2020 (relative to 2009). For Sweden, the Swedish Energy Agency (Statens Energimyndighet) has been commissioned by the Swedish government to both detail what this target implies for Sweden (concerning the specific energy efficiency target over the stipulated period), and to present a plan of how to reach the target28 . It is argued that a Swedish implementation of real time pricing, and a more efficient energy market in general, will benefit (?) the work towards these targets (see for example the SEMI report).

3

Data and summary statistics

The data used in this paper originates from a metering project commisioned by The Swedish Energy Agency. The purpose of this project was to increase the quality of data on residential electricity consumption, and to assess the potiential for energy conservation and increasing energy efficiency. In total, 389 households, selected by Statistics Sweden, had metering equipment installed on all major appliances. In total, roughly 150 different appliances were metered, and each household had a maximum of 46 appliances metered at a time. In addition, each household had individual recorders for both outdoor and indoor temperature. Both metering and temperature data were recorded at 10 minute intervals. 200 of the homes metered were detached houses and the rest, flats. In this paper, we focus exclusively on detached houses. As detached houses have a substantially higher level of consumption compared to flats, they are also more interesting to study from a policy perspective. A majority of the households were located in Mälardalen region, with only 25 households each located in northern and southern Sweden. The project was carried out between 2005 to 2008 and each household was metered for between 15 and 365 days. The figures below illustrate time distribution of metered households. 27 If households do not opt in to either flat-rate or monthly contracts they are assigned a default contracts where prices (SEK/kWh) typically are fixed on an annual basis. 28 The EU 20-20-20 target is defined to have 20 percent of renewable energy, 20 percent more energy efficiency and a reduction in emissions of 20 percent, by 2020.

9

Figure 1: The distribution of number of households metered across months and years.

Figure 2: The cumulative distribution of household metering time. (Explain, in the text, how this figure is to be read).

Figure 1 illustrates when households were metered, with number of metered households for each month illustrated on the vertical axis As can be seen, roughly 10-20 households per month were metered during 2005 and 2006, and 20-30 households per month during 2007 and early 2008. Figure 2 illustrate how many months household were metered, with duration of metered period (in months) for each household on the horisontal axis and number of households on the vertical axis As can be seen, a majority of the households were metered for one or two months, and a few households metered for three to six or twelve to fiftheen months. In addition to the metering data, survey data was collected on household characteristics such as (monthly) income, number and age of inhabitants, living area size, main heating system, building year and year of refurbishment. For some appliances, information on brand and model is also available.

10

Mixed heating

Electric heating

All heating systems

p-value

39753.86 3.313 128.4 1969.241 23.291 72

40604.78 3.170 130.773 1971.109 27.101 129

40324.2 3.219 129.974 1970.527 25.73 201

.524 .422 .644 .660 .000

1039.866 518.3613 118.6588 81.0818 321.6891 144

1392.303 948.0272 132.1041 86.61495 225.4745 158

1224.253 743.1534 125.6931 83.97662 271.3517 302

.000 .000 .014 .341 .000

Household characteristics Income (SEK) Household size (persons) Living area size Building Year Number of appliances Observations Monthly consumption in kWh Total Heating (incl water heating) Kitchen Lighting Residual Observations

Table 1: Summary statistics (mean (s.e.)) for households, grouped by heating source. Note that “mixed” heating refers to households which used some form of electric heating (e.g. portable heaters) but wherein the major source of heating was not electric. Column “p-value” refers to the p-value on the (t-) test of equality of mean of relevant characteristic across households with mixed and electric heating. Table 1 provides summary statistics for household characteristics by heating source. Income is reported in intervals, but we use interval regression to transform it to a continous variable (see section 4.1). Note that the mean income is very high, compared to Swedish population average of roughly 28000 SEK. This is might be due to measurement error, as income is self-reported. Household size (number of persons) varies between one and six, with a mean of slightly above 3, and living area size in square meters has a mean value of 130. Building year vary substantially between 1750 and 2007. This should have substantial effect on for example heating consumption, where old houses are expected to be more “leaky” and hence require more heating load. Finally, between 10 and 46 appliances were metered for each household with a mean of 25. This should cover all main appliances. It is important to point out that the number of metering devices assigned to each household may be driven by other things than how many appliances the household possess. We find no differences in household characteristics between households with different heating systems, except for the slighly larger number of appliances owned by households which are electrically-heated. This fact motivates the relatively simple load curve estimation framework, with load curves for houses which are electrically heated only an intercept shift; see section 4.1 for details. The household characteristics do not also appear to vary depending on when or how long households were metered. The monthly average for an electricly heated household corresponds to an annual load of roughly 17000 kWh, sligthly less than the 20000 kWh assumed by Statistics Sweden29 . As can be seen, both total and heating load differ substantially between different heating systems. However, households are rather homoge29 See

section §2 for details on the Statistics Sweden reference household.

11

nous in terms of kitchen and lighting consumption, irrespective of heating system. Not illustrated in the table (but evident from Fig in the Appendix) is the fact that the minimum values for some end-uses are small; this is the result of averaging across higly variable monthly consumption i.e. during warm summer months, even households with electric heating only have very small heating consumption. This seasonal pattern is illustrated in figure 3 below, where we show average monthly consumption over the studied period (2005-2008).

Figure 3: Mean consumption in kWh by month. Note that this is the mean over all households and all heating systems. As can be seen in Figure 3, there is a distinct winter peak with more than the double monthly load compared to summer months. As winters in Sweden are both dark and cold, this is expected. Heating and lighting load should, in particular, be substantially larger during the winter.

4 4.1

Results Load curve estimation

For the estimation of load curves, we start by aggregating the load from individual appliances to enduses. These end-uses includes “kitchen” as the sum of all kitchen appliances (stove, oven, microwave etc), “lighting”, and “heating” as the sum of all heating appliances – for example air heating pumps, radiators and also water heating appliances. Appliances that do not fit in any of these three catogories are aggregated into a “residual” end-use (for example TV, computers, etc). The categories are selected based on their expected share of total load, where heating by far should be the largest load followed by kitchen and lighting. For the load curve estimation we consider a subsample of the full data set, consisting of all working days in February (of all years). The choice of February is motivated by the fact that this is usually the coldest a month, making it the most interesting month to study (since heating is the end-use that consumes most electricity, see table 1). We also compare the results from this month with consumption for working days in June, June being the warmest (non-vacation) month.

12

Following B&F, we also use a SUR framework for estimation of the end-use load curves with a total of 24 hourly equations for each end-use. However, unlike in the case of B&F, a majority of the appliances are metered in our sample and, as a result, there is no scope for combining metered and unmetered data. This considerably simplifies our estimation framework, which we present next30 . The actual equation estimated for hour t is (ommitting the t superscript for notational simplicity): Y i = b + δy + βXid,t + vi,y D X

where Y i =

(1)

Yi,d

d=1

is the mean (or possibly median) daily load for hour t and end-use k = 1, ..., K for D household i = 1, .., N , Xi,d,t denotes control variables which are allowed to vary over all dimensions of the data and δy denotes a year dummy accounting for year-specific effects (if any) on daily consumption (year indices are suppressed on X and Y , since they do not play any role in our analysis). Essentially, this formulation allows us to focus on variability across households, and allows an interpretation of equation (1) as modeling the consumption of household i on an average february day31 . Thus, the percentiles of (predicted or actual) consumption, based upon equation (1), refer to those of the relevant household on an average february day, an interpretation which allows infering household behaviour. The complete list of control variables are the following: hourly mean outside temperature, building year, living area size, heating system, income (in SEK), living area size (in m2), building year, number of inhabitants and retail prices. As heating is by far the largest load, outdoor temperature should have a significant effect on electricity consumption.32 A similiar argument can be made for the inclusion of building year and living area size, as old (and presumably more leaky) houses, and larger houses, consume more electricity for space heating. Of course, whether the household has only electric heating or mixed heating should also affect (heating) consumption. We also include number of inhabitants and income, as this likely is positively correlated with both the number of appliances possessed and energy efficiency of these. Finally, 30 It is worth pointing out that the reason the B&F framework is relatively involved is due to heteroscedasticity in their framework, induced by the presence of non-metered appliances/end-uses. This leads to their using a slightly more involved, non-standard approach (an iterative, two-stage approach), detailed in Bartels and Fiebig (2000) to deal with the (in their case) issue of additive heteroscedasticity. Since all end uses–and almost all appliances–are metered in our case, we can use the standard SUR framework with hours as equations (as B&F also point out, pp 54). 31 An alternative approach would be to use the data from every day in February, which lead to the following equation for the tth hour (with obvious notation): Yi,d,y = at + αi + δy + βXi,d,y,t + ti,d,y .

This approach, estimating N + 4 + K parameters (typically K is around 8) and N × 20 observations, provides additional benefits in terms of accounting for individual-level “unobserved heterogeneity” (as is typical with “fixed effects”) or any other cause leading to violation of the usual orthogonality conditions at the individual-equation level. On the other hand, in our load curves below, the interpretation in terms of households (e.g. “median household on a average day in february”) is more complicated, since there is variation across two dimensions, now; day and household. Nonetheless, we find that both formulations of the equation for the tth hour provide almost identical load curves, both in pattern and magnitudes (the load curves for the formulation in equation (??) are available upon request). 32 Note that as outdoor temperature varies across hours (i.e. equations), this give rise to efficiency gains from using the SUR framework, compared to estimating equations for each hour separately.

13

we include retail prices to control for price variation between households.33 We then estimate equation equation (1) for each hour using standard SUR, and use the median predicted value to produce end-use specific load curves.In addition, we also produce load curves for the 20th and 80th percentile consumption (i.e. percentiles of Ybi ) to illustrate household heterogeneity in terms of hourly load. We now turn to the results of this estimation. Below we illustrate daily load curves for working days in February. Corresponding load curves for working days in June are to be found in the appendix.

Figure 4: Total electricity consumption by hour for working days in February. The solid line illustrates the median , the dashed line the 80th percentile and the dotted line the 20th percentile of hourly load.

Figure 5: Total electricity consumption by hour for working days in February. Also illustrated are mean outdoor and indoor temperatures (dashdotted and dashed lines, respectively)–both metered for each household–and typical working hours (the two vertical dotted lines), assumed to be 8 AM-5PM.

Total load is displayed in figure 4, illustrating the median consumption together with the 80th and 20th percentile consumption. Note the large difference between median and 80th and 20th percentile. This is expected and is to a large extent driven by differences in heating load. We elaborate more on this below, when we discuss the load curve for heating. As might be expected, there are two distinct peaks. The first one occurs at roughly 6 am when the household wakes up. The second one occurs around 5 pm when the household get home from work. Thus, the shape is rather intuitive. Turning to figure 5, it is clear that two two peaks correspond roughly to working hours (typically 8 am – 5 pm in Sweden), illustrated by the two vertical lines. Note also that the total load curve appears (approximately) to be the mirrorimage of outdoor temperature. Since the indoor temperature remains flat during the day, this indicates that some part of total load (i.e. heating) is used to offset outdoor temperature as to keep indoor temperature constant. Further, the mean indoor temperature is suprisingly high, compared to the recommendations of minimum indoor temperature of 18 degrees celsius by The National Board of Health and Welfare (see 33 Unfortunately the metering data set does not include retail prices. Neither do we have information on contract type (monthly, annual, etc) for each household. However, average prices for different contract types are available from Statistics Sweden which are used as an approximation. We test for different contract types but finds no change in results, and therefore use average price for monthly contracts in our main results

14

http://www.socialstyrelsen.se/publikationer2013/2013-11-29). Further, not illustrated is sunrise and sunset at approximately 7:30 am and 4:30 pm in February, roughly coinciding with working hours34 .

Figure 6: Electricity consumption for space (and water ) heating by hour for working days in February. Other details similar to those in figure 4 and figure 5.

Figure 7: Electricity consumption for lighting by hour for working days in February. Other details similar to those in figure 4 and figure 5.

In figure 6 we illustrate the estimated load curve for heating. First of all, note the large differences between households, with the median heating load roughly three times the 20th percentile load. As noted above, this is mainly due to differences in heating systems. Households with mixed heating have the possibility to reduce electricity consumption substantially by substituting electric heating with, for example, a wood stove or district heating. However, even if the 20th percentile consumption is relatively small compared to median, it is nonetheless large when compared to for example electricity consumption from lighting or kitchen (roughly 0.5 kWh thoughout the day, comparable to peak lighting load for the 80th percentile household). Hence, the heating load curves not only illustrate household heterogeneity but also just how large space heating is (in terms of kWh) relative to other end-uses. As can be seen in the figure, there are distinct morning and evening peaks for the heating load curve (for all levels of consumption, although it is naturally less pronounced for the 20th percentile). Interestingly, although there is a decrease in heating load during mid-day, this does not lead to a corresponding decrease in indoor temperature. As residents usually are away from home during mid-day (and thereby should not be concerned with a specific indoor temperature) this might then suggest that there exist potential for energy conservation by reducing heating consumption even further, without reducing utility derived from it (i.e. without reducing indoor temperature during periods when householders are at home). 34 Specifically, these are sunrise and sunset for February 14 in Mälardalen. Since we are estimating load for the mean working day in february, we use this date as it is in the middle of the month. This implies that in early February, the hours of daylight ar fewer, and that they increase towards the end of the month. Further, we use the metrological definition of sunset and sunrise; when half of the sun is visible above the horizon. It is important to point out that hours of daylight differs substantially between regions in Sweden.

15

The load curve for lighting is displayed in figure 7. Here households appear to be rather homogenous in their consumption, with both the 80th and 20th percentile close to the median. There is a smaller peak during morning, and a larger one between 4 pm and 10 pm, and these peaks corresponds to sunrise and sunset (and beyond). Furthermore, the fact that more lighting is used during evening is because there are more activities at home during this time (compared to the morning). Also worth noting is that the load is close to zero during night time, implying only limited stand-by usage. Compared to the heating load curve, we note that the lightning load is relatively small, in absolute terms.

Figure 8: Electricity consumption for kitchen by hours for working days in February. Other details similar to those in figure 4 and figure 5.

Figure 9: Electricity consumption for residual load by hours for working days in February. Other details similar to those in figure 4 and figure 5.

The load curve for kitchen is displayed in figure 8. Again, households seems to be rather homogenous in terms of kitchen electricity use. At first sight, it might seem strange that there is no morning peak. However, Swedish breakfast is by tradition cold, and even if some electricity is used for making coffee etc this load is only marginal (compared to more cooking-intense dinner). There is a distinct dinner peak between 4 pm and 8 pm, as can be expected. Clearly, some people end their working day early since the dinner peak starts at roughly 4 pm. This could be explained by some people being retired or working part time. The magnitude of kitchen load is comparable to that of lighting, and hence small in comparison with heating. Finally, we illustrate the load curve for the residual electricity consumption in figure 9. Since this category contains many different end-uses such as TV, computers, etc, it is difficult to interpret the shape. However, we can note that also this curve has a morning and afternoon peak occuring roughly at the same time as the other end-uses.

4.2

Cost savings from load shifting

Next, we turn to the cost for a utility of servicing each end-use and how these cost would change if the average household shifted the load of some end-use . First, consider the average Nord pool spot price for a

16

working day in February. Note that this spot price is for all demand and not just residential. In that sense, the price curve not only show the scope for price-induced load shifts but also gives a picture of system peak (both residential, service and industrial sectors demand).

Figure 10: Mean (solid line), min (dotted line) and max (dashed line) hourly Nord pool spot prices during February (2006-2008). The mean hourly spot price in shown figure 10 (together with hourly minimum and maximum prices). As can be seen in figure, there are two price peaks; one at roughly 9 am and then at around 17 pm. As can be seen, this peaks are naturally more pronounced for the maximum price. Comparing the spot price to the estimated load curves above, it should be obvious that these two price peaks coincide with the the household peak. Further, when residential demand is low during the early morning and late night the price is also rather low. This consumption pattern has two important implications: first, it is in line with the hypothesized behavior of “excessive” consumption during periods of high price and second, that there are potential cost savings from shifting consumption to off-peak hours. However, as the variation in the average spot price is rather small, the potential cost savings are a priori expected to be limited, on average. To estimate to cost of servicing each end-use we match the estimated load curve with corresponding spot prices, using the average spot prices for working days in february during 2006-2008. We denote this price by p¯t and the estimated predicted hourly load for end-use k as yet,k . To obtain the cost Ckt of servicing end-use k in hourt, we simply multiply the predicted hourly load with the mean spot price for that hour. Note that this cost should be interpreted as the utilities cost for servicing that particular load but could also be interpreted as the equivalent cost for the household on real time pricing, since what the consumer pays is the cost (paid by utility) plus taxes, mark-up and some fixed costs for transmission, among other things35 :

Ckt = yet,k p¯t 35 Many

parts of these fees are quantity-independent, justifying an interpretation as “fixed costs”.

17

The daily cost of servicing end-use k is then the sum of the 24 hourly costs; Ckd =

24 X

yet,k p¯t

t=1

End use

Daily cost in SEK

Heating Kitchen Lighting Total

12.603 1.247 1.225 18.800

Table 2: Daily cost (in SEK) of servicing different end uses for the median consumption level (load). table 2 illustrates the average cost of servicing each end-use for an average household during an average february working day, where this cost is assumed to be the average spot price multiplied with the average end-use load. As can be seen from table 2, heating is by far the most expensive end-use, being the largest load. For the other end-uses, even if timing of their demand (being part of peak demand) coincides with high spot prices, their cost is still small relative to the cost of heating. Load shifting Hours ahead 1h 2h 3h 4h 5h 6h 7h

Cost decrease as % of daily total cost Median household 0.003 0.313 0.77 1.22 1.58 1.92 2.15

80th percentile household -0.15 0.148 0.768 1.35 1.82 2.23 2.44

20th percentile household 0.42 1.29 2.29 3.18 3.96 4.54 4.80

Table 3: Cost savings from shifting load. We test how these costs changes when we shift load from a given end-use from expensive hours to cheap hours. We can either interpret them as the cost change for a utility, or we can think of it as some fraction of the cost saving benefiting the consumer. The conceptually most simple way of shifting the load curve is to move the whole curve a few hours ahead in time36 . This implies that evening hours now show up as morning hours (i.e. some kind of rightshift). We shift the total load curve in this way for one to seven hours ahead and calculate the cost change in percentage and SEK. When load is shifted seven hours ahead, the demand peaks occur at the lowest hourly prices. Note that by doing this we are keeping the total daily load constant, and are only re-allocating the consumption over hours. We do this experiment for the mean household, but also test how the cost changes for households at the 80th and 20th percentile of consumption. 36 Note that we only do this experiment for within-day load shifting, as compared to across-day shifting. For all of these end-uses except for laundry the substitutability for shifting load across days (for example from a working day to the week end) is limited. For example, it would not make any sense to shift space heating or lighting from one day to another, since that would imply one very cold and dark, albeit cheap, day.

18

As can be seen in table 3, the cost decreases are overall suprisingly small, and this holds for all three type of households. Even if we shift the whole load curve seven hours ahead, the cost only decrease with 2.15 percent for the median household. Further, if we shift the load additional hours ahead the cost actually increase. It is also important to bear in mind that these are the cost changes for an average february working day (using the average february prices). Hence, for some days the cost will probably decrease more, and for some days the cost may even increase. Nonetheless, the monetary incentives for households and/or utilities are quite small, and it seems unlikely that the household would alter consumption behavior significantly for such small gains. The cost savings illustrated in table 3 are a likely best cast scenario, for several reasons. First, the load shifting pattern illustrated above is not very feasable. Rather, we would like to stress that such shifting is highly unlike since it requires the household to completely change their habits. It seems reasonable to belive that such a change in habits would cause significant disutility for the household, at least in the short run. Further, when limited substitutability of load is combined with the rather small monetary incentives for such shifts, the incentives for individual households to tolerate such load shifting are very small37 . Finally, in this experiment we treat the spot price as exogenous. This assumption is reasonable when few households are on real time pricing schemes. However, if larger number of households switch to real time pricing, one would anticipate that the price variation, and thereby potential cost savings, to decrease.

4.3

Price elasticity estimation

To be written

5

Conclusions

To be written

References Allcott, H. (2011). Rethinking real-time electricity pricing. Resource and Energy Economics, 33(4):820–842. Andersson, B. (1997). Essays on the swedish electricity market. Bartels, R. and Fiebig, D. G. (2000). Residential end-use electricity demand: results from a designed experiment. The Energy Journal, (2):51–81. Bartels, R. and Fiebig, G. (1990). Integrating direct metering and conditional demand analysis for estimating end-use loads. The Energy Journal, 11(4):79–98. 37 On the other hand, Allcott (Allcott (2011)) finds – as already alluded to– that households goes to great lengths in responding to price increases by reducing demand. However, what we argue as being unreasonable is load shifting, not energy conservation.

19

Borenstein, S. (2005). The long-run efficiency of real-time electricity pricing. Borenstein, S. and Holland, S. P. (2003). On the efficiency of competitive electricity markets with timeinvariant retail prices. Technical report, National Bureau of Economic Research. Brännlund, R., Ghalwash, T., and Nordström, J. (2007). Increased energy efficiency and the rebound effect: effects on consumption and emissions. Energy economics, 29(1):1–17. Damsgaard, N. (2003). Residential electricity demand: Effects of behavior, attitudes and interest. Fiebig, D. G., Bartels, R., and Aigner, D. J. (1991). A random coefficient approach to the estimation of residential end-use load profiles. Journal of Econometrics, 50(3):297–327. Holland, S. P. and Mansur, E. T. (2006). The short-run effects of time-varying prices in competitive electricity markets. ENERGY JOURNAL-CAMBRIDGE MA THEN CLEVELAND OH-, 27(4):127. Hsiao, C., Mountain, D. C., and Illman, K. H. (1995). A bayesian integration of end-use metering and conditional-demand analysis. Journal of Business & Economic Statistics, 13(3):315–326. Kavousian, A., Rajagopal, R., and Fischer, M. (2013). Determinants of residential electricity consumption: Using smart meter data to examine the effect of climate, building characteristics, appliance stock, and occupants’ behavior. Energy. Krishnamurthy, C. K. B. and Kriström, B. (2013). Energy demand and income elasticity: a cross country analysis. Available at SSRN. Larsen, B. M. and Nesbakken, R. (2004). Household electricity end-use consumption: results from econometric and engineering models. Energy Economics, 26(2):179–200. Lundgren, J., Granström, S., and Pålsson, M. (2010). Ökat inflytande för kunderna på elmarknaden. timmätning för elkunder med abonnemang om högst 63 ampere. Swedish Energy Markets Inspectorate. Parti, M. and Parti, C. (1980). The total and appliance-specific conditional demand for electricity in the household sector. The Bell Journal of Economics, pages 309–321. Savolainen, M. and Svento, R. (2012). Real-time pricing in the nordic power markets. Energy economics, 34(4):1131–1142. SEMI (2013). Uppföljning av timmÀtningsreformen. Swedish Energy Markets Inspectorate.

20