Online Annexes: Chapter 1

International Monetary Fund. Research Dept.
Published Date:
October 2018
Show Summary Details

World Economic and Financial Surveys

World Economic Outlook

Online Annexes: Chapter 1

OCT 18

Annex 1.SF.1. Commodity Special Feature

Global Energy Demand: Additional Facts

Over the past decades, the increase in demand for electricity has led to higher usage of primary energy sources by the power sector. Electricity, in fact, is not a primary energy supply, given that it is generated from fossil fuels, mainly from coal and natural gas, nuclear, hydro, solar photovoltaic, wind, geothermal, biomass and waste, and a few other sources. Coal and natural gas contributed the most to satisfy the significant increase of electricity consumption from 1971 to 2015 (Annex Table 1.SF.1.1).

Contrary to all other primary energy sources, oil is mostly used in the transport sector and not in the power generation. The share of oil-fueled power plants has, in fact, declined dramatically over the past decades (Annex Table 1.SF.1.2).

Annex Table 1.SF.1.1.Contribution to Bectrification, 1971–2015(Percent, contribution to growth)
Other renewables and other fuels27
Total electricity growth (1971-2015)296
Source: International Energy Agency, World Energy Balances; and IMF staff calculations.Note: Individual contributions to electrification sum up to total electricity growth (1971-2015).
Annex Table 1.SF.1.2.World Energy Usage, 2015 and 1971(Millions tonnes of oil equivalent)
Year = 2015 Unit: MIOECoalOilGasNuclearRenewables and OtherTotalPercent of Total
Power Generation2,6803071,2326717545,64341
Residential and Commercial11029660108061,81313
Non-Energy Use616151600-8366
Total Primary Energy Demand3,8364,3342,9446711,85813,643100
Year = 1971 Unit: MtoeCoalOilGasNuclearRenewables and OtherTotalPercent of Total
Power Generation739302217291381,42526
Residential and Commercial20934120404631,21722
Non-Energy Use621614002364
Total Primary Energy Demand1,4372,438893297255,523100
Source: International Energy Agency, World Energy Balances; and IMF staff calculations.Note: Because of statistical discrepancies, individual data in each column do not sum exactly to total primary energy demand. Lighter color represents smaller share in total primary energy demand.

Energy and Income: The Econometric Specification

To test for the presence of an S-shaped relationship between energy demand and per capita income, the Special Feature uses an unbalanced panel of 136 countries, controlling for the size of the country (that is, population and land area) and fossil fuel abundance. Time fixed effects are used to capture worldwide gains in energy efficiency and fluctuations in global economic activity and energy prices. The sample is annual and spans 1971–2015, covering two major energy price cycles. Specifically, the exercise estimates the following specification relating (log) total energy demand E to (log) population, pop; a third-order polynomial in (log) income per capita, gdp; and a vector of control variables, X:3

in which λt are year fixed effects, while Xit includes a time-varying energy-export and coal producer dummy, distance from the equator, and the log of land area; the indices i and t refer to countries and years, respectively. An oil exporter is defined as having oil production exceeding consumption. A similar definition is used for natural gas and coal exporters. A coal producer is defined as having production able to satisfy between 60 percent and 100 percent of the country’s coal consumption. Distance from equator is the absolute value of latitude.

This empirical specification generalizes earlier contributions to the literature on energy demand, including Medlock and Soligo (2001) and van Benthem (2015), and follows the seminal work of Grossman and Krueger (1995), who study turning points in the relationship between income and environmental quality by introducing a cubic per capita income term.4 Similarly, this specification potentially allows for an S-shaped relationship between income and energy demand (β4 < 0). The special case of β3 = β4 = 0 and β1 = β2 corresponds to the case in which energy demand increases proportionally to GDP.

Results for the baseline specification, column (5), and robustness checks are reported in Annex Table 1.SF.1.3. Not surprisingly, the analysis finds that energy demand moves in lockstep with population. Point estimates suggest that having a sizable land area coupled with being a coal exporter (producer) increases energy demand by about 45 (33) percent. Given that coal is expensive to transport, coal prices tend to be relatively low in coal producer countries, thus attracting energy-intensive industries, while a large land area may require more travelling and may proxy for hydropower potential. Both factors increase energy demand given population size and other country characteristics.

Annex Table 1.SF.1.3.Total Demand Determinants by Different Specifications
GDP per Capita0.783***-1.048**-8.415***-7.103*-8.676**-5.068*-6.889***
(GDP per Capita)20.104***0.956***0.843*1.044**0.639*0.865***
(GDP per Capita)3-0.0324***-0.0293-0.0378**-0.0231-0.0330***
Oil Exporter-0.01730.00523
Gas Exporter0.0483-0.0478
Coal Exporter0.378**0.315**
Coal Producer0.251*0.132
Number of Countries1361361361368887136
Graduation from Biomass770520562321464
Static Saturation Port454,186401,087179,389323,51682,921
Dynamic Saturation Port (1% eff. gain)154,556127,28663,59074,05017,831
Dynamic Saturation Point (spec. eff. gain)173,33333,57638,41041,29825,281
Inflection Point18,69814,44710,03910,1846,204
Max Elasticity0.98920.97230.94160.82800.6660
Average Elasticity0.98450.97210.92330.81770.5888
Sources: International Energy Agency; World Bank World Development Indicators; and IMF staff calculations.Note: Energy exporters and producers are derived from the International Energy Agency. Latitude is the absolute value of latitude in degrees for national capitals. Robust standard errors clustered at the country level in parentheses. Variables are in logarithmic scale. Column (7) uses fixed effects regression. Average elasticity is calculated at 15,000 2011 international US$. “eff. gain” is efficiency gain. “spec. eff. gain” is specific efficiency gain, and is calculated using each specification’s average growth of time dummies. OLS = ordinary least squares. WLS = weighted least squares.* p<0.05, ** p<0.01, *** p<0.001

Results are qualitatively robust to different specifications, including “distance from the equator” (column (6)) and “temperature” (not shown), which serve as proxies for intrinsic demand for cooling and heating energy services, reducing the significance of income. This is not surprising, however, as Hall and Jones (1999) and other scholars have used distance from the equator to instrument output per worker. Similarly, under country fixed effects, the estimate of the cubic terms is less significant when using OLS (not shown).5 Pooling the sample, instead, allows for a better exploitation of income differences across countries to infer the form of the energy-income S-shaped relation. Finally, weighting countries by energy demand, reinforces the S-shaped relationship, reducing the saturation point, even with country fixed effects (specification 7). Further, a necessary and sufficient condition for having an S-shaped relationship is that β1<0, β2>0, and β3<0. The estimated coefficients meet this condition, which is statistically significant and robust across most specifications.

The average energy efficiency gain over the sample is calculated by dividing the time dummy in 2015 (that is, the last data point) by 45 (the number of years in the sample).6 Energy efficiency gains vary somewhat across specifications, with an average of about 1 percent a year.7 In Annex Table 1.SF.3, the baseline dynamic saturation point, thus, is calculated assuming a 1 percent annual efficiency gain. Alternatively, the dynamic saturation point is calculated assuming the average efficiency gain calculated for each specification.8 As even small differences in the assumed efficiency gains compound over time, the alternative calculation can lead to substantial differences in the estimated saturation point.

The Primary Energy Mix

To examine the primary energy mix, each primary energy share is regressed on a quadratic function of per capita income, as shown in Annex Table 1.SF.1.4. The same dataset of the energy demand analysis is used. The shares are calculated dividing the primary energy source to total energy by country over time. To control for the local relative abundance of a specific resource, either dummies or country fixed effects are used. In the energy source panel regression, country fixed effects are necessary to avoid omitted variable bias (for example, Malta uses no natural gas, given that no gas pipeline reaches the island). It is worth noting how having a relative abundance of a fossil fuel (that is, oil, natural gas, and coal) increases significantly its consumption share (that is, the exporter and coal producer dummies are significant), mostly at the expense of the other (relatively less abundant) fossil fuels. 9

Annex Table 1.SF.1.4.Primary Energy and Electricity Share Determinants
Share of Fossil FuelShare of OilShare of Natural GasShare of CoalShare of RenewableShare of Electricty
GDP per Capita0.594***0.798***0.460***0.466***-0.1240.1650.258***0.158-0.395***-0.729***-0.0567***-0.0468
(GDP per Capita)2-0.0327***-0.0395***-0.0290***-0.0237**0.00927-0.00792-0.0129**-0.007590.0195***0.0343***0.00565***0.00387
Oil Exporter0.0460.0747*0.0681*-0.0901**-0.0268-0.02
Gas Exporter0.00155-0.0646**0.113**-0.04060.0279-0.00231
Coal Exporter-0.00387-0.135***-0.0627*0.186***0.0181-0.00854
Coal Producer0.0359-0.0704*-0.00920.103***-0.038-0.00468
Number of Countries136881368813688136881368813688
Share Peak8,93024,3102,75518,455NANA22,377NANANANANA
Sources: International Energy Agency; Wold Bank World Development Indicators; and IMF staff calculations.Note: Fixed effects regression. Robust standard errors clustered at the country level in parentheses. Share is in percentage. GDP per capita is in logarithmic scale. NA = not applicable.* p<0.05, ** p<0.01, *** p<0.001

The Demand for and Supply of Renewable Energy: Additional Information

As discussed in Box 1.SF.1, renewable energy has grown rapidly in recent years. Annex Figure 1.SF.1.1 visualizes renewable energy growth between 1980–2015 by region using five-year moving averages.

To explain the demand and outlook for renewable energy, Box 1.SF.1 proposes a simpler version of the main specification used in the Special Feature: 17

in which REit measures the renewable energy capacity for electricity generation; gdpit and popit are log income and log population, respectively; Xi is a set of controls including distance from the equator and the logarithm of land area; and yeart is a trend term capturing global trends in technology. The renewable energy capacity for electricity generation (in gigawatts [GW]) includes hydropower, solar, wind, and geothermal (except biofuels) covering 177 countries over 1990–2015 (using data from the US Energy Information Administration).

Annex Figure 1.SF.1.1.Renewable Total Primary Energy Supply Growth, by Select Regions

(percent, five-year year moving average)

Source: International Energy Agency; and IMF staff calculations

Note: Renewable total primary energy supply includes hydropower, geothermal, and solar, wind, and other. FSU-15 is former Soviet Union, consisting of 15 countries. Global average includes FSU-15, Latin America and the Caribbean, Middle East and North Africa, and South Asia.

Given that the focus in Box 1.SF.1 is on forecasting the demand for renewable energy, the above equation is estimated in levels using a weighted nonlinear least squares estimator, which is identical to Poisson pseudo-maximum likelihood (PPML) as in Silva and Tenreyro (2006). An estimation of the log-log form of equation (1) using weighted ordinary least squares (with REit as weight) gives very similar out-of-sample predictions, but the within-sample predictions of weighted nonlinear least squares appear superior.18

For the out-of-sample prediction presented in Box 1.SF.1, OECD (2014) long-term forecasts for GDP per capita and population size are used. The Organisation for Economic Co-operation and Development (OECD) publishes GDP and population forecasts for 45 major economies, including all OECD and Group of Twenty countries. For the remaining group of countries, the average sample growth rates are used. Given that most renewable energy capacity is installed in the 45 countries covered by the OECD, the forecasts are not sensitive to these assumptions.

To calculate how much energy the increase in renewable energy capacity can deliver to the electricity grid under the conservative and baseline scenarios shown in Figure 1.SF.1.1 of Box 1.SF.1, it is assumed that renewable energy capacity has a utilization rate of 35 percent, on average. Furthermore, to calculate how much fossil-fuel-based electricity generation can potentially be replaced by this new renewable energy capacity, total demand for fossil fuels by the electricity sector is calculated by summing over the categories coal, oil and gas in the ‘Power Generation” row of Annex Table 1.SF.1.2 for the year 2015. To account for the loss of energy that occurs during conversion from chemical energy to electricity in thermal power plants, it is assumed that the conversion efficiency of primary fossil fuels is 35 percent, on average.

Energy demand (in million tons of oil equivalent) is the sum of electricity and primary energy supply (that is, coal, oil, natural gas, hydropower, nuclear energy, and renewables). Energy data are from the International Energy Agency; data on population, GDP per capita (in 2011 US dollars), and country area size (in square kilometers) come from the World Bank’s World Development Indicators. Latitude is from the GeoDist database by Centre d’Etudes Prospectives et d’Informations Internationales.

Energy is consumed to produce energy services. These are defined as “those functions performed using energy which are means to obtain or facilitate desired end services or states” (Fell 2017). Familiar examples are heating, cooling, lighting, refrigeration, and transport. The non-energy demand for fossil fuels stems mainly from the petrochemical industry, representing about 6 percent of total primary energy demand.

Results for the specifications with country fixed effects estimated with ordinary least squares are available upon request.

The time dummy in the first year of the sample, 1971, is normalized to 0.

As for energy efficiency, there is long-standing debate in resource economics on whether energy efficiency implies a reduction in energy consumption. Jevons (1865) asserts: “[I]t is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to a diminished consumption. The very contrary is the truth […] Every improvement of the engine when effected will only accelerate anew the consumption of coal.” Most recent studies, however, do not find the Jevons’ effect—instead, energy consumption declines, all else equal, as energy efficiency improves (Greening, Greene, and Difiglio 2000; Sorrell 2009). By example, the United States “graduated” from biomass in the last quarter of the 19th century when wood was replaced by coal as the primary resource (EIA 2018).

This is labelled “spec. eff. gain” in Annex Table 1.SF.1.3.

There is a positive correlation between the natural gas and oil exporter dummies, given that about half of natural gas exporters are also oil exporters. This relation is much weaker, instead, for coal.

The shorter time span used here makes it difficult to fit an S-curved relationship between renewable energy capacity and income, necessitating a simpler specification.

For the several advantages of PPML relative to ordinary least squares for estimating nonlinear relationships, see Silva and Tenreyro (2006).

    Other Resources Citing This Publication