Can consumer attitudes forecast household spending in the United States? Further evidence from the Michigan survey of consumers.

1. Introduction

For many years, indices of consumer sentiment have been used to provide government policy makers, economic forecasters, and business managers with timely and important information on consumer attitudes. This interest in consumer attitudes reflects a widespread belief that the sentiments and expectations of individual consumers directly affect the direction of the U.S. economy. Reinforcing this belief is the fact that consumer spending accounts for about two thirds of the nation's Gross Domestic Product (GDP).

Thus far, most analyses of consumer attitudes as a leading indicator of household spending have focused primarily on the predictive power of the Michigan Index of Consumer Sentiment (ICS). The results of these studies have, however, been mixed. For example, an early study by Lovell (1975) finds that measures of consumer attitudes based on the Michigan Survey of Consumers are unreliable predictors of future consumption. (1) Mishkin (1978), using a stock adjustment model, shows that the ICS provides good explanatory power for changes in consumer durables. Carroll, Fuhrer, and Wilcox (1994, henceforth CFW) find that the Michigan Index has some incremental predictive power as regards forecasting household spending. Souleles (2001), using the microdata of the Michigan Survey, reports that consumer sentiment is useful in forecasting future consumption, even when controlling for a number of macroeconomic variables. On the other hand, Howrey (2001) finds that both lagged and current-quarter monthly values of the ICS are generally insignificant when control variables are present in the equations of total personal consumption expenditures (PCE), consumer spending on durable goods as well as on services.

Lovell (2001) recently suggests that the Index of Consumer Expectations (ICE) developed by the University of Michigan may be a better proxy for consumer confidence than the ICS. This is because the ICE is derived solely from a subset of forward-looking questions, in contrast to the ICS, which is based on both forward-looking questions and current-conditions questions. (2) In view of Lovell's (2001) insightful suggestion, the main objective of this article is to empirically examine the predictive power of the ICE in forecasting U.S. consumption growth. Moreover, it would he useful to compare the informational content of the ICE and the ICS to determine whether indices of consumer confidence reflect consumers' perception of future economic conditions.

In this article, we use the reduced-form equation given in CFW (1994) to examine the forecasting ability of the ICE and the ICS. Our empirical results indicate that the lagged values of the ICE predict changes in total PCE much better than those of the ICS. Furthermore, when tested separately in the reduced-form equation, the forward-looking questions are generally significant, suggesting that they contain valuable information about consumers' expectations of future economic outlook. We also extend our analysis to the study by CFW (1994). The results of this analysis confirm the view that the ICE has greater incremental predictive power than the ICS.

The remainder of this article is structured as follows: Section 2 describes the five core questions used in the Michigan Surveys of Consumers. Section 3 discusses the econometric methodology and data. Section 4 reports our main empirical results. Section 5 is a case study based on CFW's (1994) data set. Section 6 presents some conclusions.

2. The Michigan Surveys of Consumers

The ICS, produced by the Michigan Surveys of Consumers, is derived from the following five core questions: (3)

QF[P.sup.r] (Financial Position realization). We are interested in how people are getting along financially these days. Would you say that you (and your family living there) are better off or worse off financially than you were a year ago?

QDurs (Durables purchases). About the big things people buy for their homes such as furniture & refrigerator, stove, television, and things like that. Generally speaking, do you think now is a good or bad time for people to buy major household items?

QF[P.sup.e] (Financial Position expectation). Now looking ahead--do you think that a year from now you (and your family living there) will be better off financially or worse off, or just about the same as now?

QBCm12 (Business conditions, 12 months). Now turning to business conditions in the country as a whole--do you think that the next twelve months we'll have good times financially, or bad times, or what?

QBCy5 (Business conditions, 5 years). 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 5 years or so, or that we will have periods of widespread unemployment or depression, or what?

The ICS is computed using the relative scores (the percentage giving favorable replies minus the percentage giving unfavorable replies, plus 100) for each of the five core survey questions.

It is important to note that the five core questions vary in nature. Two of the five questions, QF[P.sup.r] and QDurs, ask respondents how they view current economic conditions, whereas QBCm12, QBCy5, and QF[P.sup.e] ask them how they view future business conditions--both during a 1-year and a 5-year period--as well as changes in their own financial situation during the following year.

In order to further gauge consumer confidence, the University of Michigan also designed the ICE using only the relative scores of the three forward-looking questions QBCm12, QBCy5, and QF[P.sup.e]. Because of its usefulness in predicting the future course of the U.S. economy, the ICE is included in the Leading Indicator Composite Index developed by the U.S. Department of Commerce.

3. Econometric Methodology and Data

We examine the predictive ability of various measures of consumer confidence by using the reduced-form equation given in CFW (1994, p. 1400):

(1) [DELTA] log([C.sub.t]) = [[alpha].sub.0] + [N.summation over i=1] [[beta].sub.i][S.sub.t-i] + [gamma][Z.sub.t-1] + [[epsilon].sub.t],

where [C.sub.t] is consumer spending, [S.sub.t] is consumer confidence, [Z.sub.t] is a vector of control variables, and [[epsilon].sub.t] is a non-autocorrelated error term. Following CFW (1994), we include the following control variables in Z: four lags of the dependent variable (past consumption growth) and four lags of the growth in real labor income. In this article, we use three proxies for [S.sub.t]: the ICS, the ICE, and the five individual survey questions (QF[P.sup.r], QDurs, QBCm12, QBCy5, and QF[P.sup.e]). We also consider four categories of consumption [C.sub.t]: total personal consumption expenditures (PCE), durable goods, nondurable goods, and services. Data for labor income and the four categories of consumption are obtained from the Web site of the Bureau of Economic Analysis (BEA). The other data series that include the ICE, the ICS, and the relative scores for the five survey questions are provided by the Survey Research Center of the University of Michigan.

Our empirical analysis has been carried out using quarterly data for the sample period 1960Q1-2002Q2. (4) The choice of the starting date of our sample period is constrained by data availability. This is because, prior to 1960, the Michigan Surveys of Consumers was conducted only 2 or 3 times a year. (5) More importantly, some of the original data on the pre-1960 ICE are no longer kept by the University of Michigan. (6) Consequently, it is very difficult to compare data between the ICS and the ICE using the pre-1960 survey results.

4. Empirical Results

The empirical results of the reduced-form equation are presented in Tables 1 and 2. To conserve space, we report only the increment to the [R.sup.2] (or the adjusted [R.sup.2]) provided by the lagged values of the various measures of [S.sub.t] and the p-value of the joint significance of the lags of consumer confidence. We also estimate the reduced-form equation without the control variables in order to examine whether the confidence indicator, by itself, has good forecasting ability. As CFW (1994) point out, this part of our investigation allows us to examine the validity of Hall's (1978) random walk hypothesis. (7) Finally, due to the lag structure of the model, the effective sample size for Equation 1 is from 1961Q1 to 2002Q2.

Table 1 displays the predictive power of the ICE, the ICS, and the five individual survey questions when the control variables are excluded from the reduced-form equation. From this table, we find that the ICS, by itself, is a leading indicator in all four consumption categories. For example, we observe that in the case of total PCE, the sentiment index is statistically significant at the 1% level. Moreover, adding the last 4 quarters of data from the ICS to the prediction equation explains 9.5% of the variation in the next period's growth in total PCE (row 1). A similar finding is also detected for the three subcategories of total PCE: durable goods (row 2), nondurable goods (row 3), and services (row 4), where the 4 lags of the index in these cases are statistically significant at the 5% level or better. A careful inspection of the results of Table 1 further indicates that the incremental [R.sup.2]s for nondurable goods as well as for services are quite high, ranging from 8.9% to 9.9%. As regards the durable-goods case, lagged consumer sentiment contributes only 2.8% of the l-quarter-ahead variation in the growth of this consumption category.