Economists are slightly better at predicting inflation than consumers
The jump in consumer price index (CPI) inflation in April 2021—to 4.2 percent over the previous 12 months—has sparked numerous stories in the press about the potential inflationary impact of President Joseph R. Biden Jr.’s American Rescue Plan and his proposed American Jobs Plan and American Families Plan. A survey of consumers finds a modest increase in expected inflation over the next year, with little change to longer-term expectations. A survey of professional economic forecasters, which has a better track record of predicting inflation, shows little change in expected inflation over both the short term and the long term. Inflation has been declining or stable over almost the entire historical record of these surveys, however, so that neither has been tested for its ability to predict a significant and sustained increase in inflation.
Two previous posts (here and here) showed that bond yields (back to the 1950s) and the inflation component of bond yields (available since the 1990s) are not good predictors of future inflation. This post examines surveys of consumers and professional economic forecasters. Consumer expectations are shown to have little useful information about future inflation except for a brief period in the 1980s and then only for one year ahead. Expectations of professional forecasters help to predict inflation over both the short term and the medium term (up to five years), and this predictive power is not limited to the 1980s.
Figure 1 displays the median consumer expectations for inflation one year and five years ahead from the University of Michigan Surveys of Consumers. Also shown is the percent change in the CPI in the four quarters prior to each survey date. Consumers missed the runup of inflation in the late 1970s, but they correctly anticipated the decline of inflation in the early 1980s. The short-term expectation of inflation fell somewhat ahead of the permanent decline of inflation in 1991, but the long-term expectation did not. Since then, both measures of expectations typically moved concurrently with inflation rather than in anticipation of inflation. Consumers have overpredicted inflation on average by 0.6 percentage point (one year ahead) since 1990.
Figure 2 displays the median expectations of professional economic forecasters for inflation one year and ten years ahead from the Federal Reserve Bank of Philadelphia’s Survey of Professional Forecasters. The figure also displays an alternative measure of inflation based on the CPI excluding food and energy (core CPI), as this less volatile measure may be easier to forecast. Since 1990, long-term expectations overpredicted subsequent 10-year inflation by 0.6 percentage point on average for CPI and 0.7 percentage point for core CPI, but short-term expectations have been very close to one-year-ahead inflation, overpredicting by only 0.2 percentage point on average for CPI and core CPI. Since the late 1990s, long-term expectations have been remarkably stable, whereas short-term expectations seemed to move mostly concurrently with actual inflation.
Table 1 presents statistical analysis of the predictive power of consumer surveys. Column 1 displays the results of a regression of 12-month inflation on the value of 12-month inflation lagged by 12 months and on the survey measure of one-year-ahead inflation lagged by 12 months. If the survey measure were the best possible predictor of future inflation, one would expect it to have a coefficient of 1 and the constant term and lagged inflation term to have coefficients of 0. This is not far from the case, as the survey measure is highly statistically significant with a coefficient of 1.07, whereas the coefficient on lagged inflation is small and not significantly different from 0. The constant term is negative and is economically, though not statistically, significant. A large negative constant term is further evidence that the survey measure overpredicts inflation on average. The regression in column 1 finds that movements in survey expectations tend to be good predictors of movements in future inflation, even if the level of future inflation generally tends to be lower than the level of expectations. The overall predictive power is fairly high, with an R2of 0.65.
|Table 1 Regressions of inflation on lags of inflation and consumer survey expectations|
|Full sample||Post-1989 expectations only|
|12-month inflation||12-month inflation||60-month inflation|
|Lagged 12-month inflation||0.16||0.35**||-0.02|
|Lagged 1-year-ahead inflation expectations||1.07**||-0.77|
|Lagged 60-month inflation||0.34|
|Lagged 5-year-ahead inflation expectations||0.00|
|Note: Variables are lagged 12 months when regressing 12-month inflation and 60 months when regressing 60-month inflation. The first column uses monthly observations from January 1979 to April 2021, the second column from January 1991 to April 2021, and the third column from April 1995 to April 2021. Newey-West standard errors in parentheses. ** and * denote significance at 1 and 5 percent levels, respectively.|
|Sources: Authors calculations using data from U.S. Bureau of Labor Statistics (via Macrobond) and University of Michigan Surveys of Consumers.|
Column 2 presents the same regression limited to a sample that drops expectations surveyed prior to 1990. This removes the large swings of inflation in the late 1970s and early 1980s. Since 1990, consumers have not been good predictors of inflation. Indeed, inflation tends to move in the opposite direction from consumer expectations, as denoted by the negative (though not statistically significant) coefficient. The R2 is much smaller than that in column 1, indicating little predictive power.
Column 3 repeats the regressions of columns 1-2 using the five-year-ahead survey measure instead of the one-year-ahead measure. This medium-term projection became a regular part of the survey in 1990. Two measures of lagged inflation are included, a 12-month measure and a 60-month measure that corresponds to the time period of inflation being explained. A longer lag of past inflation may provide a better guide to future inflation over a longer horizon. Both inflation measures and the survey expectations are lagged 60 months, because the goal is to measure predictive power over a five-year horizon. None of the variables is statistically significant, but the long lag of inflation has a much larger coefficient than the short lag and the survey measure. Overall, it appears that consumers are not good forecasters of inflation over the short term and the medium term since 1990.
Table 2 repeats the exercise of table 1 using the one-year-ahead and ten-year-ahead survey measures of expected inflation from professional economic forecasters. Column 1 displays a regression of four-quarter inflation on lagged four-quarter inflation and lagged expectations of one-year inflation. The coefficient on expectations is 0.74 and highly significant, while the coefficient on lagged inflation is near 0 and not significant. Column 2 shows a regression of eight-quarter inflation on lagged four-quarter and eight-quarter inflation and lagged expectations. The R2 is even higher than in column 1, and the coefficient on expectations is also higher, 0.89, and strongly significant, while the coefficients on lagged inflation are near 0 and not significant.
|Table 2 Regressions of inflation on lags of inflation and professional forecaster survey expectations|
|Full sample||Post-1989 expectations only|
|4-quarter inflation||8-quarter inflation||20-quarter inflation||40-quarter inflation||4-quarter inflation||8-quarter inflation||20-quarter inflation||40-quarter inflation|
|Lagged 4-quarter inflation||-0.11||-0.06||-0.03||-0.03||-0.06||-0.07||-0.01||-0.06*|
|Lagged 8-quarter inflation||-0.21||0.00|
|Lagged 1-year-ahead inflation expectations||0.74**||0.89**||0.72*||0.59|
|Lagged 20-quarter inflation||-0.37**||-0.01|
|Lagged 40-quarter inflation||0.25||0.57**|
|Lagged 10-year-ahead inflation expectations||0.92**||0.11||0.52||-0.15|
|Note: Variables are lagged 4 quarters when regressing 4-quarter inflation, 8 quarters when regressing 8-quarter inflation, 20 quarters when regressing 20-quarter inflation, and 40 quarters when regressing 40-quarter inflation. The first column uses quarterly observations from 1982Q3 to 2021Q1, the second column from 1983Q3 to 2021Q1, the third column from 1986Q1 to 2021Q1, the fourth column from 1991Q1 to 2021Q1, the fifth column from 1991Q1 to 2021Q1, the sixth column from 1992Q1 to 2021Q1, the seventh column from 1995Q1 to 2021Q1, and the eighth column from 2000Q1 to 2021Q1. Ten-year expectations for 1981Q1-1991Q3 are from the Hoey survey. Newey-West standard errors in parentheses. ** and * denote significance at 1 and 5 percent levels, respectively.|
|Sources: Authors calculations using data from U.S. Bureau of Labor Statistics (via Macrobond), Federal Reserve Bank of Philadelphia Survey of Professional Forecasters, and Federal Reserve Board.|
Columns 3 and 4 display regressions of 20-quarter and 40-quarter (5-year and 10-year) inflation on lags of inflation and lagged expectations. The coefficients on lagged inflation are generally small and not significant, except that the coefficient on lagged 20-quarter inflation in column 3 is significantly negative. The coefficient on expectations is 0.92 and strongly significant in column 3 but close to 0 and not significant in column 4.
Columns 5-8 repeat the regressions of columns 1-4 after dropping expectations surveyed before 1990. The coefficients in column 5 are nearly identical to those in column 1. The coefficients on expectations in columns 6 and 7 are moderately smaller than those in columns 2 and 3, and they are significant only at the 10 percent level. The coefficient on expectations in column 8 is small and not significant, similar to that in column 4. The R2s are much lower in columns 5-8 than in columns 1-4, indicating less predictive accuracy. It seems that predicting inflation got harder after 1989 for both consumers and economists. Overall, economists are slightly better at predicting inflation than consumers since 1990, especially over a one-year horizon.
In 2021Q2, the median of professional forecasters’ expectations for CPI inflation over the coming year is 2.4 percent. Using this survey expectation and the April 2021 12-month inflation rate in place of the four-quarter inflation rate for 2021Q2, column 1 of table 2 implies a prediction of inflation over the next four quarters of 2.1 percent. Under the same assumptions, column 2 implies an eight-quarter inflation rate (through 2023Q2) of 2.2 percent.
In 2021Q2, the median of professional forecasters’ expectations for CPI inflation over the next ten years is 2.3 percent. Column 3 of table 2 implies a prediction of inflation over the next five years of 2.0 percent, and column 4 implies inflation over the next ten years of 1.9 percent. Thus, as of April 2021, professional economic forecasters did not foresee any notable increase in future US inflation.
1. The American Rescue Plan, passed in March 2021, authorizes $1.9 trillion for public health measures and support to households, businesses, and states as the economy gradually emerges from the COVID-19 pandemic this year. If enacted, the other plans would spend several hundred billion dollars per year over the next five to ten years and raise taxes to cover most of the additional spending.
2. The consumer survey is conducted monthly. The figure displays quarterly averages of survey results and quarterly inflation to smooth through minor monthly variations.
4. These results are not sensitive to including a lagged change in gasoline prices, which is significantly correlated with consumer expectations of inflation.
5. None of the results in table 2 are sensitive to including lagged changes in gasoline prices.