Overview
Energy demand elasticities are understandably popular. They are a convenient top down methodology to summarize past economic behaviour. Income elasticities have the potential to help us forecast increasing energy use as consumers get richer. Own price elasticities can help us anticipate the energy consumption response if energy or carbon is taxed. Or they can indicate price response to energy disruptions. Cross price elasticities show how fuels may be phased in or out as their costs of production change. Elasticities relating to structural and demographic variables indicate how energy needs change as countries industrialize or de-industrialize, citizens migrate to the city, or populations get older. Weather elasticities can indicate consumption response across seasons or climate variation. Such top down historical behavioural summaries can also be used as a cross check on more detailed technical bottom up models.
As a result of their usefulness, hundreds of studies have focused on demand elasticity estimates. However, for these past studies to be useful for forecasting and policy analysis, energy elasticities must follow predictable, discoverable patterns. Before moving forward with yet more study, now is a good time to look back at this large base of work and see what we have learned from it and develop an agenda for ongoing work. This presentation is a progress report on a large energy demand survey that seeks to do just that.
Methods
My data set to date is a collection of around 18,000 econometric equations in about 1,000 studies. I include any available study with demand estimates for energy products for any aggregate, for any sector, for any region or sub-region in the world. I include any demand elasticity reported: income, price, cross price, temperature, etc. My intent is to survey this heterogeneous data for the lessons it might teach us, make it public as the studies get processed and analyzed, and to develop and maintain an ongoing searchable database of studies. Both good and bad studies have been included to indicate what to emulate and what to avoid.
The database (Dahl Energy Demand Database (DEDD) is quite heterogeneous. Studies vary by product and sector is studied as well as when, where and how they are studied. I am interested in whether these varying factors influence energy demand elasticities - particularly price and income in discernible ways. My first and ongoing task has been to summarize and categorize this heterogeneity which I include in the longer paper and my presentation. Less than 2% of the studies were published before 1970 with about 40% published in the 1980s. Very few studies yet include data since 2000 in their samples but I am expecting that the recent price volatility along with climate legislation will spawn even more studies that in the 1970s and 1980s.
DEDD includes estimates for more than 100 countries, but estimates are dominated by the U.S. at almost 40 %, as these are most accessible to me, trailed by Canada and Australia. The estimates include total demands for energy, electricity, natural gas, coal, fossil fuels, bio-fuels, oil products, as well as total demands for separate oil products: gasoline, diesel fuel, fuel oil, distillate/gasoil , heavy fuel oil, jet fuel, kerosene, and liquid petroleum gases including propane. These demands may be disaggregated by sector including residential, commercial, electricity generation, individual industry, highway use, and highway transport. Electricity has been the most studied product followed by oil and total energy. Diesel and biomass are the least studied.
Time series is the most frequently used data type with a bit over half of the estimates, strict cross sectional data is under 10%. Cross section time series data are used for well over a third of the estimates. The sector the most demand equations estimated is the industrial with about 9% of estimates on aggregate industrial energy demand and about another quarter on demand for individual or subsets of industries. Another 23% are estimates for the residential sector. The transportation sector comes next with 18% of the estimates. These include estimates for gasoline, diesel, and jet fuel as well as aggregates of transportation fuels. Electricity generation and the comercial sector, which often include the government sector, have a few percentage points each.
Different types of models are used. The most popular has been the flexible functional form with more than 1/4 of the estimates. These are most often applied to industrial or electricity generation sectors. The simple static models are the second most popular. They comprise about 1/4 of the estimates. About 1/4 of these static models contain some sort of capital stock variable. Studies that include a lagged endogenous variable, other lagged variable or both comprise somewhat over 1/4 of the estimates. Expenditure system models are used for less than 5% of the estimates. Error correction models although still less than 5% of the estimates are becoming increasingly popular. It will be interesting to watch how these newer time series techniques will advance our understanding of the drivers of energy consumption.
Once the studies are summarized and categorized, increasingly the database is being analyzed looking for evidence supporting hypotheses or searching for patterns that might suggest testable hypotheses. For example, one could look for evidence that poor countries have higher income elasticities than rich countries, cross sectional data yields higher price elasticities than time series, estimates on monthly data yield lower long run elasticiticies than estimates on annual data. Although all available elasticities are being collected, my focus to date has been on own price and income elasticities with some attention given to cross price elasticities. Histograms and summary statistics including means, medians and percentiles are being developed by category starting with more aggregate data and moving increasingly to more disaggregate data. I do not eliminate any studies but keep all the extreme values and look for patterns as to when or why they occur. Later analysis will limit the estimates to preferred specifications and where sufficient data exist, testing will be performed to find if suspected patterns have statistical support. Descriptive and anecdotal evidence will maintained on less studied variables or categories. The results of the tests on this large heterogenous data set will be used to develop a research agenda and hypotheses to be tested on more heterogenous data and studies consistently designed to investigate the topic.
Results
Results of the study will indicate any summary estimates or ranges of income, price and other elasticities that might be discernible from the database, suggested patterns about how they may change across categories, a discussion of what the studies suggest about energy demand, and a research agenda of questions that have important implications for energy producers, consumers, or policy makers.
Conclusions
As I have found in earlier survey work, at first blush, one is struck by the amount of heterogeneity in the elasticities. However, with more scrutiny of summary statistics patterns start to appear. For example, estimates on monthly and quarterly data do seem to find smaller long run income and price elasticities than those on annual data as are estimates on household data. Aggregate income elasticities seem to fall as countries get richer, but the patterns diverge across sectors. Commercial energy GDP elasticities increase as countries get richer and the commercial sector grows, industrial income elasticities are low for very poor countries, increase as the countries take off and are building up infrastructure, and then fall back again as country structures move towards the commercial sector. Transportation fuel demand tends to grow at a more predictable rate than either gasoline or diesel fuel especially where strong policies have aimed at the fuel mix. Most of the very extreme elasticity outliers are on models estimated with a lagged endogenous model. The conference paper and presentation will include these and other results gleaned from the analysis to date.