Submitted by Erico Matias Tavares via Sinclar & Co (co-authored with DegreeDays.net [5]),
A year ago we wrote about [6] how electricity consumption could provide clues to the performance of the US economy, which generated a lot of interest and comments.
A relationship between the two variables makes sense, but needs to be framed in the proper context. Genuine economic (and population) growth should translate into more electricity consumption, as we have more activity and transactions taking place throughout the economy.
However, factors such as energy efficiency and the weather can muddle this relationship:
- An increase in efficiency means that the same output can be obtained with less inputs. Therefore, a small-ish reduction in electricity consumption versus a prior period may not necessarily be indicative of a sluggish economy over that time. And we know that this efficiency has been on the rise in recent years (just look at the power rating of your new appliances).
- Likewise, a warmer winter versus the prior year may also cause a drop in electricity consumption, simply due to home heaters not being used as hard, not necessarily because the economy is doing badly.
So can we adjust electricity consumption to take these factors into consideration and get a better measure of its relationship with economic growth?
We developed an indicator to do just that together with DegreeDays.net [5], an energy systems data company. We provide a brief technical explanation of our proposed methodology below (for a much better overview please visit this supporting article [7]). Bear with us, the analysis is quite interesting!
To account for weather variations, we regressed the weekly electricity consumption data against US-population-weighted heating degree days (“HDD”) and cooling degree days (“CDD”) that we calculated for matching weekly periods:
- Degree days are a specialist type of data derived from detailed temperature readings and used for analysis of heating and cooling energy consumption. This is exactly what we need to normalize the effects of changes in the weather on electricity consumption.
- Energy required to heat a building over a period is proportional to the HDD over that period; energy required to cool a building over a period is proportional to the CDD over that period. Calculating this is pretty straightforward for one building in a specific location.
- But in order to account for the millions of different buildings and variations in weather patterns across the country we calculated population-weighted HDD and CDD using census per city and weather data (again, this is much better explained in that article [7]).
Accounting for energy efficiency is a bit trickier. We did not find any recent data (and which is updated regularly) that could measure this effect. So we decided to run those regressions over much shorter time frames. The rationale being that longer periods could lead to an apples and oranges comparison since the installed power demand base changed. For instance, a new refrigerator consumes much less electricity than a decade ago.
So at the start of each calendar year we ran the regression over the prior 12 months. We then estimated electrical consumption using the regression coefficients and the updated population-weighted HDD and CDD variables, compared it to the actual electricity consumption each week and voilà, we came up with an adjusted indicator based on the percentage differential between the two.
If actual electricity consumption is higher than the one predicted by our regression, which as we have seen should be adjusted for weather changes and some efficiencies, this is a sign that the economy should be doing well. If it is considerably lower then the economy may not be doing so great (or everyone is buying those new refrigerators).
OK! Let’s look at the actual results.
First, some context. Here’s the actual electricity consumption in the US without any adjustments since 1995:
Weekly US Electricity Consumption (MM kw hrs): Jan 95 - Present
Source: EEI.
As we pointed out last year, this is not exactly a picture of economic dynamism. Since then electricity consumption across the US has pretty much remained in the doldrums:
- The red line shows the weekly historical peak, reached all the way back in the summer of 2006. Notice how far we have been from it in recent years, despite all the GDP and population growth that has occurred since then.
- The black line, which is the smoothed consumption data over time (the longer-term trend if you prefer), has recently turned negative.
It is curious to note that financial commentators regularly gauge China’s economic performance by looking at its electricity consumption [8], but this is not done for the US. Perhaps there is something there that does not fit with the prevailing narrative. Sure, China remains a manufacturing economy while the US is largely a services economy; but as far as we know restaurants, insurance companies, hospitals and IT service providers still use electricity.
In fact, growth in data has been so significant – requiring ever more power-hungry data centers – that IT now consumes some 10% of world electricity production [9]. All the iPhones, clouds, tablets, chats, likes and whatever else have materially increased our demand for electricity, both on the front- and the back-end. So it is surprising that electricity consumption statistics haven’t been a bit perkier in the US, given all the IT development there.
And here’s the evolution of annual electrical generation in the world’s leading economies/blocks since 2006:
Annual Electrical Generation Index (‘06=100): ’06-‘14
Source: BP World Energy Review.
The US (dotted line) has pretty much underperformed everyone in the group in terms of growth (except in 2014 when OECD ex-US, Canada dropped below it). Even Canada, its northern neighbor, has done better. What does this tell you about the US “decoupling and leading global growth"? That Americans must have much more efficient refrigerators and data centers than everyone else? Or maybe there is more to this story.
With that in mind, let’s finally put our “energy indicator” to use. We aggregated weekly data into quarters because as you can imagine we are dealing with quite a noisy data series. Here are the results, compared to year-on-year (“y-o-y”) Real GDP growth (red line):
Quarterly Real GDP y-o-y Growth and the Energy Indicator: 1Q’01 – 1Q’15
Source: EEI, BEA, degreedays.net.
The graph brings out some interesting points:
- Overall, the percentage differential between actual and predicted (by our methodology) electricity consumption reflects the ebbs and flows of the GDP cycle.
- At time the energy indicator precedes GDP turns, although it may take some quarters for this to become evident (again, we are dealing with noisy, unsmoothed data, unlike quarterly GDP which has a host of seasonal adjustments). Notice the jump in 3Q’02, before GDP growth recovered in earnest. And in the run up to the Great Recession in 2008, we had the biggest drop in the indicator in the series while official GDP statistics were still showing economic growth.
- An advantage of our methodology is that we can update the indicator on a weekly basis, and so at the end of each quarter we can have a sense of how the economy performed almost in real-time. The last (red) bar is the accumulated reading to date for 2Q’15, which being positive suggests also a positive y-o-y Real GDP growth in 2Q.
- However, a divergence has formed since early 2013: the trend in the energy indicator is going down, while y-o-y GDP growth seems to be moving higher (notice the two arrows).
OK, so what accounts for that divergence? We can’t say for sure, but back in 2013 the Bureau of Economic Analysis “modernized [10]” its GDP accounting methodology, to include things like R&D, copyrights and pension deficits. As a result total GDP increased by 3%. Presumably the series was updated as far back as 1929 but we can certainly debate whether such variables reflect actual transactions and human activity, which is what our indicator picks up.
Other countries have been even more creative in GDP accounting revisions, adding estimates for illegal activities such as prostitution and drug consumption. Ah, but we already have these covered, as presumably they also consume electricity (even if under very dim lights).
You can take your pick as to which measure you would like to focus on to gauge real economic performance.
But just looking at electricity consumption, adjusted or otherwise, things are definitely looking very sluggish in the US economy right now.
