This is the first in what I’m planning will be a series of posts discussing the contribution that the energy price spike of 2008 made to our present economic difficulties. In this first installment, I revisit a very interesting research paper on the response of consumer spending to energy price increases written by Lutz Kilian (Professor of Economics at the University of Michigan), and Paul Edelstein (Senior Economist for Decision Economics). I first brought this paper to the attention of Econbrowser readers in the spring of 2007. I thought now would be a good time to take a look at how well the equations in Edelstein and Kilian’s paper can describe what we saw happen in the later part of 2007 and first half of 2008.
Edelstein and Kilian summarized the historical correlations between energy prices and economic activity in terms of some basic forecasting equations. They started with a simple model that could be used to predict real personal consumption expenditures on the basis of past consumption spending and past changes in the relative price of energy. Their measure of energy prices was scaled by the share of energy purchases in total spending, so that if their energy-price measure goes up by one unit, it means that a consumer who tried to consume the same amount of energy as before would have to cut total real spending on other goods and services by 1%. In the graph below, I have reproduced their calculation for how their forecast of real consumption expenditures in month t+s would change if an energy price increase reduced disposable income by 1% in month t. This calculation is plotted as a function of s, how many months into the future we’re trying to forecast, so that as you move to the right along the graph, you’re looking at the consequences farther and farther into the future of an event (an energy price increase) that occurs now.
The graph shows, not surprisingly, that when energy prices go up, consumer spending falls. But there are two surprising things about the quantitative character of this response. The first surprise is the delay– energy prices go up at time t, but the biggest consequences for consumption spending aren’t seen until s = 12 months later. The second surprising feature of these results is the magnitude. If consumers continued to purchase the same number of gallons of gasoline as they had before, a shock of the size analyzed in this graph would require them to reduce spending on other items by 1%. Yet eventually they historically would be predicted to reduce spending by 2.2%. Why do consumers cut spending by so much more than the shock itself?
Edelstein and Kilian proposed to find some answers to these questions by looking at specific components of consumption spending. The three graphs shown below break spending down into services, nondurables, and durables. The magnitude of the response of the first two categories– about a 1.5% reduction in spending on services or nondurables– is more in line with what we might have expected. The real action is in durables. Purchases of items such as appliances and automobiles respond very dramatically to a change in energy prices.
Looking further at the composition of durable goods, Edelstein and Kilian found that by far the most dramatic response is in terms of spending on motor vehicles and parts. Unlike the other categories of spending, this response is huge and immediate.
Once we state the basic facts this way, the results seem a lot less mysterious. The response of spending on autos to an energy price increase is not simply a matter of consumers having less disposable income, but instead involves a lot of other factors, such as consumers switching to smaller cars or postponing purchases until they have more confidence about where gasoline prices are going to settle down. As a result of decreased spending, the income of those employed in the auto sector would decline, and for these individuals, their loss in purchasing power is much greater than their personal increase in fuel expenses. As people who derive their income from manufacture and sale of automobiles and parts reduce their spending, the aggregate consequences can accumulate over time, accounting for both the magnitudes and the delays in the response of other categories of consumption spending.
Another interesting finding by Edelstein and Kilian was that consumer sentiment is quite sensitive to energy prices, with a 1% loss in purchasing power quickly translating into a 15% drop in the Reuters/Michigan index of consumer sentiment.
The economics literature is filled with statistical models that seem to do a very good job at fitting the data in hand, but then fall apart when applied to data that arrive after the paper is published. I was curious to see how well the Edelstein-Kilian forecasting equations would have performed over the period since their paper was written. The heart of the system used to generate the graphs above is an equation that predicts a category of consumption spending one month ahead on the basis of the current and past consumption spending and energy prices. I took the parameters as estimated from the historical data set used by Edelstein and Kilian (1970:M1 to 2006:M7), and formed forecasts of each new observation between 2006:M8 and 2008:M9, one month at a time. I then calculated the average squared error of those forecasts. These mean squared forecast errors are reported in the first column of the table below. These are compared in the second column with the errors from a model that forecasts the spending component from its own lagged values alone. For each of the equations used in the graphs above, the Edelstein-Kilian specifications exhibit significant post-sample success.
One can also use these equations to estimate how much of a contribution energy prices made to recent developments. I used the Edelstein-Kilian regressions to form a forecast for the path of real consumption expenditures looking forward from September 2007. That forecast is indicated by the blue line in the graph below, and basically called for consumption spending to continue along the same sort of path we saw earlier in 2007 and 2006. The actual path for spending, indicated by the black line, grew much more sluggishly than this after September 2007 and began to decline in June 2008.
By the nature of this system, one can by definition decompose the gap between the black and blue lines into a series of individual forecast errors, namely the sequence of errors we made trying to predict energy prices one month ahead, and the month-by-month errors we made predicting consumption given those energy prices. One can then use such calculations to ask the following question: What if there had been no surprises with energy prices, and the only errors had been those associated with trying to predict consumption spending conditional on knowing energy prices? [This is an easy calculation to perform using the RATS “history” command]. The result of that “what-if” calculation is plotted as the green line above. The difference between the green line and the blue line could be interpreted as the fraction of the slowdown in consumption spending that one might attribute to energy prices alone. The graph suggests that energy prices played a very significant role in the slowdown in consumption spending.
I performed a similar calculation for spending on motor vehicle and parts, and found that the oil shock made a major contribution to the problems for this sector for 2007:Q4 through 2008:Q2.
Energy prices were surely also a significant factor in the dramatic deterioration of consumer sentiment in the early part of this year.
My conclusion is that the oil price shock made an important contribution to consumption spending in general and purchases of motor vehicles and parts in particular in the later part of 2007 and first half of 2008. In my next post, I intend to discuss the implications of this for the economy as a whole.