Demand Elasticity for Gasoline

by Doug Short and John Lounsbury

What is the relationship between retail gasoline prices and the volume of gasoline sales? The first chart below shows the monthly data for U.S. Prime Supplier Sales Volumes, courtesy of the Depart of Energy’s Energy Information Administration (EIA). The numbers are updated monthly with about a two month lag. The numbers are highly volatile and have a distinct seasonality, so I’ve added a 12-month moving average (MA) to facilitate our analysis.

The next chart includes an overlay of monthly retail gasoline prices, all grades and formulations. The retail prices are updated weekly, so the price series is the more current of the two.

As we would expect, the rapid rise in gasoline prices in 2008 was accompanied by a significant drop in sales volume. With the official end of the recession in June 2009, sales reversed direction … slightly. But the 12-month MA of volume for the latest month (January 2011) is still about 10.3% below the pre-recession level. The dramatic rise in gasoline prices since last September has yet to be seen in sales volumes because of the 2-month lag. But the average of daily sales for January 2011 was the lowest January number in a decade.

The next chart adjusts the 12-month MA of sales volume for population growth based on the monthly adult Civilian Non-Institutional Population data from the Bureau of Labor Statistics, via the St. Louis (FREDrepository. What we see here is that gasoline sales volume, on a per-capita basis, peaked in September 2009. In fact, our per-capita consumption of gasoline is slightly lower than it was at the end of the Great Recession.

What does this analysis suggest about the state of the economy? From an official standpoint, the Great Recession ended 22 months ago. But if we want confirmation that the economy is in recovery, gasoline sales is the wrong place to look.

In the previous graph there is a clear inverse realtionship between consumption per capita and price.  The same relationship exists in the second graph but is more hidden by noise in the data.  There is a clear implication that there is some elasticity in the demand for gasoline over multi-year spans of time.

The following graph from Wikipedia displays the classical picture of price elasticity of demand:  as prices rise sales fall.  Demand adjusts downward in response to higher prices.

There are two extreme conditions for PED plots:  One is perfect inelasticity and the other is perfect elasticity.  When there is perfect price inelastic demand, the demand is constant at all prices, as shown in the first graph below.  The other extreme is perfect price elastic demand where all levels of demand command the same price, as shown in the second graph below.

Editor’s note:  In the lower graph above Wikipedia has printed D1 when they should have printed P1.

Using the same data as the first three graphs, the following three graphs have been created.  The first is for the two years 2009 and 2010, and includes the latest data for January 2011.

The graph is similar to the perfect price elastic demand graph – there is little price change for all levels of demand change.  The weak elasticity effects change from a negative slope for the most negative demand changes (the normal relationship) to a positive slope for higher levels of demand change ( an abnormal relationship).  Some of the situations that produce such positive slopes are discussed at Wikipedia (here and here), but neither of these examples fit the gasoline situation we are discussing. 

Looking at the two preceeding years (2007-08) we find the following graph.

In this graph also there is a situation which is close to the perfectly elastic demand graph.  This illustrates a relatively elastic demand for gasoline during the years 2007-08 which encompassed the record oil price spike in 2008.  Again there is a region where the slope of the PED curve trend line is positive (declining demand), which the authors consider an anomaly.

One question that occurs is that gasoline demand has significant seasonal variations, such as the peak summer driving season.  These seasonalities may be distorting an underlying elasticity of demand in the month to month change data, as discussed at the beginning of this article.  In the following graph the price and  sales changes are plotted year-over-year.

Surprisingly, this attempt to reduce the seasonality factors in the data has produced a large region of positive slope.  The more rise in price the more the demand tends to increase.  This might make sense if a price increase produces fear of further price increases and therefore increases consumption.  It might be called a hoarding syndrome.  However, there are also periods of falling prices.  To avoid having the expectation factor during price increases cancelled out during price declines, sales would have to fall (or increase significantly less than they rose when prices were rising).

Is the graph above telling us that fear and hoarding has greater effect than rejoicing with increasing consumption when prices fall?

In conclusion, the implication of some multi-year price elasticity of demand is supported with month-to-month data, although anomolous (abnormal) relationships are seen.  The YoY data displays an abnormal positive slope tendancy which implies rising demand with rising prices over much of the data range.  The YoY data also shows less price elastic demand (steeper slope) than the month-to-month data studied.

All observations here must be qualified with recognition that the scatter about the trend lines is large.  The R-squared values for the three graphs (not displayed on the graphs) were 0.026, 0.1063 and 0.3326.

5 replies on “Demand Elasticity for Gasoline”

  1. Doug and John —
    This is very interesting. Thanks for all of the hard work. I have a question and an observation.

    Why not seasonally adjust the data? You acknowledge the pronounced seasonal effect, and everyone knows about it. The moving average approach is fine for smoothing unexplained volatility to get the trend, but here you are losing valuable information. The seasonally adjusted month-to-month data would be much better for capturing the elasticity effect.

    Concerning the conclusion, I think we are all interested in the elasticity question, but I don’t understand why you interpret this as some measure of the economy. The decrease in fuel consumption reflects changed behavior in many ways including greater use of mass transit, more care in taking discretionary trips, and purchases of more fuel-efficient vehicles. We expect prices to stimulate changed consumer behavior, and it does not imply economic strength or weakness. There is also a long-term trend toward fewer miles driven, especially miles driven per licensed driver. This is probably a better measure than overall population.

    As with most economic topics, there are many choices and approaches.


  2. Jeff – – –

    I agree that the statement about economic strength makes an implicit assumption that consumer behavior is not changing at the baseline. That may well be an achilles heel.

    I also agree that it would be worthwhile to also look at the seasonally adjusted month-to-month data. However, I see no merit in using seasonally adjusted data for year-to-year analysis. In fact, if there is a different behavior of the year-to-year data if seasonally adjusted numbers are used, I would interpret that to impugn the validity of the seasonal adjustment methodology. If there are differences over 16 years of data then I would propose that all Januaries do not average out to the same seasonality, and the same for all other months. My interpretation is that seasonality would have parameters that extended beyond season.


  3. John — Concerning seasonal adjustment, I specifically had in mind the two tables entitled Gasoline Sales vs. Price. One is from 2007-08 and the other from 2009-10. This is only four years of data and you are trying to make an interpretation of month-to-month changes. If ever there is a case for using seasonally adjusted data, this would seem to be it.

    I don’t mind the year-over-year comparisons on the other part of the analysis, for the reasons you give. Sorry I was not more clear.


Comments are closed.