Expected income growth and the great recession.
French, Eric ; Kelley, Taylor ; Qi, An 等
Introduction and summary
The Great Recession of 2008-09 was characterized by the most severe
year-over-year decline in consumption since 1945. The consumption slump
was both deep and long-lived. In this article, we document this decline
in aggregate consumption and look at an important determinant of
consumption: consumers' expectations about their future income. The
analysis uses microeconomic data from the Michigan Surveys of Consumers
(hereafter, Michigan Surveys) to study expected income growth. (1) These
data show that consumers' expected income growth declined
significantly during the Great Recession. It was the most severe drop in
income expectations ever observed in these data, and expectations have
not yet fully recovered to pre-recession levels.
The decline is widespread: It exists for all age groups, education
levels, and income quintiles. Furthermore, we show that expected income
growth is a strong predictor of actual future income and consumption
growth. For example, we show that expected income growth has
considerable added forecasting power above and beyond lagged consumption
and income growth, lagged Treasury bill rates, and lagged stock market
returns. For this reason, forecasts of near-term consumption and income
growth based on this series suggest sluggish income and consumption
growth over the next year. For example, forecasts based on expected
income growth suggest that consumption and income will likely grow about
1 percent next year, whereas forecasts that do not account for expected
income growth suggest that consumption and income will likely grow more
than 2 percent next year. Given the usefulness of these data for
forecasting, these data strongly suggest lackluster consumption and
income growth in the year ahead.
Usefulness of consumer expectations data for forecasting
There are at least two reasons why survey data on expected income
growth might be useful for predicting future income. First, people might
have some advance knowledge of their future economic circumstances that
is not available in other data. For example, they may know whether their
employer is about to reduce its work force. Second, expected income
growth might affect aggregate activity through a causal channel.
Pessimistic expectations about future income reduce current aggregate
demand and thus firms' demand for workers, actually causing a
reduction in labor income. In this sense, pessimistic income
expectations may create a self-fulfilling prophesy. (2)
Carroll, Fuhrer, and Wilcox (1994) study why consumer expectations
might be useful for explaining consumption growth. As they point out,
the simplest version of the permanent income hypothesis, the benchmark
model in consumption theory, suggests that consumer expectations should
not be useful for forecasting consumption growth. The reason for this is
that the permanent income hypothesis says that consumption should be
proportional to one's assets plus the present discounted value of
all expected labor income. High expected consumption growth should be
reflected in the current level of consumption. Consumption should change
over time only because of asset price changes or changes in the present
discounted value of future labor income. (3) Under certain departures
from the partial equilibrium permanent income hypothesis, future income
expectations may also affect the subsequent growth rate of consumption.
These departures include slow adjustment due to habit formation;
nonseparability between nondurable consumption and the consumption of
durables in the presence of adjustment costs; and the presence of
rule-of-thumb consumers (i.e., consumers who spend a fixed percentage of
their income), and/or liquidity-constrained consumers who respond to
income when it arrives, not when it becomes expected. In a general
equilibrium sense, income expectations are also determinants of the
growth rate of consumption in the extension of the permanent income
hypothesis to the case of variable interest rates (see, for example,
Hall, 1988), in which case consumption growth depends on the real
interest rate, which in turn depends on expected future income relative
to the present level. This paper does not identify the reasons why
expectation data help forecast future income and consumption. It simply
shows that they do.
Some previous studies have argued that data from the Michigan
Surveys are not very helpful for forecasting consumption (for example,
Leeper, 1992, and Ludvigson, 2004) beyond other commonly used variables.
For example, Ludvigson (2004) shows that once the econometrician
accounts for other variables--such as lagged consumption and income
growth, lagged Treasury bill rates, and lagged stock market returns--the
Index of Consumer Expectations and the Index of Consumer Sentiment (both
indexes created using variables from the Michigan Surveys) have little
added forecasting power. We find that the Index of Consumer Expectations
and the Index of Consumer Sentiment have modest added forecasting power,
and expected income growth has greater forecasting power, for
consumption after accounting for these other variables.
Consumer spending and the Great Recession
Figure 1 displays the level of real personal consumption
expenditures (PCE) from 1962 to 2012:Q4. Even over this long horizon,
the figure shows a distinct flattening out of the consumption growth
rate in 2008-09. The fact that this pattern is clearly visible even with
the perspective of a 50-year window highlights the severity and
persistence of the Great Recession and the very slow recovery that has
ensued. It took almost 12 quarters for total real PCE to go back to its
level at the previous peak in 2007:Q4.
Figure 2 compares the time path of real PCE over several
recessionary time periods. For each recession, the level of PCE is
normalized to 1 at the peak, as defined by the National Bureau of
Economic Research (NBER), prior to the recession. The NBER dates for the
recession peaks are 1973:Q4, 1980:Q1, 1981 :Q3, 1990:Q3, 2001:Q1, and
2007:Q4.
Figure 2 highlights that in the 2008-09 recession, consumption
dropped 3.1 percent and was slow to recover afterward. This pattern
contrasts with every recession since 1974. During all previous
recessionary periods, either consumption fell only modestly or it
increased following the peak. In this article, we consider consumer
expectations as a possible explanation for why the recovery in
consumption was so anemic.
Expected income in the Michigan Surveys of Consumers
We use data from the Michigan Surveys database, which is a
nationally representative probability sample of households. While the
surveys began in 1946, their current monthly format began in 1978. Since
the mid-1980s, about 500 households have been interviewed per month.
Those interviewed are asked a large number of questions about their own
current economic conditions and expectations of the future. Responses in
these surveys are used to construct the Index of Consumer Sentiment and
the Index of Consumer Expectations, both of which are widely followed in
both the business and financial communities. These indexes have also
been the subject of some academic research. (4) The questions underlying
these surveys are designed to capture the public's confidence in
the economy and are thus thought to be leading indicators of the
economy. An attractive aspect of these variables is that they are
publicly available and are released very quickly after the survey is
conducted.
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
[FIGURE 3 OMITTED]
In this article, we generate a measure of expected income growth
using the microdata from the Michigan Surveys database. To the best of
our knowledge, only De Nardi, French, and Benson (2012) have used this
measure in published research. The Michigan Surveys ask two questions to
identify consumers' expected income growth.
1) "During the next 12 months, do you expect your income to be
higher or lower than during the past year?"
2) "By about what percent do you expect your income to
(increase/decrease) during the next 12 months?"
The resulting index of expected income growth ranges between +95
and -95 and reflects the expected percentage change in nominal income in
the next year. Given the phrasing of the question, it is not clear if it
refers to labor income or total income. We assume that it refers to
disposable income, which is total income from all sources minus taxes.
Figure 3 compares realized and expected nominal disposable income and
shows that the two series track each other well, although nominal
disposable income is more volatile. For example, consumers'
expected income growth started its decline in 2007, well before the fall
in disposable income.
The surveys also ask about expected changes in the price level over
the next 12 months. This number is historically very similar to realized
Consumer Price Index (CPI) inflation. These two inflation series have
diverged in the past, but since the late 1970s the differences between
them have been minor. At the start of the Great Recession, however, a
large gap opened up, which makes for the largest discrepancy between
these two data series. In the second quarter of 2008, expected inflation
was reported at +6 percent, compared with -1 percent actual CPI
inflation. The two measures have since moved closer together (see figure
4). Clearly, the gap in these two measures affects measured real income
growth expectations, as we document in the next section.
Income growth expectations
Figure 5 shows nominal expected income growth around different
recessions. The time period "0" represents the NBER peak of
economic activity, "-4" represents expected income growth four
quarters before the peak, and "4" represents expected income
growth four quarters after the peak. The figure shows that expected
income growth was much lower around the Great Recession than around
previous recessions.
Nominal income growth during the Great Recession was low, but
inflation was also low. To study the behavior of real income
expectations, we must measure inflation expectations, which we do in two
ways. First, we use actual CPI inflation over the 12-month period
covered by the survey question, which assumes that consumers have
perfect foresight for next year's inflation. Second, we use the
answer to the survey question about the individual's expectations
about growth in prices over the next 12 months. Using these two
measures, we construct individual-level expected real income growth and
then aggregate up to population means by quarter. We construct expected
real income growth by subtracting each individual's inflation
expectations from his expected nominal income growth. The data begin in
1978 and go through 2012:Q4.
[FIGURE 4 OMITTED]
[FIGURE 5 OMITTED]
[FIGURE 6 OMITTED]
In figure 6, there is no clear cyclical pattern prior to the Great
Recession in real income expectations when deflating by CPI inflation.
Before the most recent recession, real income growth was rather flat;
several quarters before the peak, it dropped into negative territory;
and four quarters after the peak, it went up to about 4 percent. After
that point, however, it recorded a large drop, reaching -3 percent five
quarters after the peak. In summary, for the most recent recession, real
income growth expectations deflated by CPI show a deterioration and
lower average growth than during previous recessions.
Figure 7 shows that consumers' perceived real income growth
using consumers' inflation expectations provides an even more
pessimistic outlook about purchasing power during the Great Recession.
Consumers' perceived real income growth was dipping in and out of
negative territory well before the recession started; and it sustained a
large drop starting four quarters before the peak. This drop was
abnormal, both in terms of size and duration. The recovery in expected
income growth also stands out historically, in terms of its length and
sluggishness. Even 21 quarters from the peak, expected income growth is
still well below its pre-recession levels.
An attractive feature of the Michigan data is that they allow us to
examine responses by age, education, and income. Figure 8 shows that
after the late 1970s, younger individuals began to expect higher income
growth than their older counterparts around all recessions. All age
groups expect roughly equal declines in income during recessions.
Relative to previous recessions, all age groups expected greater
declines in income during the Great Recession.
Figure 9 shows that around all previous recessions, people with
higher levels of education (some college and above) have expected faster
income growth than people with lower levels of education (high school
and below). It also shows that during the Great Recession, the college
educated were more pessimistic relative to previous recessions. Although
the college educated still expect faster income growth than the less
educated, the differences are less stark than in the past. This is
consistent with Petev, Pistaferri, and Eksten's (2010) findings.
First, they find that increased government transfers supported income
among the poorest-income households during the Great Recession. (5)
Second, using the Consumer Expenditure Survey, they find that survey
respondents in the top decile of the wealth distribution (who are mostly
college educated) are the ones who decreased spending the most during
the Great Recession (which is also consistent with the findings of Meyer
and Sullivan, 2013). This finding also holds for the subcategories of
nondurables and services. This drop in consumption might be due to the
large negative wealth effect experienced by these households as a result
of the decrease in house values and stock market valuation.
[FIGURE 7 OMITTED]
Figure 10 shows that the Great Recession reduced the expected
income growth of all income quintiles.
Our main findings from the analysis of the microdata are as
follows. First, expected nominal income growth declined significantly
during the Great Recession. It is the worst drop ever observed in these
data, and it has not yet recovered to pre-recession levels. Second,
expectations for real income growth have also declined, and the decline
in expected real income growth is more severe when personal inflation
expectations are used instead of actual CPI inflation. Third, the
decline exists for all age groups, education levels, and income
quintiles. Relative to previous recessions, the latest recession found
those with higher levels of income and education feeling more
pessimistic about their prospects than their poorer and less-educated
counterparts.
Do the Michigan Surveys have predictive power for future income and
consumption growth?
Next, we explore the predictive power of the Michigan Surveys data
for future disposable income and consumption growth. (6) We estimate the
regression for disposable income first:
([Y.sub.t+k+4] - [Y.sub.t+k]/[Y.sub.t+k]) = [[alpha].sub.0] +
[[alpha].sub.1] ([Y.sub.1] - [Y.sub.t-4]/[Y.sub.t-4]) +
[[alpha].sub.2][g.sub.Mt] + [[epsilon].sub.t+k],
where [[alpha].sub.0], [[alpha].sub.1], [[alpha].sub.2] are
parameters to estimate and [[alpha].sub.1] and [[alpha].sub.2] are
reported in table 1. The variable
([Y.sub.t+k+4] - [Y.sub.t+k]/[Y.sub.t+k])
is next year's annual income growth k quarters from now, so k
is 0 when forecasting income growth over the next year and 4 when
forecasting income growth over the subsequent year. The variable
([Y.sub.t] - [Y.sub.t-4]/[Y.sub.t-4])
is income growth over the past year, and [g.sub.Mt], is expected
real income growth deflated using expected inflation, both from the
Michigan Surveys. As our results show, deflating by expected inflation
produces the most accurate forecasts of future income and consumption
growth.
[FIGURE 8 OMITTED]
As shown in table 1, lagged actual income growth has a negative
coefficient, while expected income growth has a positive coefficient.
For income growth over the next year, the coefficient on expected income
growth is 0.74, indicating that a 1 percent decline in expected income
growth reduces next year's income growth 0.74 percent, controlling
for the past year's income growth. The third column shows that
predicted income growth over the next year (2012:Q4 to 2013:Q4), using
lagged income growth and expected income growth, is 0.5 percent, well
below its average of 2.8 percent over the 1978-2012 sample period.
Income growth between 2013:Q4 and 2014:Q4 is also forecasted to be low.
Expected income growth is also a good predictor of consumption
growth. Table 1 also presents regressions using future consumption
growth as the left-hand-side variable and lagged income growth and the
Michigan expectations variable as the right-hand-side variables. The
consumption forecast for 2012:Q4 to 2013:Q4 is for 0.9 percent growth.
[FIGURE 9 OMITTED]
Table 2 shows forecasts of alternative measures of consumer
expectations. In this table, we focus on the predictive power of
different measures of consumer expectations for forecasting consumption
growth one year ahead. Table 1 showed that our benchmark expectations
measure is not particularly good for forecasting beyond one year, as
measured by the R-squared statistic. Furthermore, the consumer
expectations variables that predict consumption well also predict income
well. Thus, for brevity, we do not report the results for income and
just focus on consumption.
The first row of table 2 shows results for consumption growth using
the benchmark expectations measure that we also used in table 1. The
second row deflates expected income growth by CPI inflation over the
past year. This measure produces a consumption forecast of 2.5 percent
growth for the four quarters ending in 2013:Q4. This higher forecast is
not surprising given that, when deflating by CPI (as in figure 6) rather
than expected inflation (as in figure 7), expected income growth over
the next year is much stronger. However, the R-squared statistic shows
that deflating by lagged CPI inflation yields a lower R-squared, meaning
that it is somewhat less useful for predicting future consumption
growth. For this reason, we prefer deflating by consumers' expected
inflation.
[FIGURE 10 OMITTED]
In the third row, we take median expected nominal income growth
minus median expected inflation. This measure has the attractive feature
that these data are publicly available. Using this measure produces a
forecast of 1.4 percent, modestly higher than our benchmark estimate.
However, the R-squared statistic suggests that this model fits the data
worse than the benchmark model.
The fourth row takes the median of real (that is, deflated using
inflation expectations) expected income growth as the expected income
growth measure. The forecast from this model is similar to the benchmark
specification and fits the data about as well as the benchmark.
The next three rows use measures of consumer confidence from the
same survey. Row 5 uses the log of the Index of Consumer Expectations.
Row 6 uses the Index of Consumer Sentiment. Row 7 uses the change in the
log of consumer expectations. Row 8 uses the log of one of the
subcomponents of the Index of Consumer Expectations, the measure of
whether the next five years will be good economic times. (7) All of the
consumer expectations and consumer sentiment measures suggest relatively
stronger consumption growth ahead than is suggested by the expected
income growth measures. However, these measures have less predictive
power than the expected income growth measure.
Row 9 shows that lagged income growth has modest predictive power
for consumption growth, and row 10 shows that expected income growth
alone has very good predictive power.
Comparing the R-squared statistics in row 1 and row 10 indicates
that virtually all of the forecasting power from lagged income growth
and expected income growth combined is coming from expected income
growth alone.
When including multiple consumer expectations measures in the same
regression, our preferred measure (expected income growth deflated using
expected inflation) is still positive and statistically significant.
Furthermore, forecasts based on regressions that include our preferred
measure always yield a forecast of low consumption growth for next year.
For example, when regressing consumption growth on our preferred
measure, as well as lagged income growth and the log Index of Consumer
Expectations, our forecast for next year's consumption growth is
1.2 percent.
Previous research suggests that consumer expectations are not very
helpful for forecasting consumption growth after conditioning on other
variables, such as multiple lags of consumption, income, the Treasury
bill rate, and stock price growth (for example, Ludvigson 2004). In
table 3, we assess whether expected income growth is still useful for
forecasting after accounting for these other variables. In order to
assess the forecasting power of our measures, we begin by presenting
results from a regression of annual consumption growth using four
quarterly lags of consumption growth, income growth, the Treasury bill
rate, and growth in the S&P 500 stock market index, as well as a
single expectations measure. The table reports the sum of the four
coefficients (or the one coefficient in the case of the expectations
measure) and the p-values for the joint marginal significance of the
lags of each variable. We report the R-squared statistic and also the
adjusted R-squared statistic. The adjusted R-squared statistic penalizes
the R-squared statistic for the total number of parameters using in the
regression. Since adding parameters mechanically increases the R-squared
statistic, even though it might not improve the ability of the model to
forecast out of sample, the adjusted R-squared is thought to give a
better sense of the model's ability to forecast out of sample.
The top row of table 3 shows that when using no expectations
measure, lagged consumption, income, the Treasury bill rate, and the
stock market can explain 40 percent of the variance of one-year-ahead
consumption growth. Lagged consumption, stock prices, and Treasury bill
rates all turn out to be statistically significant. Because of the
number of parameters, the adjusted R-squared measure is somewhat lower
though, at 0.32. Forecasted consumption growth from 2012:Q4 to 2013:Q4
is 2.2 percent.
The next row includes our preferred expected income growth measure.
When we add this measure, only the stock market measures and expected
income growth remain statistically significant. For example, the
coefficient on the expectations measure is 0.46 with a p-value of 0.02,
meaning that if the expectations measure has no predictive power, there
is only a 2 percent chance that we would have estimated such a large
coefficient in our sample. The adjusted R-squared rises from 0.31 to
0.39, showing again that the measure does improve the fit of the model.
Furthermore, the forecast of next year's consumption growth, 1.3
percent, is still very low.
The next two rows use the Index of Consumer Expectations and the
Index of Consumer Sentiment. Consistent with Ludvigson (2004), we show
that these measures add modest predictive power relative to lagged
consumption, income, Treasury bill rates, and stock market performance.
The bottom row shows that, relative to the case of no expectations
variable, adding the log of the Index of Consumer Sentiment barely
increases the adjusted R-squared.
Table 4 provides more evidence of the forecasting power of the
expectations variables. It shows the root mean squared forecast error
(RMSE) and U-statistic associated with each of the four different
forecasting models shown in table 3, plus four of the different
forecasting models in table 2. (8) A U-statistic of below 1 suggests
that forecasts from the estimated model have a lower RMSE than would
result from using a random walk as the forecast of consumption growth.
(9) The U-statistic for row 1 shows that over the 1992:Q4-2012:Q4 and
2002:Q4-2012:Q4 periods, the regression model that uses four lags of
consumption, income, Treasury bill, and stock market information does
worse than just using last year's consumption growth as a forecast.
Furthermore, rows 2-4 indicate that adding in any of the expectations
measures, if anything, increases the RMSE and the U-statistic. Models
with many parameters frequently have poor out-of-sample performance.
Table 4 provides another example of this. Row 5 considers a simple model
that uses last year's income growth to predict next year's
consumption growth. The U-statistic for this model is also above 1.
However, rows 6-8 show that models that use just income growth and
consumer expectations measures usually have U-statistics below 1,
suggesting that these models are better for forecasting consumption
growth than just using last year's consumption growth. Finally, row
9 shows that using only expected income growth performs as well as any
of the models we consider, and always seems to perform better than a
random walk.
Taken together, tables 2-4 show that real income expectations are
useful for forecasting consumption growth, both within sample (as shown
using the R-squared statistic) and out of sample (as shown using the
U-statistic).
Table 5 decomposes the total consumption growth regression
estimates in table 1 into durables and nondurables. Table 5 shows that a
1 percent increase in expected income growth increases next year's
durables spending by 1.47 percent and nondurables and services spending
by 0.58 percent. (10) Despite the larger coefficient on durables, the
R-squared statistic shows that we are less able to predict the
variability in durables spending than in nondurables and services
spending. Lagged income and expected income growth can jointly explain
44 percent in the variability in nondurables and services spending
growth but only 12 percent of the variability of durables spending
growth. Table 5 shows that the two measures have virtually no predictive
power for durables spending growth beyond one year, although the
measures have some predictive power for nondurables and services
spending growth. Table 5 also shows that expected income growth is much
more useful for forecasting nondurables and services spending growth
than the Index of Consumer Expectations, at least as measured by the
R-squared statistic.
In short, the low expected income growth in the Michigan Surveys
data suggests that the U.S. will experience low income and consumption
growth over the next two years.
Conclusion
This article documents the decline in aggregate consumption during
and after the Great Recession. It also explores the relationship between
the decline in consumption and the decline in consumers'
expectations about their future income. The analysis uses microeconomic
data from the Michigan Surveys of Consumers to study expected income
growth. These data show that expected income growth declined
significantly during the Great Recession for all age, income, and
education groups. It is the worst drop ever observed in these data, and
it has not yet fully recovered to pre-recession levels. Furthermore, we
show that expected income growth is a strong predictor of actual future
income and consumption growth. For this reason, forecasts of near-term
consumption and income growth using these data suggest sluggish income
and consumption growth over the next year.
Policymakers are still debating which actions, if any, should be
taken to stimulate the economy. Although this article does not give any
clear direction on the path that should be taken, the results discussed
here suggest that actions undertaken to stimulate the economy are
unlikely to lead to an overheated, high-inflation economy in the near
future.
REFERENCES
Barsky, Robert B., and Eric R. Sims, 2009, "Information,
animal spirits, and the meaning of innovations in consumer
confidence," National Bureau of Economic Research, working paper,
No. 15049, June.
Carroll, Christopher D., Jeffrey C. Fuhrer, and David W. Wilcox,
1994, "Does consumer sentiment forecast household spending? If so,
why?," American Economic Review, Vol. 84, No. 5, December, pp.
1397-1408.
De Nardi, Mariacristina, Eric French, and David Benson, 2012,
"Consumption and the Great Recession," Economic Perspectives,
Federal Reserve Bank of Chicago, Vol. 36, First Quarter, pp. 1-16,
available at www.chicagofed.org/digital_assets/publications/
economic_perspectives/2012/1Q2012_part1_denardi_ french_benson.pdf.
Flavin, Marjorie A., 1981, "The adjustment of consumption to
changing expectations about future income," Journal of Political
Economy, Vol. 89, No. 5, October, pp. 974-1009.
Hall, Robert E., 1988, "Intertemporal substitution in
consumption," Journal of Political Economy, Vol. 96, No. 2, April,
pp. 339-357.
Leeper, Eric M., 1992, "Consumer attitudes: King for a
day," Economic Review, Federal Reserve Bank of Atlanta, Vol. 77,
No. 4, July, pp. 1-15.
Ludvigson, Sydney C., 2004, "Consumer confidence and consumer
spending," Journal of Economic Perspectives, Vol. 18, No. 2,
Spring, pp. 29-50.
Meyer, Bruce, and James Sullivan, 2013, "Consumption
inequality and the Great Recession," American Economic Review,
forthcoming.
Petev, Ivalyo, Luigi Pistaferri, and Itay Saporta Eksten, 2010,
"Consumption and the Great Recession: An analysis of trends,
perceptions, and distributional effects," Stanford University,
mimeo.
Shapiro, Matthew D., 2010, "The effects of the financial
crisis on the well-being of older Americans: Evidence from the cognitive
economic study," University of Michigan, Michigan Retirement
Research Center, working paper, No. WP 2010-228, September.
Souleles, Nicholas S., 2004, "Expectations, heterogeneous
forecast errors, and consumption: Micro evidence from the Michigan
Consumer Sentiment Surveys," Journal of Money, Credit and Banking,
Vol. 36, No. 1, February, pp. 39-72.
NOTES
(1) The official name is the Thomson Reuters/University of Michigan
Surveys of Consumers; see http://thomsonreuters.com/products_
services/financial/financial_products/a-z/umichigan_surveys_
of_consumers/.
(2) Barsky and Sims (2009) suggest that the main reason questions
in the Michigan Surveys are useful for predicting future income growth
is that individuals have some information about their future financial
circumstances, not that their optimism causes fluctuations in future
output.
(3) The simple partial equilibrium permanent income hypothesis
implies that, conditional on initial assets, consumption is roughly
proportional to the capitalized value of expected future labor income.
Denoting [c.sub.t] as consumption at time t, r as the interest rate,
[W.sub.t] as assets at time t, [Y.sub.t+i] as nonasset income at time t
+ i, [beta] as a variable that accounts for an individual's
discounting of the future, and [E.sub.t] as an individual's
expectations of future variables, Flavin (1981) shows that [c.sub.t] =
[beta][r[W.sub.t] + (r/1(l + r))[E.sub.t][[infinity].summation over
(i=0)][(1 + r).sup.-1] [Y.sub.t+i]]. Dividing both sides by current
income y, and log-linearizing, we see that the log consumption/income
ratio is (approximately) a present value of all future expected income
growth rates. Thus, in this benchmark theory, expected future income
matters because (conditional on current income) it is the key
determinant of the current level of consumption.
(4) Examples include Leeper (1992), Carroll, Fuhrer, and Wilcox
(1994), Souleles (2004), Ludvigson (2004), and Barsky and Sims (2009).
(5) As a possible explanation for the pessimism of the wealthy,
Shapiro (2010) finds that these households were exposed more to the
stock market and experienced larger declines in wealth as a consequence.
The median decline in wealth was 15 percent in Shapiro's data, and
those who lost at least 10 percent of their net worth had almost twice
the mean wealth and 3.5 times the median wealth of the sample.
(6) See Leeper (1992), Carroll, Fuhrer, and Wilcox (1994), Souleles
(2004), Ludvigson (2004), Barsky and Sims (2009), and De Nardi, French,
and Benson (2012) for more on the predictive power of the Michigan
Surveys.
(7) It uses the response to "Looking ahead, which would you
say is more likely--that in the country as a whole we'll have
continuous good times during the next five years or so, or that we will
have periods of widespread unemployment or depression, or what?"
(8) The RMSE for the 1992:Q4-2012:Q4 period is calculated by first
estimating the regression model using data up to 1991:Q4. Next, we use
the estimated coefficients to forecast consumption growth over the
1991:Q4-1992:Q4 period. We then subtract from this realized consumption
growth over the 1991:Q4-1992:Q4 period and square the value. We repeat
this for 1992:Q 1-1993:Q1,..., 2011:Q4-2012:Q4. We then take the square
root of the mean of the squared errors over the forecasting period,
which is our measure of RMSE. The U-statistic divides RMSE by the RMSE
of a model that uses last year's consumption growth as a forecast
of this year's consumption growth.
(9) The random walk forecast uses last year's consumption
growth as a forecast of this year's consumption growth
(10) We also tried regressing nondurables and services spending
growth on expected income growth separately. The estimates for both
subcomponents were very similar to each other so we do not report them.
For example, the coefficient on expected income growth is 0.62 in the
one-year-ahead nondurables equation and 0.57 in the one-year-ahead
services equation (when focusing on growth over the next year), versus
0.58 for nondurables and services together.
Eric French is a senior economist and research advisor, Taylor
Kelley is an associate economist, and An Qi is an associate economist in
the Economic Research Department of the Federal Reserve Bank of Chicago.
The authors thank Dick Porter and Robert Barsky for helpful comments.
TABLE 1
Income and consumption growth regression results
Lagged income Michigan Surveys
Dependent variable growth variable income expectations
Annual income growth
One year ahead -0.35 0.74
(0.08) (0.17)
Two years ahead 0.03 0.36
(0.09) (0.14)
Three years ahead -0.27 0.47
-0.28 (0.21)
Annual consumption growth -0.29
One year ahead -0.30 0.71
-0.31 (0.20)
Two years ahead -0.32 0.54
-0.33 (0.17)
Three years ahead -0.34 0.33
-0.35 (0.22)
Forecasted annual
Dependent variable growth (%), Q4/Q4 R-squared
Annual income growth
One year ahead 0.51
(2012:04 to 2013:04) 0.30
Two years ahead 1.70
(2013:04 to 2014:04) 0.10
Three years ahead 1.28
(2014:04 to 2015:04) 0.09
Annual consumption growth
One year ahead 0.87
(2012:04 to 2013:04) 0.34
Two years ahead 1.26
(2013:04 to 2014:04) 0.14
Three years ahead 1.91
(2014:04 to 2015:04) 0.05
Notes: The regressions were run with data from 1978:01 to 2012:04.
Newey-West standard errors are in parentheses. Average annual income
and consumption growth are 2.78 percent and 2.91 percent,
respectively.
Sources: Authors' calculations based on data from Haver Analytics and
Thomson Reuters/University of Michigan Surveys of Consumers.
TABLE 2
Alternative expectations measures, consumption growth, 2012:04 to
2013:04
Lagged income Michigan Surveys
growth expectations
Expectations measures variable measure
1) Mean of nominal income growth -0.08 0.71
deflated by inflation expectations (0.12) (0.20)
2) Mean of nominal income growth 0.05 0.43
deflated by lagged CPI inflation (0.08) (0.12)
3) Median expected nominal income 0.03 0.79
growth minus median expected (0.09) (0.21)
inflation (a)
4) Median of expected income 0.03 1.24
growth deflated by inflation (0.10) (0.40)
expectations (b)
5) Log of Index of Consumer -0.01 5.43
Expectations (0.09) (1.53)
6) Log of Index of Consumer 0.01 5.49
Sentiment (0.11) (2.07)
7) Change in log of Index of 0.29 4.97
Consumer Expectations (0.10) (2.39)
8) Log of expectations five 0.09 3.15
years ahead (0.07) (1.01)
9) No expectations measure 0.28
(0.09)
10) Mean of nominal income growth 0.66
deflated by inflation expectations, (0.17)
no lagged income variable
Forecasted
annual growth
Expectations measures (%), Q4/04 R-squared
1) Mean of nominal income growth 0.87 0.34
deflated by inflation expectations
2) Mean of nominal income growth 2.51 0.29
deflated by lagged CPI inflation
3) Median expected nominal income 1.41 0.31
growth minus median expected
inflation (a)
4) Median of expected income 0.81 0.34
growth deflated by inflation
expectations (b)
5) Log of Index of Consumer 2.62 0.28
Expectations
6) Log of Index of Consumer 2.53 0.23
Sentiment
7) Change in log of Index of 3.40 0.13
Consumer Expectations
8) Log of expectations five 3.10 0.20
years ahead
9) No expectations measure 3.00 0.07
10) Mean of nominal income growth 1.04 0.34
deflated by inflation expectations,
no lagged income variable
(a) Takes difference between median of expected income growth and
median of expected inflation, quarter by quarter.
(b) Deflates expected income growth by expected inflation for every
member of the sample, then takes the median over all members
interviewed that quarter.
Notes: The regressions were run with data from 1978:01 to 2012:04.
Newey-West standard errors are in parentheses.
Sources: Authors' calculations based on data from Haver Analytics and
Thomson Reuters/University of Michigan Surveys of Consumers.
TABLE 3
Annual consumption growth forecasts with alternative expectations and
additional quarterly covariates
Four lags of Four lags Four lags
Expectations measure consumption of income of S&P 500
No expectations measure 2.16 -0.19 0.05
(0.01) (0.67) (0.06)
Mean of nominal income growth 0.99 -0.35 0.08
deflated by inflation (0.45) (0.70) (0.01)
expectations
Log of Index of Consumer 1.55 -0.60 0.05
Expectations (0.12) (0.73) (0.17)
Log of Index of Consumer 1.68 -0.54 0.05
Sentiment (0.13) (0.66) (0.12)
Four lags Michigan Surveys
of Treasury expectations
Expectations measure bill rate measure
No expectations measure -1.09
(0.01)
Mean of nominal income growth -0.85 0.46
deflated by inflation (0.16) (0.02)
expectations
Log of Index of Consumer -0.92 3.15
Expectations (0.02) (0.03)
Log of Index of Consumer -1.01 2.74
Sentiment (0.01) (0.12)
Forecast (%),
Adjusted 2012:Q4--
Expectations measure R-squared R-squared 2013:04:00
No expectations measure 0.39 0.31 2.18
Mean of nominal income growth 0.47 0.39 1.30
deflated by inflation
expectations
Log of Index of Consumer 0.43 0.36 2.30
Expectations
Log of Index of Consumer 0.41 0.32 2.21
Sentiment
Notes: The table reports the sum of the coefficients on the quarterly
lags of the variable indicated; the probability that the variable can
be excluded from the prediction equation appears in parentheses.
Hypothesis tests were conducted using a heteroskedasticity and serial
correlation robust covariance matrix.
Sources: Authors' calculations based on data from Haver Analytics and
Thomson Reuters/University of Michigan Surveys of Consumers.
TABLE 4
One-year-ahead consumption growth forecasts with alternative
expectations, U-statistics, and RMSE
1992:04-2012:04
Covariates used RMSE U-statistic
1) Table 3 covariates,
no expectations variable 1.78 1.08
2) Table 3 covariates and
real income expectations 1.81 1.10
3) Table 3 covariates and log of
Index of Consumer Expectations 1.78 1.09
4) Table 3 covariates and log of
Index of Consumer Sentiment 1.78 1.08
5) Lagged income growth,
no expectations variable 1.78 1.09
6) Lagged income growth and
real income expectations 1.53 0.93
7) Lagged income growth and log of
Index of Consumer Expectations 1.53 0.93
8) Lagged income growth and log of
Index of Consumer Sentiment 1.58 0.96
9) Real income expectations,
no other covariates 1.51 0.92
2002:04-2012:04
Covariates used RMSE U-statistic
1) Table 3 covariates,
no expectations variable 2.12 1.11
2) Table 3 covariates and
real income expectations 2.18 1.14
3) Table 3 covariates and log of
Index of Consumer Expectations 2.11 1.10
4) Table 3 covariates and log of
Index of Consumer Sentiment 2.11 1.11
5) Lagged income growth,
no expectations variable 2.14 1.12
6) Lagged income growth and
real income expectations 1.93 1.01
7) Lagged income growth and log of
Index of Consumer Expectations 1.83 0.96
8) Lagged income growth and log of
Index of Consumer Sentiment 1.90 0.99
9) Real income expectations,
no other covariates 1.90 0.99
Notes: Table 3 covariates include four quarterly lags of consumption,
labor income, S&P 500 index growth, and Treasury rates. Lagged income
growth is annual disposable income growth. U-statistic = [square root
of [[[summation].sup.T.sub.t=1][([??].sub.t+4] -
[C.sub.t+4]).sup.2]]/[[summation].sup.T.sub.t=1][([C.sub.t+4] -
[C.sub.t]).sup.2]], where [[??].sub.t+4] is the model's forecast for
next year's annual consumption growth using data through time t,
[c.sub.t+4] is the realized value for next year's annual consumption
growth, and [c.sub.4] is last year's consumption growth. Root mean
squared forecast error (RMSE) = [square of
[1/T][[summation].sup.T.sub.t-1][([[??].sub.t-4] - [c.sub.t-4]).sup.2].
Sources: Authors' calculations based on data from Haver Analytics and
Thomson Reuters/University of Michigan Surveys of Consumers.
TABLE 5
Consumption subcomponents growth regression results
Lagged income Expectations
Dependent variable growth variable variable
Expectations variable is
expected real income growth
Annual consumption growth
One year ahead -0.08 0.71 ***
(0.12) (0.20)
Two years ahead -0.13 0.54 ***
(0.13) (0.17)
Three years ahead -0.21 0.33
(0.13) (0.22)
Annual durables growth
One year ahead -0.35 1.47 **
(0.44) (0.73)
Two years ahead -0.28 0.68
(0.48) (0.57)
Three years ahead -0.64 * 0.19
(0.32) (0.61)
Annual nondurables and services growth
One year ahead -0.04 0.58 ***
(0.09) (0.15)
Two years ahead -0.11 0.50 ***
(0.08) (0.13)
Three years ahead -0.15 0.33 *
(0.12) (0.18)
Expectations variable is log
of Consumer Expectations Index
Annual consumption growth
One year ahead -0.01 5.43 ***
(0.09) (1.53)
Two years ahead 0.01 2.17
(0.13) (1.85)
Three years ahead -0.10 0.79
(0.13) (1.62)
Annual durables growth
One year ahead -0.41 14.89 ***
(0.35) (5.41)
Two years ahead -0.17 4.06
(0.51) (6.87)
Three years ahead -0.63 ** 1.56
(0.31) (4.89)
Annual nondurables and services growth
One year ahead 0.05 3.78 ***
(0.07) (1.26)
Two years ahead 0.04 1.69
(0.08) (1.24)
Three years ahead -0.03 0.52
(0.12) (1.29)
Forecasted annual
Dependent variable growth (%), Q4/Q4 R-squared
Expectations variable is
expected real income growth
Annual consumption growth
One year ahead 0.87 0.34
Two years ahead 1.26 0.14
Three years ahead 1.91 0.05
Annual durables growth
One year ahead 0.87 0.12
Two years ahead 3.19 0.02
Three years ahead 4.89 0.03
Annual nondurables and services growth
One year ahead 0.95 0.44
Two years ahead 1.08 0.22
Three years ahead 1.59 0.08
Expectations variable is log
of Consumer Expectations Index
Annual consumption growth
One year ahead 2.62 0.28
Two years ahead 2.78 0.05
Three years ahead 2.90 0.01
Annual durables growth
One year ahead 4.26 0.15
Two years ahead 5.00 0.01
Three years ahead 5.36 0.03
Annual nondurables and services growth
One year ahead 2.45 0.29
Two years ahead 2.51 0.06
Three years ahead 2.59 0.00
* Significant at the 90 percent level.
** Significant at the 95 percent level.
*** Significant at the 99 percent level.
Notes: See table 1 notes. Expected real income was deflated using
consumers' expected inflation. Average annual durables and nondurables
plus services growth are 5.1 percent and 2.6 percent, respectively.
Sources: Authors' calculations based on data from Haver Analytics and
Thomson Reuters/University of Michigan Surveys of Consumers.
