Do Energy Prices Drive the Long-Term Inflation Expectations of Households?
Between July 2014 and January 2015, the average price of gasoline fell by more than 33 percent. This decline in gas prices has significantly impacted the Consumer Price Index (CPI), which has actually fallen every month since October 2014. Both the dramatic fall in gasoline prices and the recurrent drops in the CPI raise the question of how these developments are affecting the long-term inflation expectations of households. This is a question worth pursuing, because anchored long-term inflation expectations are important for promoting short-run inflation stability and for facilitating central bank efforts to achieve output stability.
Over the past several months, two noted measures of long-term inflation expectations have slipped somewhat: the expectations of households from the Thomson Reuters/University of Michigan Surveys of Consumers (UM Survey) and the expectations estimated by the Federal Reserve Bank of Cleveland (FRBC) based upon financial market data and professional forecasts. Are these expectations anchored or are they driven by energy-price changes? Some prominent analysts have argued that household inflation expectations respond strongly to changes in energy prices. But previous analysis has focused mainly on shorter-term inflation expectations. We examine the role played by energy prices in influencing long-term inflation expectations relative to the impact of movements in the CPI and other macroeconomic variables
Household inflation expectations are measured in a national survey. Each month, the Survey Research Center at the University of Michigan surveys about 500 households and asks them questions about economic conditions. One survey question asks consumers about their inflation expectations five to ten years ahead. (“By about what percent per year do you expect prices to go (up/down) on average, during the next 5 to 10 years?”) We study the median response.
Each month, the Federal Reserve Bank of Cleveland estimates inflation expectations at various horizons. These estimates are based upon inflation swap data in conjunction with nominal Treasury yields and survey information from professional forecasters. Using the published five-year and ten-year inflation expectations estimates, one can construct inflation expectations that match the implied horizon of the UM Survey.
We examine the potential influence of seven different variables on long-term inflation expectations. Examining the role of these seven variables simultaneously allows us to properly determine the role played by energy-price inflation. To conduct the analysis, we use a particular type of statistical model that is used to inform forecasting and policy analysis at the Cleveland Fed, a structural Bayesian vector autoregression (SBVAR). This is a statistical model which simultaneously measures the cross-correlations between many different variables—including their cross-correlations across time—and entails assumptions about contemporaneous influences between variables. (For details, see Binder, Higgins and Verbrugge, forthcoming.) In our analysis, we assume that energy prices are not contemporaneously influenced by any other variable within the current month. This is the conventional assumption.
Four of the seven variables are elements of the CPI: energy prices, food prices, shelter prices, and the overall CPI. To measure overall CPI trends, we use smoothed monthly changes in the median CPI, which performs well as a predictor of future CPI trends and is far superior to the so-called “core CPI” in this regard. We measure energy-price movements, food-price movements, and shelter-price movements relative to the overall CPI movement.
The remaining three variables are macroeconomic variables. We include these because inflation expectations may respond to economic activity and to monetary policy. The macroeconomic variables are: the Chicago Fed National Activity Index (CFNAI), which measures economic activity changes on a monthly basis; the unemployment rate; and the federal funds rate. The inflation expectations variable is either the UM Survey or the FRBC measure.
We relate movements in inflation expectations (measured in percent) to movements in each of these variables (measured in percent). Our analysis is based upon monthly changes. We estimate from mid-1990 through the first month of 2015; UM Survey data on long-term expectations are not available on a monthly basis prior to mid-1990.
The figure below plots the response of UM Survey expectations to a typically sized shock to energy prices. The impulse response is plotted for 12 months after the shock, along with error bands. If zero does not lie in between the error bands at a particular month, then—on a statistical basis—there is reliable evidence that the impact is distinguishable from zero. We see that energy-price shocks do indeed have a noticeable and statistically significant influence on UM Survey expectations for at least 12 months. However, the response is quite small. A typically sized positive energy-price shock raises UM Survey inflation expectations by just over 0.03 percentage points the month it happens; thus, for example, if UM Survey inflation expectations had been 3.0 percent without the shock, this shock would raise long-term inflation expectations to 3.03 percent. But 12 months later, the effect of the shock is only 0.01 percentage points. We do not depict any of the other impulse responses, as there is very little evidence for an influence of any of the other six variables—including movements in the median CPI.
The figures below plot the response of FRBC inflation expectations to a typically sized shock to energy prices and to the CFNAI. As with UM Survey expectations, the impacts are statistically significant, but economically small. Energy-price shocks impact FRBC inflation expectations, with the estimated effect rising to a little above 0.03 percentage points two months after the shock and then falling to slightly below 0.02. The effect of an energy-price shock is, however, indistinguishable from zero after seven months. Conversely, shocks to CFNAI have a persistent impact on FRBC inflation expectations. A typically sized CFNAI shock raises FRBC inflation expectations by roughly 0.03 percentage points, an effect that persists for a year. None of the other variables has an appreciable impact on FRBC inflation expectations.
The impulse responses relate to typically-sized shocks, but the most recent six-month movement in energy-price inflation is far from typically sized. Suppose we do a simple exercise and assume that the entire drop in energy prices actually happened over the period of one month. This amounts to a shock that is about eight times bigger than is typical. A back-of-the-envelope calculation says that such a shock, all by itself, would lower UM Survey inflation expectations the next month by about 0.24 percentage points and would lower FRBC inflation expectations by about 0.22 percentage points. Since June 2014, both measures of expectations have dropped by about 0.2 percentage points. Hence, recent drops in energy prices can potentially explain the recent slippage in both of the inflation expectations measures we examine. Further, this suggests that a rebound of energy prices would lead to a similar rebound in long-term inflation expectations.
We next go on to look at variance decompositions, which measure how much of the variation in inflation expectations ultimately derives from shocks to another given variable over various time horizons. We focus on the impact of energy-price shocks as compared to “other macro variables” (the combined effects of the three macroeconomic variables—the CFNAI, the unemployment rate, and the federal funds rate), “other price variables” (the combined effect of shocks to food inflation, shelter inflation, and overall CPI inflation), and “unexplained” (the effect of shocks to expectations themselves). We first look at the variance decomposition for UM Survey inflation expectations.
|Impact of Shock (percent)|
|Area||Energy Prices||Macro Variables||Price Variables||Unexplained|
|3 months later||7||0||3||90|
|6 months later||8||0||3||89|
|12 months later||8||1||4||87|
While energy-price shocks clearly matter, over the 1990-2014 period they generally account for less than 10 percent of the variance of UM Survey inflation expectations, despite being the single biggest explanatory factor. While long-term household inflation expectations do respond to incoming data, this response is muted. Even at the 12-month horizon, the vast majority of the variance is unexplained by other variables.
Next we consider the variance decomposition for FRBC inflation expectations.
|Impact of Shock (percent)|
|Area||Energy Inflation Shocks||Shocks to Macro Variables||Shocks to Other Price Variables||Unexplained|
|3 months later||5||5||1||89|
|6 months later||5||7||1||87|
|12 months later||5||9||1||85|
For FRBC inflation expectations, energy-price shocks are usually even less important, accounting for 5 percent of the variance at most. In fact, they are generally less important than shocks to macroeconomic variables, chiefly CFNAI shocks. At all horizons, the vast majority of the variance of FRBC inflation expectations is unexplained by other variables.
Such results suggest that both UM Survey and FRBC inflation expectations are well-anchored, in the sense that they are “relatively insensitive to incoming data,” as former FOMC chair Ben Bernanke defined “anchored.” Of course, very dramatic movements in energy prices will still shift inflation expectations to some extent, as we noted above.
What drives inflation expectations? While energy prices have a greater influence on the long-term inflation expectations of households (UM survey) than on the FRBC measure of inflation expectations, their quantitative influence is generally quite modest. Shocks to energy prices explain very little of the usual variation in either UM Survey inflation expectations or FRBC inflation expectations. But the recent drops in energy price inflation are far from usual, and we show that these unusual energy-price movements can potentially explain the recent drops in both inflation expectations measures.
While both UM Survey inflation expectations and FRBC inflation expectations appear to be well-anchored, in the sense that they are relatively insensitive to incoming data, these expectations are not identical. Household long-term inflation expectations are much higher. Also, these expectations are differently influenced by shocks; for instance, while macroeconomic shocks impact FRBC inflation expectations, they do not appear to impact UM Survey inflation expectations. What accounts for such differences in expectation formation? These are questions left for future work, but we briefly mention two conjectures.
One potential explanation is that a feature of the UM Survey might be unduly influencing the estimated expectations. When asked about their inflation expectations, households are required to give an integer response. We do not view this restriction as innocuous. Previous research indicates that this kind of rounding can influence actual aggregate estimates and outcomes (see, e.g., Schweitzer and Severance-Lossin (1996) and Knotek (2011).) However, recent evidence from the New York Fed’s Survey of Consumer Expectations suggests that when consumers are asked about inflation, they often give integer responses, even if they need not do so.
Another possibility is suggested by the recent research of Carola Binder. She reviews a large number of surveys given to consumers from the 1950s until the present and finds that consumers are often confused about the link between monetary policy and inflation and that many fear the return of the high inflation experience of the 1970s. In studying the UM Survey data, she also finds that households differ in their uncertainty about future inflation—likely due to differences in financial literacy—and that the expectations of the more certain households are more accurate (Binder 2014a). It is possible that the expectations of more certain or more informed households are closer to those estimated by the Federal Reserve Bank of Cleveland.
Binder, Carola Conces, Amy Higgins, and Randal Verbrugge (2015). “What Drives the Long-Term Inflation Expectations of Households?” Manuscript in preparation.
Knotek (2011) “Convenient Prices and Price Rigidity: Cross-Sectional Evidence” Review of Economics and Statistics 93(3), 1076â€“86.
Schweitzer, Mark and Eric Severance-Lossin (1996) “Rounding in Earnings Data.” Federal Reserve Bank of Cleveland Working Paper, no. 96-12.