The Relationship Between Consumer Sentiment and Stock
Prices
by
Kevin P. Christ
Assistant Professor of Economics, Rose-Hulman
Institute of Technology
and
Dale S. Bremmer
Professor of Economics, Rose-Hulman Institute of
Technology
To be presented at
the “Financial Economics” Session of the
78th
Annual Conference of the Western Economics Association International
in
The Relationship Between Consumer
Sentiment and Stock Prices
I. INTRODUCTON
Testable hypotheses about the relationship
between stock prices and consumer confidence seem to appear in the business
press whenever new consumer survey data is released. Headlines like “Rise in Consumer Sentiment
Sends Share Prices Higher”[1]
and “Stocks Tumble, Spurred by Dive in Consumer Confidence”[2]
suggest that stock prices respond directly to measures of consumer
confidence. Previous research into the
matter, however, suggests that the headline writers have it backwards – that
the direction of influence runs only one way, from stock prices to consumer
confidence, and that any stock price response to new information about consumer
confidence is ephemeral.
This paper focuses on
the short-run and long-run relationship between stock indices and measures of
consumer sentiment, and it presents three key empirical results. First, cointegration test confirm that there
is no long-run relationship between different stock indices and the
Following this
introduction, the second section of the paper presents a brief review of the
literature about the relationship between measures of consumer confidence and
other economic variables. Unit root
tests, cointegration tests, and Granger causality tests are discussed in the
third section. The fourth section of
paper presents a model to predict consumer confidence. These predictions and their errors are used
to explain stock prices. The final
section of the paper summarizes the results and offers conclusions.
II. LITERATURE REVIEW
As interest in the
fortunes of the stock market have increased, so too has interest in the links
between stock market indexes and other indicators of economic activity. Recently, researchers have turned their
attention to potential links between stock indexes and measures of consumer
confidence. Generally, stock indexes and
measures of consumer confidence appear to be contemporaneously correlated, with
the direction of influence running from stock price movements to consumer
confidence, but not the other way (Otoo, 1999; Jansen and Nahuis, 2002). Despite this empirical evidence of causation,
there are reasonable theoretical links between stock indexes and consumer
confidence. Moreover, it is possible
that the direction of causation simultaneously works in both directions,
thereby making detection of causal relationships extremely difficult.
The two published
measures of consumer confidence in the
There are two channels
through which stock movements may influence consumer confidence. The first linkage is the traditional wealth
effect, in which movements in stock indexes translate into changes in current
wealth, thereby influencing consumer sentiment directly. The second channel is the "leading
indicator" linkage, in which consumers interpret current changes in stock
indexes as reliable indicators of future income changes (Poterba and Samwick,
1995; Morck, Shleifer and Vishny, 1990).
Either scenario -- changes in current wealth or anticipated changes in
future income --may reasonably be expected to directly influence consumer sentiment.
Using individual observations from the
Jansen and Nahuis (2002)
extend Otoo’s analysis to eleven European countries. With few exceptions, they find that stock
returns and changes in consumer confidence are positively correlated. Like Otoo, they find that stock prices
Granger-cause consumer confidence, but consumer confidence does not
Granger-cause stock prices. Their
empirical results confirm Otoo’s finding that higher stock prices are a leading
indicator that increases consumer confidence.
Jansen and Nahius characterize this leading indicator link as the
“confidence channel,” that is independent of the traditional “wealth effect.” The empirical results of both Otoo and Jansen
and Nahius suggest that the confidence channel is a separate transmission mechanism
that is not part of the conventional wealth effect.
Explanations of possible
causal relationships between consumer confidence and equity prices that work in
the other direction are theoretically reasonable, but empirically
unsupported. Again, there are two
channels of potential influence. The
first channel is the link between consumer spending and corporate profits. Several studies show that measures of lagged
consumer sentiment are statistically significant explanatory variables in
explaining the behavior of current household spending.[4] If this is so, then there should be an
indirect link between consumer sentiment and expected corporate profits, thus
providing a link (albeit tenuous) between consumer sentiment and stock
prices. However, the relationship
between consumer sentiment and measures of output differs considerably across
countries and across the different measures of consumer confidence. In terms of predicting future output,
measures of consumer confidence have less explanatory power than measures of
business confidence.[5] The second potential channel of influence is
the so-called "publication effect" (Jansen and Nahuis, 2002), whereby
publication of consumer survey data exerts a psychological effect on the
market. Such an effect, if it were to
occur, is most likely highly transitory.[6]
Three different stock indices were used to explore the relationship between stock prices and measures of consumer sentiment. While Otoo’s study only used the Wilshire 5000 index, this study uses three stock indices that are regularly reported in the news: the Dow Jones Industrials, the S&P 500, and the NASDAQ.
The
index of consumer confidence is generated by the
All of the data consist of monthly time series. With the exception of the NASDAQ stock index, the data sample is 1978:01 – 2003:01. However, the NASDAQ stock index is only available for the 1984:10 – 2003:01 sample.[7]
To avoid regressions with spurious results, each time series is tested for a unit root. First, all the variables are expressed in their natural logs. Table 1 reports the augmented Dickey-Fuller tests (Dickey and Fuller, 1979) for both levels data and first-differenced data. Data plots indicate that the underlying regressions used to generate the Dickey-Fuller test statistics for level data should include both an intercept and a trend variable. On the other hand, data plots of the first differences indicate that neither an intercept nor a trend variable is necessary in the regressions that generate the Dickey-Fuller test for the first-differenced data. Each test statistic is derived by including the number of lagged dependent variables that minimizes the Akaike Information Criteria (Akaike, 1973) for each specification.
Referring
to Table 1, the augmented Dickey-Fuller tests indicate that the null hypothesis
of a zero root cannot be rejected for measures of consumer confidence, the
three stock indices, and the unemployment rate.
The null hypothesis of zero roots is rejected at the five percent level
for the interest rate on the ten-year
Given that the consumer confidence and stock indices have zero roots, cointegration tests (Johansen and Juselius, 1990) are preformed to determine whether a long-run relationship exists between consumer confidence and each of the stock indices. If consumer confidence and a stock index are cointegrated, that implies a long-run relationship between the two variables exists. Three independent statistical tests were performed to determine whether consumer confidence and the Dow Jones Industrial, consumer confidence and the S&P 500, and consumer confidence and the NASDAQ were pair wise cointegrated. The results of the cointegration tests are reported in Table 2. The natural log of the consumer sentiment and stock index variables were used in deriving the statistics in Table 2.
The
test statistics reported in Table 2 are based on the null hypothesis that a
cointegrating vector between consumer confidence and a given stock index does
not exist. In other words, the null
hypothesis is there is no long-run relationship between a measure of consumer
sentiment and a given index of equity prices.
The trace test indicates that such a long-run relationship does not
exist. In each of the three cases, the
test statistic was less than the critical value associated with a five-percent
level of significance. Hence, the null
hypothesis of no long-run relationship cannot be rejected. Likewise, in all three cases, the
max-eigenvalue test statistic was less than the five-percent critical value. This adds additional evidence that the null
hypothesis of no long-run relationship cannot be rejected. The outcomes of these cointegration tests are
similar to the results that Jansen and
Nahuis found with their European data.
Given that there is no long-run statistical relationship between consumer confidence and the stock indices, the nature of the short-run relationship was explored. Granger-causality tests (Granger, 1969) were performed by estimating a two-equation, vector autoregressive system. Let the natural log of the index of consumer confidence in month t be denoted by ct, while the natural log of a given stock index in month t is denoted by st. There are two equations in the VAR. The first equation in the two-equation system of seemingly unrelated equations is
,
and the second equation of the VAR is
.
Notice that both equations (1) and (2) have the same lag structure, that is, there are N lagged explanatory variables for both the stock index in question and the consumer sentiment. Referring to equation (1), if γi = 0 for every i, then one could conclude that the stock index does not Granger cause consumer confidence. Likewise, if θi = 0 for every i, then consumer confidence does not Granger cause the stock index.
The results of the Granger causality tests are reported in Table 3. The regressions are preformed on first-differences of the natural logs of the variables to avoid the problems of spurious correlations caused when regressing time series with unit roots on each other. The length of the lag was chosen by picking that value of N that minimized the Akaike Information Criterion for each VAR. Regardless the stock index used, the Dow Jones, the S&P 500, or the NASDAQ, the results were the same. In all three cases, the null hypothesis that a given stock index did not Granger cause consumer sentiment was rejected at the one-percent level of significance. Likewise, in each of the three cases, the null hypothesis that measures of consumer confidence do not Granger cause measures of the stock index could not be rejected.
The finding that stock indices affect consumer confidence, but not the reverse, is consistent with the findings of Otoo and Jansen and Nahius. However, neither of these studies investigate whether the impact of expected changes in consumer confidence and unexpected changes in consumer confidence differ. If the stock market is efficient, then stock prices should reflect only expected changes in consumer confidence
IV. EFFECT OF EXPECTED AND
UNEXPECTED CONSUMER CONFIDENCE
The typical approach to estimating the relationship between movements in stock prices and movements in measures of consumer confidence is to regress changes in the logs of stock indexes on changes in the logs of consumer confidence indexes. A formal model is
where, as before, st is the natural log of a stock index for the current period, and ct and ct - 1 are the natural logs of the contemporaneous and previous period's consumer confidence index. Such regressions usually yield little predictive value, with R2s of less than 0.10, and statistical significance for only the contemporaneous consumer confidence variable. Our own estimations of such regressions yielded statistically significant results for the coefficient on the contemporaneous consumer confidence variable, and R2s ranging between 0.04 and 0.08.
Our alternative approach is to distinguish between expected changes in the consumer confidence index and unexpected changes. In an efficient market, movements in stock prices should only reflect expected changes in consumer confidence.
To investigate the proposition that only expected changes in the consumer confidence are related to stock indexes, it is necessary to first settle upon a forecasting model for consumer confidence. A naive approach would be to assume that the current period index is a function of its most recent value. The third column of Table 4 presents the results of estimating
.
As a baseline forecast, such a model performs well, as indicated by the high R2 and low Root Mean Square Error (RMSE).
A more sophisticated approach would be to assume that an index of consumer confidence is influenced by economic indicators that consumers use to formulate their personal views regarding the health of the economy. The last three columns of Table 4 present the results of such a regression using different stock indexes:
where ut – 1 is the natural log of the previous period's unemployment rate. As indicated by the high R2s and low RMSEs, these forecasting models also perform well.
The predicted values from regressions such as equation (4) or equation (5) are used to forecast expected changes in consumer confidence, while the residuals of these regressions serve as proxy for unexpected changes in consumer confidence. The intuition behind this approach is that markets should fully anticipate expected changes in consumer sentiment.
Table 5 presents the estimation results of two different specifications. The first specification, the basic model, assumes the change in stock prices is a function lagged changes in interest rates and the actual change in consumer confidence, with no attempt to distinguish between expected and unexpected changes in consumer confidence. This specification is
where it is the natural
log of the yield on the ten-year
.
In each case, the basic model yields results that are similar to those found in previous research: the coefficient on the change in consumer confidence is statistically significant, but the overall model yields little explanatory value as exhibited by very low R2s. The revised model has considerably more explanatory power, and the importance of the expected change in consumer confidence is ten times greater than that for the actual change in consumer confidence in the basic model. Forecasted changes in consumer confidence are priced into the market. Surprise changes in consumer confidence are not. This is precisely what one would expect from an efficient market.
V. CONCLUSIONS
Similar
to the results of Jansen and Nahuis, this paper finds that measures of consumer
confidence and stock indices exhibit unit roots. The finding that no long-run relationship
between
However, in a departure from the studies of Jansen and Nahius and Otoo, this paper finds that stock prices reflect expected changes in consumer confidence. There is no statistically significant correlation between unexpected changes in consumer confidence and stock prices. This finding complements the theoretical conclusions of the efficient markets literature. Forecasts of expected changes in consumer confidence are based on commonly available data that are also incorporated in the determination of stock prices. Expected increases in consumer confidence lead to increases in the demand for stock and higher equity prices. Our empirical results are consistent with the notion that unexpected changes in consumer confidence exhibit no sustained correlation with stock prices.
References
Akaike, H., “Information Theory and an
Extension of the Maximum Likelihood Principle,”
2nd International Symposium on Information Theory, B. N. Petrov
and F. Csaki (eds.),.
Bram, Jason and Sydney Ludvigsun, “Does
Consumer Confidence Forecast Household Expenditure? A Sentiment Index Horse Race,” Federal
Reserve Bank of New York Economic Policy
Review, June 1998, 4(2), 59-78.
Carroll, Christopher, Jeffrey Fuhrer, and
David Wilcox, “Does Consumer Sentiment Forecast Household Spending? If So, Why?” American Economic Review,
December 1994, 84(5), 1397-1408.
Dickey, David and Wayne Fuller, “Distribution
of the Estimators for Autoregressive Time Series with a Unit Root,” Journal
of the American Statistical Association, June 1979, 74(366), 427-431.
Fama, Eugene, “Efficient Capital Markets: A
Review of Theory and Empirical Work,” The Journal of Finance,
May 1970, 25(2), 383-417.
Granger, C. W. J., “Investigating Causal
Relations by Econometric Methods and Cross-Spectral Methods,” Econometrica,
July 1969, 37(3), 424-438.
Jansen, W. Jos and Niek J. Nahuis, “The Stock
Market and Consumer Confidence: European Evidence,” Monetary and Economic
Policy Department, De Nederlandsche Bank, July 2002,
http://www.dnb.nl/monetair_beleid/pdf/serie2002-11.pdf .
Johansen, Soren and Katarina Juselius,
“Maximum Likelihood Estimation and Inference on Cointegration - - With
Applications to the Demand for Money,”
Morck, Randall, Andrei Shleifer, and Robert
Vishny, “The Stock Market and Investment: Is the Market a Sideshow?” Brookings
Papers on Economic Activity, 1990, 2, 157-202.
Otoo, Maria Ward, “Consumer Sentiment and the
Stock Market,” Board of Governors of the Federal Reserve System, November 1999,
http://www.federalreserve.gov/pubs/feds/1999/199960/199960pap.pdf .
Poterba, James M. and Andrew A Samwick,
“Stock Ownership Patterns, Stock Market Fluctuations, and Consumption,” Brookings
Papers on Economic Activity, 1995, 2, 295-357.
Santero, Teresa and Niels Westerlund,
“Confidence Indicators and Their Relationship to Changes in Economic Activity,”
Organization for Economic Cooperation and Development, Economic Department,
Working Papers, No. 170, 1996, http://www.oecd.org/pdf/M00001000/M00001194.pdf
Souleles, Nicholas, “Consumer Sentiment: Its
rationality and Usefulness in Forecasting Expenditure - - Evidence from the
Michigan Micro Data,” National Bureau of Economic research Working Paper
Series, Working paper 8410, August 2001, http://www.nber.org/papers/w8410.
Table 1
Augmented Dickey-Fuller Tests for Unit Roots
|
|
|
|
Levels Data: Regressions Include Intercept
and Trend |
|||
Variable (in natural log) |
Test Statistic |
Lags |
Sample |
Consumer Confidence |
-3.12 |
0 |
1978:02 – 2003:01 |
Dow Jones Industrial |
-2.16 |
0 |
1978:02 – 2003:01 |
S&P 500 |
-1.36 |
0 |
1978:02 – 2003:01 |
NASDAQ |
-1.32 |
1 |
1984:12 – 2003:01 |
Unemployment Rate |
-2.91 |
6 |
1978:08 – 2003:01 |
10-Year Government Bond |
-3.66** |
3 |
1978:05 – 2003:01 |
|
|
|
|
First-Differenced Data: Regressions Don’t
Include Intercept or Trend |
|||
Variable (in natural log) |
Test Statistic |
Lags |
Sample |
Consumer Confidence |
-9.06*** |
5 |
1978:08 – 2003:01 |
Dow Jones Industrial |
-16.91*** |
0 |
1978:03 – 2203:01 |
S&P 500 |
-6.67*** |
4 |
1978:07 – 2003:01 |
NASDAQ |
-13.12*** |
0 |
1984:12 – 2003:01 |
Unemployment Rate |
-4.35*** |
5 |
1978:08 – 2003:01 |
10-Year Government Bond |
-8.83*** |
2 |
1978:05 – 2003:01 |
|
|
|
|
***
indicates the null hypothesis that the time series has a unit root is rejected
at the 1% level, while ** indicates the same null hypothesis is rejected at the
5% level.
Table 2
The Long-Run relationship between Consumer Confidence and a Stock Index : Johansen Cointegration Test Results
|
|
|
|
Null Hypothesis: A cointegrating equation for the following two variables does not exist. |
|||
|
|
|
|
|
Pair-Wise Combination of Consumer Confidence and Stock Index |
||
|
|
|
|
|
(1) Consumer Confidence And Dow Jones |
(2) Consumer Confidence And S&P 500 |
(3) Consumer Confidence And NASDAQ |
|
|
|
|
Sample |
1978:04 - 2003:01 |
1978:04 - 2003:01 |
1985:02 - 2003:01 |
Eigenvalue |
0.03 |
0.04 |
0.06 |
Number of lags |
2 |
2 |
3 |
Akaike Information Criteria |
-6.71 |
-6.74 |
-6.04 |
|
|
|
|
Trace Test |
|||
|
|
|
|
Test Statistic |
11.40 |
13.13 |
15.74 |
5% Critical Value |
15.41 |
15.41 |
25.32 |
1% Critical Value |
20.04 |
20.04 |
30.45 |
|
|
|
|
Max-Eigenvalue Test |
|||
|
|
|
|
Test Statistic |
10.60 |
11.28 |
13.71 |
5% Critical Value |
14.07 |
14.07 |
18.96 |
1% Critical Value |
18.63 |
18.63 |
23.65 |
|
|
|
|
Table 3
Pair-wise Granger Causality Tests
Model A:
Model B:
|
|
|
|
All dependent and explanatory variables
are first differences of natural logs. |
|||
|
|||
Pair of Variables: Consumer Confidence (Δct
) and Dow Jones Industrial (Δst) |
|||
Null Hypothesis |
Test Statistic |
Lags |
Sample |
Dow Jones Does Not Granger Cause Consumer Confidence |
15.39*** |
2 |
1978:04 – 2003:01 |
Consumer Confidence Does Not Granger Cause Dow Jones |
0.29 |
|
|
|
|
|
|
Pair of Variables: Consumer Confidence (Δct
) and S&P 500 (Δst) |
|||
Null Hypothesis |
Test Statistic |
Lags |
Sample |
S&P 500 Does Not Granger Cause Consumer Confidence |
15.38*** |
2 |
1978:04 – 2003:01 |
Consumer Confidence Does Not Granger Cause S&P 500 |
0.17 |
|
|
|
|
|
|
Pair of Variables: Consumer Confidence (Δct
) and NASDAQ (Δst) |
|||
Null Hypothesis |
Test Statistic |
Lags |
Sample |
NASDAQ does not Granger Cause Consumer Confidence |
14.32*** |
2 |
1985:01 – 2003:01 |
Consumer Confidence Does Not Granger Cause NASDAQ |
0.18 |
|
|
|
|
|
|
***
indicates the null hypothesis can be rejected at the 1% level.
Table 4
Forecasting models for consumer confidence
Regression Parameter |
Associated Intercept Or Explanatory Variable |
Naïve Forecast (1) 1978:02 – 2003:01 |
Complex Forecast
(2) 1978:03 – 2003:01 |
Complex Forecast (3) 1978:03 – 2003:01 |
Complex Forecast (4) 1984:12 – 2003:01 |
α0 |
Intercept |
0.198** (0.762) |
0.977 (0.890) |
2.082*** (0.639) |
3.452*** (0.387) |
α1 |
Lagged dependent variable, |
0.956*** (0.017) |
|
|
|
α2 |
Contemporaneous Dow Jones
Index, |
|
0.181** (0.056) |
|
|
α3 |
Lagged Dow Jones Index, |
|
0.253*** (0.056) |
|
|
α2 |
Contemporaneous S&P 500
Index, |
|
|
0.188*** (0.057) |
|
α3 |
Lagged S&P 500 Index, |
|
|
0.232*** (0.057) |
|
α2 |
Contemporaneous NASDAQ Index, |
|
|
|
0.074** (0.035) |
α3 |
Lagged NASDAQ Index, |
|
|
|
0.132*** (0.035) |
α4 |
Lagged Unemployment Rate, |
|
-0.191* (0.094) |
-0.191** (0.094) |
-0.224** (0.098) |
r |
Autoregressive parameter |
|
0.991*** (0.008) |
0.987*** (0.009) |
0.964*** (0.017) |
|
|
|
|
|
|
R2 |
|
0.913 |
0.921 |
0.921 |
0.884 |
DW |
|
1.967 |
2.139 |
2.140 |
2.183 |
F-test |
|
3,136.56† |
861.99† |
857.13† |
404.59† |
Forecast RMSE |
|
0.046 |
0.044 |
0.044 |
0.038 |
Dependent variable: ct
= natural log of the
Table 5
Relationship Between Expected and Unexpected Changes in Consumer Confidence
|
= change in the natural log of the Dow Jones Industrial
Index |
= change in the natural log of the S&P 500 Index |
= change in the natural log of the NASDAQ Composite Index |
||||
Regression Parameter |
Basic model using actual |
Revised model: and generated from forecast (2) in Table 4 |
Basic model using actual |
Revised model and generated from forecast (3) in Table 4 |
Basic model using actual |
Revised model and generated from forecast (4) in Table 4 |
|
a0 |
0.007** (0.002) |
0.008*** (0.002) |
0.007** (0.002) |
0.008*** (0.002) |
0.001 (0.005) |
0.008** (0.004) |
|
a1 |
-0.221*** |
-0.123*** (0.043) |
-0.186*** |
-0.090** (0.043) |
-0.148 (0.131) |
-0.150 (0.105) |
|
a2 |
0.190*** (0.057) |
|
0.195*** (0.056) |
|
0.316** (0.127) |
|
|
a3 |
|
1.820*** (0.125) |
|
1.910*** (0.131) |
|
2.936*** (0.336) |
|
a4 |
|
-0.017 (0.045) |
|
-0.012 (0.045) |
|
0.034 (0.117) |
|
r |
-0.076 (0.060) |
-0.414*** (0.059) |
-0.072 (0.060) |
-0.373*** (0.060) |
0.045 (0.071) |
-0.259*** (0.077) |
|
|
|
|
|
|
|
|
|
R2 |
0.076 |
0.417 |
0.067 |
0.422 |
0.042 |
0.252 |
|
DW |
2.002 |
2.158 |
2.012 |
2.130 |
1.992 |
2.054 |
|
F-test |
8.04† |
52.29† |
7.02† |
53.37† |
3.09 |
17.79† |
|
Sample |
1978:04 – 2003:01 |
1978:04 – 2003:01 |
1978:04 – 2003:01 |
1978:04 – 2003:01 |
1894:12 – 2003:01 |
1894:12 – 2003:01 |
|
The basic regression model
is . The revised regression model is . *, **, and *** indicate that the null hypothesis that
the slope coefficient is equal to zero may be rejected at the 10%, 5%, and 1%
level of significance, respectively.
These t-tests are two-tail
tests. † indicates that the
null hypothesis that all the slope coefficients are simultaneously equal zero
is rejected at the 1% level.
Appendix
The Index of Consumer Sentiment (ICS)
Calculated by the Survey Research Center, University of Michigan
The ICS is derived from the responses to the following five questions. |
|
x1 |
“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?” |
x2 |
“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?” |
x3 |
“Now turning to business conditions in the country as a whole - - do you think that during the next twelve months we’ll have good times financially, or bad times, or what?” |
x4 |
“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?” |
x5 |
“About the big things people buy for their homes - - such as furniture, a 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?” |
xi is equal to the percent of favorable replies minus percent of unfavorable replies plus 100, rounded to nearest whole number.
The denominator, 6.7558, is the 1966 base period total, and the addition of 2.0 corrects for survey design changes that occurred in the 1950s.
Source:
[1] New York Times,
[2] New York Times,
[3] For a comparison of these two measures of consumer confidence, see the discussion by Bram and Ludvigson (1998).
[4] See Carroll, Fuhrer, and Wilcox (1994), Bram and Ludvigson (1998), and Souleles (2001).
[5] See Santero and Westerland (1996).
[6] On
[7] Consumer sentiment and the yield on