Demand for Electric Power in Norway : Estimating price and substitution elasticities

The main goal of this master thesis is to estimate how the prices of electricity and heating oil affect the aggregate demand for electric power in Norway. The sample period is 2000-2010. I find that aggregate demand is responding to prices. But the effect is limited. The thesis finds that the price elasticity during the summer becomes stronger and more significant if one control for a structural break in late 2004. This indicates that the mandatory setup of automatic hourly consumption reporting systems have influenced electricity demand. The results also improves in quality if one correct for the months of the greatest economic turmoil during the financial crisis of 2008-09. Compared to existing studies the thesis makes use of more recent data and it takes longer time periods into account. I estimate my models over a much longer sample period than have been usual. My models also control for more explanatory variables than the existing studies (I am aware of). For example, earlier studies do not bring the price of heating oil into their models. As heating oil is a substitute to electrical heating, these studies may suffer from omitted variables bias. By controlling for more variables, my models may be more reliable in describing electric power demand. The longer sample period also makes it possible to extract longer term effects and to model market dynamics in greater detail. Information about consumer behavior in the period may be useful for future investment decisions in infrastructure and production capacity. Accurate information about price and substitution elasticities may also be of interest for improving climate policies and tax regimes. In the next two sections I present my methods and the main results of the thesis. Methods I estimate a demand equation for electricity combining instrumental variable regression methods and autoregressive distributed lag models. To identify strong instruments, important explanatory variables and relevant market dynamics, I use the automatic model selection software implemented in OxMetrics 12. The software is further used for identifying large outliers, split sample analysis, forecast tests and in estimating the steady state solution to the model. The dataset and code is available for other researchers upon request. A central decision to make in a study of the Norwegian market for electricity is the choice and measurement of an electricity price variable. In the deregulated market after 1991, producers and consumers were free to establish bilateral contracts and several contract types (and prices) therefore exist. The price can also vary between different regions within the country. Some consumers therefore follow long term fixed price contracts while others buy in the Nord Pool spot market. Standard variable price contracts are also common. In these contracts, the supplier must notify price hikes two weeks in advance. I argue that the standard variable price contract is the best contract for my purposes. The main arguments are that: i) the prices in the different contracts tend to follow each other. ii) the price is the same across all regions if one use this variable. To identify valid instruments for the price of electricity, broad and accurate information about the market is necessary. A section describing key statistics of the Nord Pool area is therefore an important part of the thesis. I use three different variables as instruments: Inflows, reservoir contents and the price of coal. Inflows to hydropower reservoirs are probably not affecting electricity demand directly. Reservoir content levels will affect the present value of the water reservoirs and thus the behavior of hydropower producers. Furthermore, coal is barely used for other purposes than electricity production. In the thesis I argue that different transformed versions of these variables only affects the supply side of the market, and that they therefore could be used as instruments for the price of electricity. The instruments are thus useful for the identification of supply and demand side effects. Accurate information about the market is also important for categorizing exogenous explanatory variables correctly. In Norway, electricity is much used for heating during the long and cold Norwegian winter. The winter season is also dark and electricity is therefore also used for lighting. This supports the inclusion of a variable related to temperature, as well as seasonal controls that capture the changing lighting conditions. As electricity is mostly used for technical equipment during the summer season, one may also expect that the price elasticity is different during the summer. I therefore include a term to adjust for this potential seasonal effect. During national holidays such as Christmas and Easter, manufacturing is reduced. The model should therefore include variables capturing this. Furthermore, the income and the economic activity level also affect the demand for electricity. The gross domestic product of Norway (excl. offshore activity) is therefore included in the models. Another potentially crucial factor for understanding the electricity market is how often consumption is reported. Prior to 2005, only units consuming more than 400,000 kWh of electricity annually were required to report their electricity consumption at hourly intervals. Systems consuming less generally reported their consumption monthly and their bill was determined according to a typical consumption pattern. Hence, only very large consumers had an incentive to adjust demand according to short time price fluctuations. In 2005, the requirement was made mandatory to systems consuming 100,000 kWh or more annually as well. Increased awareness about the possibility to taking advantage of short term price fluctuations may have caused a structural change in the market after this period. In about five years, the system is planned to be required on practically all systems. How the demand side has responded to the 100,000 kWh changes of 2005 may therefore give us useful information about how hourly reporting in all units will affect the market. Future analysis could adjust for regional price differences in electricity prices and grid rents. One could also take the distribution of contract types into account, i.e. how many percent of the consumers that were on spot price contracts. Furthermore, one could include more explanatory variables. The price of carbon dioxide emission and natural gas are obvious candidates. But wood related heating products could also be used. Carbon dioxide emission allowances and natural gas prices data are available. But these variables are difficult to model. Good data on the price of wood related products are hard to find. Micro level data on consumption could also yield alternative to the approach I am using. Main results My final model estimates the long-run price elasticity at -0.1235 in the winter season. In the summer season the price elasticity is estimated at -0.0173. The seasonal difference is likely due to greater substitution availability during the winter. The final model further estimates the substitution elasticity with regards to heating oil at 0.0486. A one percent increase in the price of heating oil thus increase the demand for electricity by approximately 0.05 percent. As explained in chapter 2, the electricity price amounts to roughly one third of the total electricity cost. One could therefore approximate the total electricity price effect by multiplying the estimated price elasticities by a factor of three. Johnsen (2001) estimates the price elasticity to be between -0.05 and -0.35. He finds the price elasticity to be the highest when price levels are high. He thus also finds the greatest price elasticities during the winter season. The fact that he finds the price elasticities to be greater than I do may have several reasons. I include several variables that he does not control for. Furthermore, he used data from the mid-1990s, and the market may have structurally changed since then. The sample length of my study is much longer than the sample period he used. The estimated elasticities in my study are greater than those found by Bye and Hansen (2008). They estimate the long-run direct spot price elasticity to be -0.02 in the winter and generally zero (inelastic) during the summer. They look at a shorter period of time than I do, and my results may therefore capture more long term effects more accurately (due to sample size). They are furthermore analyzing the spot price market, which has greater price volatility. As the actual price paid is generally based on a monthly average price, these fluctuations are not generally of practical importance for small or medium sized consumers. My analysis does indeed find that something happened to the market in late 2004. The substitution effect became stronger and more significant. The summer price elasticity also increased. The introduction of mandatory hourly report systems is the likely reason behind this structural change. Adjusting for this as well the financial crisis, the parameters remain stable in sub-sample tests. This suggests that the model also can be used for forecasting purposes. The conclusion is thus that the demand side of the Norwegian electricity market responds to price movements. But the response is limited. This suggests that a tax on electricity is probably quite efficient, because consumers will not substitute consumption away from electricity to a great extent. Furthermore, the results indicate how challenging it is to reduce electricity demand significantly for, i.e., eco-political causes