The first one showed the strong correlation between energy use and wealth, and ended with the question of which direction the causality went: does using more energy cause countries to be rich, or does being rich cause countries to use more energy?
Either one is theoretically plausible.
Rich countries eat more food per capita (measured in calories) and they eat more expensive food (measured as a greater portion of their food coming from animal sources—meat and dairy). It’s unlikely that they’re rich because they eat more food and eat more animal-based foods. Rather, they eat the way they do because people like to eat more, people like to eat animal-based foods, and people in rich countries can afford to do those things.
The analogous argument would run that people like doing things that use energy, and people in rich countries can afford to do more of those things, so they use more energy.
I’m inclined toward the opposite causality, that rich countries generally get that way by using more energy.
The theoretical basis for that view is that every economic activity requires energy. Some require more than others (earning $40,000 cutting hair takes less energy than earning $40,000 making steel). And the energy requirements of a given activity can change over time as technology changes. But every economic activity requires some energy, so at some level, having more economic activity (a higher GDP) should require more energy.
Faced with a theoretical argument pointing in each direction, it would be good to have some empirical indication of which force is more likely in play.
A test suggests itself from the fact that energy prices have shown significant volatility over the last 45 years, with oil spikes in 1973 and 1979, a crash in 1986, a trough in 1998, a new peak in 2008 and then continued ups and downs more recently.
If rich countries are buying energy sort of as a luxury, because they can afford it, we should see them buying less energy when prices go up, and more energy when prices go down. We should see more energy use per GDP in times of low energy prices, and less energy use per GDP in times of high prices.
If countries are getting rich because they’re using lots of energy, then we should see no particular connection between the energy/GDP relationship and energy prices. High energy prices may cause countries to buy and use less energy, but that will also be reflected in reduced economic activity, rather than in decreased energy per unit of GDP.
The BP energy dataset (see the post about the sources of the data) provides price information for oil, coal, and natural gas, which together accounted for 86% of the world’s energy use in 2016, and 94% in 1965.
The oil price series goes all the way back to 1861, but their gas prices only start in 1984 and their coal prices only start in 1987.
I constructed an index of fossil fuel prices from 1987 through 2016 that combines the oil, coal, and gas prices. Rather than simply averaging them, I used a weighted average, where each fuel’s price has a weight that reflects the share of global energy it provided in a given year.
The results are shown in Figure 1, along with the price of oil from 1965 through 2016. The price is given in 2016 US dollars per tonne of oil equivalent (toe)—that is, the prices are adjusted for inflation, and shown in terms of dollars that have the purchasing power that the dollar had in 2016.
|Figure 1. Data from BP Statistical Review|
You can see that the price of oil hasn’t moved exactly in lockstep with the overall price level including coal and gas. In particular, during the 2000’s coal and gas didn’t rise as dramatically as oil, so the weighted average price is similarly muted in its movements during that time. But in terms of broad movements, the oil price and the weighted average tell the same story.
Figure 2 is a repost from part I of this discussion, showing the relationship between energy and GDP for 1970 and 2014, with the scatter of points for 2014 looking like it’s flatter than 1970.
|Figure 2. Data from BP Statistical Review and Penn World Tables|
To be more systematic about it, I constructed a set of 52 countries that were in the data in 1970 and stayed in it through 2014. This unfortunately left out big countries like the Soviet Union, which ceased to exist after 1991, and Russia, which didn’t exist as an independent entity until 1992.
It also left out some small members of the Organization for Economic Cooperation and Development like the Czech Republic and Slovakia, which don’t have their own data until 1990 (just before they emerged from the former country of Czechoslovakia).
And it omits many poor countries whose data systems were late to develop.
But it does include the U.S., China, Canada, Japan, India, and the major countries of Western Europe, so it covers a large portion of the world’s population and of the world’s economic activity.
And it has enough poor and middle-income countries to have a good spread of wealth levels.
Most importantly, by sticking to the same sample for those 45 years, I avoid the possibility that changes over time are due to looking at a different mix of countries over time.
For each of those 45 years, I ran a simple-minded regression which attempts to explain GDP per capita as a function of energy use per capita.
The adjusted r-squared is, roughly speaking, a measure of what portion of the variation in GDP can be explained by the variation in energy use: if all you knew was the energy use, how reliably could you predict the GDP?
That statistic ranged from 0.63 to 0.76, meaning that energy use explains roughly 70% of the variation in GDP.
Figure 3 shows the results of those 45 regressions.
|Figure 1. Data from BP Statistical Review and Penn World Tables|
The orange line in the middle is central estimate of the slope of each cluster (for a given year, as energy use goes up by 1 toe per capita, how much will GDP per capita go up).
The gray lines above and below the orange show the lower and upper bounds of the 95% confidence interval: we don’t know exactly what the true slope is, but we’re 95% confident that it lies somewhere between those gray lines.
This is the thing that we need to compare to the price indicators shown in Figure 1.
If the causality goes from being wealthy to using a lot of energy, then we should a relationship between this estimated coefficient and the energy price.
More specifically, low energy prices should be connected with countries buying more energy for a given level of wealth, which would mean a smaller amount of extra GDP for each extra unit of energy consumed.
At low prices, our estimated slope should be low (not that much more GDP from more energy), while at high prices our estimated slope should be high.
Figure 4 shows the slope estimate and the energy-price indicator on the same chart.
And there is no visible relationship.
|Figure 4. Data from BP Statistical Review and Penn World Tables|
From 1970 to 1995, the estimated slope rises at a modest, remarkably even rate. It is equally unaffected by the great rise in prices from 1973 to 1980 and by the fall from 1980 to 1985 or the crash in 1986.
Then in 1995 the slope begins a 12-year run of increasing more quickly: a greater increase in energy efficiency of these 52 economies.
The first eight years of this rise coincide with continued low energy prices, continuing unaffected even through 1998, which had the lowest prices since 1973, the year of the Arab oil embargo that launched the modern widespread concern about energy reliability. If low energy prices encouraged more use, we should have seen the orange line tending down during this period, not starting to rise faster.
Rapidly increasing prices from 2003 to 2008 coincide with increasing efficiency, but that’s just the continuation of a trend begun when prices were low.
The strongest connection is 2008-2010, where energy prices plummet and bounce back, and the GDP/energy slope dips down and rises back. But in the context of the chart as a whole, that’s precious little evidence of a link.
The absence of a link between energy prices and the GDP/energy slope points away from the hypothesis that wealth causes us to use more energy, and towards the alternative idea that using more energy is what makes us wealthy.
From the perspective of doing something about greenhouse-gas emissions, this is not an encouraging conclusion, but I’ve put aside the more promising fact revealed in the data so far, which is that there is an increase in energy efficiency revealed in the rising level of the orange line in Figures 3 and 4. And the lower bound in the last 12 years of the data is higher than the upper bound in the first 19, which means that we can be highly confident that the estimated increase in efficiency is something real, rather than statistical noise.
This post has provided some evidence that our energy use isn’t simply a luxury that should be easy to give up if we finally decide we have to. But if we can increase the energy efficiency of our economies enough, maybe we can make real progress on climate by moving further and faster in that direction.
The next post will return to that question.