# Can consumer confidence forecast household spending? Evidence from the European Commission Business and Consumer Surveys.

1. IntroductionThe analysis of the impact of consumer attitudes on consumer spending has been an area of sustained interest in the macroeconomics literature. This interest in consumer attitudes reflects a popular belief that macroeconomic outcomes depend heavily on consumers' expectations of future economic conditions. Reinforcing this belief is the fact that consumer spending accounts for 60% to 70% of the Gross Domestic Product of highly industrialized countries such as the G-7.

During the past few decades, a number of single-country investigations have attempted to assess the predictive power of consumer confidence in forecasting household spending. For example, in the United States, Mishkin (1978) reports that the Index of Consumer Sentiment (ICS) published by the University of Michigan possesses good explanatory power for changes in durable goods. Carroll, Fuhrer, and Wilcox (1994) also find that the Michigan ICS has good predictive capability with regard to household expenditure but that its forecasting power decreases considerably when the Index is used along with other macroeconomic variables. (1) Kwan and Cotsomitis (2004) examine the robustness of Carroll, Fuhrer, and Wilcox's (1994) findings when alternative measures of consumer confidence are used in their prediction equations. They report that use of the Michigan Index of Consumer Expectations improves on the estimation results of Carroll, Fuhrer, and Wilcox (1994). Bram and Ludvigson (1998) and Ludvigson (2004) compare the forecasting performance of the Michigan ICS and the U.S. Conference Board's Index of Consumer Confidence (ICC). Their findings indicate that the Conference Board's index provides additional information about future consumer consumption that is not captured by the Michigan index. However, Chopin and Darrat (2000), utilizing a multivariate causality framework, report that the Conference Board's ICC is an unreliable predictor of U.S. retail sales.

As regards other studies using non-U.S. data, Acemoglu and Scott (1994) employ Granger causality and regression analyses to determine if consumer confidence, as measured by the Gallup Poll in the United Kingdom, can predict future consumption. They find that consumer confidence is a leading indicator of future consumption growth. In a subsequent study, Delorme, Kamerschen, and Voeks (2002) examine whether consumer confidence and rational expectations are the same in the United Kingdom as they are in the United States. They report that the predictive ability of the U.K consumer confidence index is greater than that of the United States. Belessiotis (1996), using French data, also reports that the consumer confidence index provides decent explanatory power for future consumer spending. In Canada, Kwan and Cotsomitis (2005) find that although consumer sentiment is a reliable predictor of consumer expenditures at the national level, results obtained using regional data are quite mixed. Fan and Wong (1998) find that, unlike the case in the United States or the United Kingdom, confidence indicators in Hong Kong have little or no explanatory power in forecasting household spending.

The existing studies on the ability of consumer confidence to predict household spending have produced mixed results. The results vary, depending on the sample period chosen, the frequency of observations (monthly or quarterly), and the different survey methodologies employed to gauge consumer attitudes. Moreover, most of these studies confine their attention to single-country investigations. Given the diversity of results, it is important to investigate the predictive power of consumer confidence on the basis of comparable time series data for as many countries as possible.

This paper presents a first formal attempt to study the ability of consumer confidence to forecast household spending within a multicountry framework. To this end, we use two confidence indices, namely the Consumer Confidence Indicator (CCI) and the Economic Sentiment Indicator (ESI), both of which are derived from the European Commission Business and Consumer Survey (ECBCS). An important advantage of using the ECBCS data is that both the business and consumer surveys are harmonized because identical questionnaires are used in the 15 EU countries surveyed. Also, to our knowledge, the ECBCS is the most comprehensive survey of consumer and business attitudes in use today. Because of the coverage, measurability, and representativeness of these survey data, we are able to assess the predictive performance of consumer confidence with regard to household spending on the basis of data that have a reasonable degree of comparability across countries and over time and that are available for a number of countries. Our main empirical findings indicate that: (i) household spending has a slightly higher contemporaneous correlation with the ESI than with the CCI; (ii) the in-sample incremental predictive power of these confidence indicators varies significantly in our samples when important macroeconomic variables are included in the prediction equation; and (iii) the out-of-sample forecasting ability of the confidence indicators examined is quite low.

The remainder of this paper is structured as follows. Section 2 describes the construction of the two confidence indicators. Section 3 discusses the econometric methodology and the data used in our empirical analysis. Section 4 presents our main empirical findings. Section 5 reports our out-of-sample forecasts. Section 6 offers some concluding remarks.

2. Confidence Indicators

The ECBCS consists of five monthly surveys: Industry, Construction, Consumers, Retail Trade, and the Service sector. Since the early 1960s, the results of these surveys have provided decision makers, researchers, and managers with relevant information for the evaluation of current and future economic conditions. The rising demand for these results both by the public and the business sectors, as well as the expanding coverage in the press, seems to confirm their importance and usefulness.

As summarized in Table 1, the CCI for each of the surveyed countries is derived from four different attitudinal questions that form part of the European Commission's consumer survey. (2) The questions inquire about the respondents' financial position, expected changes in economic situation, unemployment level, and savings attitudes over the next 12 months. For each question, there are six answers (PP, P, E, N, NN, and NA), which range from very favorable to very unfavorable. Consumer Confidence is defined as the difference between the percentages of favorable and unfavorable replies to the four questions, where PP and NN scores receive weight 1, P and N scores receive weight 1/2, and E and NA scores receive weight 0. The score for each question is equal to (PP + 1/2 P) - (NN + 1/2 N). The CCI is the arithmetic average of the scores on the four questions.

In 1985, the European Commission designed a broader measure of confidence, namely the ESI, in order to better reflect the public's perception of future economic activity. This confidence indicator is constructed on the basis of four different confidence indicators:

1. Industrial confidence indicator [weight 40%]

2. Consumer confidence indicator [weight 20%]

3. Construction confidence indicator [weight 20%]

4. Retail trade confidence indicator [weight 20%]

Like the CCI, the other three confidence indicators are also calculated as the arithmetic average of the scores on the questions chosen among the full set of questions of the respective survey. (3,4) The ESI is defined as the weighted mean of the scores of the four confidence indicators (indices 1-4). Because the ESI combines the judgment and attitudes of both consumers and producers, it is generally considered to be a composite leading indicator that should be more informative than the CCI in anticipating changes in the direction of the EU economy and the economies of its member countries.

3. Econometric Methodology and Data

To examine the forecasting ability of consumer confidence on household spending, we use the prediction equations given in Carroll, Fuhrer, and Wilcox (1994), Brain and Ludvigson (1998), and Ludvigson (2004):

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

and

[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 real total personal consumption expenditures (PCE), [S.sub.t] is proxied by either the CCI, each of the four attitudinal questions (QFPE, QESE, QUE, and QSE), or the ESI, [Z.sub.t] is a vector of control variables, and [[epsilon].sub.t] is a nonautocorrelated disturbance term. Equation 1 examines whether lagged consumer confidence by itself can forecast changes in real total PCE. (5) Equation 2 is used to test the incremental predictive power of the past values of [S.sub.t] in the presence of control variables [Z.sub.t]. In this paper, the control variables, [Z.sub.t], include four lags of the dependent variable, four lags of the growth in real labor income, four lags of the log first difference in the real stock price index, fours lags of the first difference of the short-term interest rate, four lags of unemployment rate, and four lags of the confidence indicators of Germany, France, and the United Kingdom. We anticipate that consumption growth is positively associated with past labor income growth. The stock price index and the interest rate variables are used to capture important information from financial markets. The unemployment rate serves as a proxy for labor market conditions. The four lags of the confidence indicators of Germany, France, and the United Kingdom are included in our prediction equations to capture potential EU intercountry effects. (6,7) Note that the inclusion of these control variables is expected to minimize the likely effect of the "omission-of-variables" problem.

To measure the predictive capability of each confidence indicator, we compute the increment to the [[bar.R].sup.2] (or the adjusted R-squared) provided by the lagged values of [S.sub.t] given in Equations 1 and 2. If the incremental [[bar.R].sup.2] is statistically significant at the chosen level of confidence, we can say that [S.sub.t] is incrementally informative about future consumption expenditure.

Our empirical investigation has been conducted using quarterly data for nine of the EU countries surveyed by the European Commission: Belgium, Denmark, France, Italy, Germany, Portugal, Spain, the Netherlands, and the United Kingdom. We exclude the other five EU countries (Austria, Greece, Ireland, Finland, and Sweden) from our study because of data limitations. (8) With the exception of the CCI, the four attitudinal questions, and the ESI, all data used in Equations 1 and 2 are obtained from the Datastream International Database. It is important to note that consistent measures of the short-term interest rate for the countries examined are not available. We, therefore, employ the 3-month treasury bill rate for Belgium and the United Kingdom, the 3-month discount rate for Denmark and Portugal, and the 3-month lending rate (prime rate) for France, Germany, Italy, the Netherlands, and Spain. The stock price data used in our estimation are the Brussels All Share Index for Belgium; the Copenhagen SE-DS General Index for Denmark; the XETRA DAX Index for Germany; the CAC 40 Index for France; the Milan MIB Storico General Index for Italy; the CBS All Share General Index for the Netherlands; the Portugal PSI General Index for Portugal; the Madrid SE General Index for Spain; and the FTSE Index for the United Kingdom. Data for the confidence indicators used are provided by the European Commission. (9) The sample periods employed in this paper are: 1987:IV-2002:III for Belgium, Denmark, France, Germany, Italy, Spain, and the United Kingdom; 1988:I-2002:III for the Netherlands; and 1988:II-2002:III for Portugal. Because consumption spending, labor income, and stock returns are measured in real terms, we use the consumer price index (CPI) to deflate these data.

4. Empirical Results

Correlation Analysis

Before we report the empirical results of our two prediction equations, it would be useful to first examine the statistical association between consumer spending and our two confidence indicators. The first column in Table 2 reports the contemporaneous correlation between quarterly growth in real total PCE and the CCI. As can be seen from this table, these two series reveal a close connection in five countries--Belgium, France, Italy, Portugal, and the United Kingdom--where we can reject the null hypothesis that the contemporaneous correlation between these two series is zero at the 10% level. We further note that the UK case displays the strongest correlation (0.476). As for the other four countries examined (Denmark, Germany, Spain, and the Netherlands), we fail to find any discernible relationship between consumption growth and the CCI.

Column 2 of Table 2 presents the contemporaneous correlation between the quarterly growth in real total PCE and the ESI. Our correlation results indicate that in six out of nine countries canvassed (Belgium, France, Italy, Portugal, Spain, and the United Kingdom), the estimated correlation coefficients are not only positive but also statistically significant, at least at the 10% level. (10) It is interesting to mention that among the six "significant" cases, four of them (Belgium, Italy, Spain, and the United Kingdom) have correlation values that are higher than those based on the CCI (see Column 1). Furthermore, the United Kingdom again reveals the strongest correlation between household spending and economic sentiment (0.491). As regards the other five "significant" cases, the size of the correlation ranges from 0.244 for Spain to 0.421 for France. Overall, the results presented in Table 2 indicate that our indices of consumer confidence and household spending are correlated, with the ESI exhibiting a slightly stronger association than does the CCI.

Although suggestive, the existence of a contemporaneous correlation between consumption growth and our two confidence indicators does not necessarily imply that lagged confidence is able to forecast movements in current consumption. The regression results offered below allow us to formally assess the ability of consumer confidence to predict future household spending.

Regression Results Based on Individual-Country Confidence Indicators

Tables 3 and 4 report the in-sample predictive power of the confidence indicators employed in our prediction equations. To conserve space, we report only the increment to the [[bar.R].sup.2] provided by the lagged values of the confidence measures and the p-value of the joint significance of the lags of these variables. (11) Following Carroll, Fuhrer, and Wilcox (1994), Bram and Ludvigson (1998), and Kwan and Cotsomitis (2004, 2005), the hypothesis tests adopted are conducted using a heteroscedasticity and serial correlation robust covariance matrix. (12)

Table 3 displays the results of the in-sample forecasting capability of the confidence indicators when the control variables are excluded from the prediction equation (i.e., Eqn. 1). On the basis of these results, the following points are worthy of note.

First, we find that the CCI alone is able to predict future household spending in only three out of nine countries examined: France, Spain, and the United Kingdom; where the coefficients on the lags of the confidence indicator are statistically significant, at least at the 10% level. (13) As for Belgium, Denmark, Germany, Italy, Portugal, and the Netherlands, we notice that the CCI does not provide any explanatory power for variations in the next period's consumption growth.

Second, the incremental [[bar.R].sup.2] varies considerably in the three "significant" cases. For example, in the case of the United Kingdom, adding the last four quarters of data from the CCI to the prediction equation can explain more than 19% of the variation in the next period's growth in real total PCE (row 9). However, for Spain (row 8), the confidence index has a predictive power of only 1.9%.

Third, the forecasting ability of the attitudinal questions is fairly weak. We find that the attitudinal questions are statistically significant at the 10% level or above in only 14 out of 36 cases. Further, even in these "significant" cases, there are only eight entries where the incremental [[bar.R].sup.2] is larger than 10%. As regards the relative performance of the individual questions, we notice that the question on financial position expectation (QFPE) performs best in the case of Portugal, (14) Spain, and the United Kingdom, whereas the question on economic situation expectation (QESE) is found to be the most reliable predictor of France's real total PCE. In the case of Belgium, the question regarding savings expectation (QSE) is the only question that carries a significant p-value. Nevertheless, its predictive power is barely 0.3%, suggesting that QSE contains extremely small forecasting ability for movements in future consumer spending.

Fourth, the ESI by itself provides explanatory power in predicting future consumer spending in five of the nine countries surveyed, namely Belgium, Denmark, France, Italy, and the United Kingdom, where the four lags of the sentiment variable are statistically significant at the 10% level. We note that in these "significant" cases, the sentiment indicator exhibits decent predictive power, with the incremental [[bar.R].sup.2] ranging from 8% (Denmark) to the 19.9% (the United Kingdom). With respect to Germany, Italy, Portugal, and the Netherlands, we find that the four lags of the ESI are not statistically significant at the 10% level, suggesting that this confidence indicator is not a useful predictor of future consumption growth in these four EU countries.

Table 4 presents the in-sample incremental forecasting performance of various confidence indicators when the control variables are included in the prediction equation (i.e., Eqn. 2). On the basis of these results, several interesting points emerge:

First, the incremental predictive power of the CCI is quite limited in the presence of other macroeconomic variables. We find that in only four out of nine countries examined (Portugal, Spain, the Netherlands, and the United Kingdom), the coefficients on the four lags of the CCI are statistically significant at the 10% level or better. We also find that the predictive value of the CCI as measured by the incremental [[bar.R].sup.2] varies considerably, ranging from only 0.4% for the Netherlands to 12.8% for Spain.

Second, the incremental forecasting power of the CCI is sensitive to the inclusion of the control variables in our prediction equation. For instance, we find that in the case of France, the CCI is now unable to track next period's household spending. This finding is in sharp contrast to our previous result that the confidence indicator by itself has good forecasting ability for future consumption growth. As for Portugal and the Netherlands, we observe that the confidence indicator becomes statistically significant at the 1% level when the control variables are present in the prediction equation.

Third, in terms of individual attitudinal questions, the question on savings expectations (QSE) performs best in four out of the nine countries investigated (Germany, Netherlands, Spain, and the United Kingdom), with the incremental [[bar.R].sup.2] varying from 1.3% for the Netherlands to 30.5% for Spain. On the other hand, the question about economic situation expectation (QESE) is found to be the most useful predictor of future consumption growth in three European countries (Belgium, Denmark, and France), where the incremental [[bar.R].sup.2] ranges from 1.4% for Denmark to 6.4% for France. As regards Italy and Portugal, we note that none of the questions appear to have good predictive capability. (15)

Fourth, there are six cases (Denmark, Germany, Portugal, Spain, the Netherlands, and the United Kingdom) where the coefficients on the four lags of the ESI are statistically significant, at least at the 10% level. However, the forecasting ability of this sentiment indicator in the four "significant" cases is quite weak; the incremental [[bar.R].sup.2] in the case of Denmark, Portugal, the Netherlands, and the United Kingdom is consistently smaller than 1.7%. With respect to Belgium, France, and Italy, our regression results clearly indicate that the ESI does not have any additional explanatory power for the future path of household spending in these three European economies.

On the whole, we find that, analogous to previous single-country investigations, there is much diversity in the in-sample incremental forecasting performance of our two confidence indicators. This diversity in outcomes suggests that economic forecasters and policy makers in the EU should be careful when using either the CCI or ESI to predict consumption growth in these countries.

Regression Results Based on an Aggregate Consumer Confidence Index

The ECBCS also reports an aggregate CCI for the EU on the basis of the surveyed results received from each of the 15 member countries. This aggregate CCI is derived from aggregate replies to the questionnaires that are calculated as weighted averages of the country-aggregate replies. (16) Because the aggregate CCI contains valuable information on consumers' expectations of the future economic trend of the EU as a whole, it would be interesting to examine the ability of this Index to forecast the consumption growth of the countries surveyed. (17) In this subsection, we rerun the results of Table 4 using the aggregate CCI as a proxy for individual-country consumer confidence. Our main results are summarized in Table 5.

The main message of Table 5 is that the in-sample incremental predictive performance of the aggregate CCI (denoted as [CCI.sup.EU]) is generally significant. We find that in six out of nine cases (Belgium, France, Portugal, Spain, the Netherlands, and the United Kingdom), the coefficients on the lags of the aggregate CCI are statistically significant at the 1% level. Also, among these six "significant" cases, four of them display stronger incremental forecasting power than that of the individual-country CCI. (18) As a result, we believe that econometric models currently used to forecast consumption growth in these four countries might be made more accurate by using the aggregate CCI as a leading indicator instead of the individual country CCI. As for Denmark, Germany, and Italy, the results presented in Tables 4 and 5 indicate that neither the aggregate CCI nor its individual-country index is able to predict future consumption growth.

5. Out-of-Sample Forecasts

In this section, we examine the ability of Equation 2 to predict out of sample. In particular, we want to see whether the inclusion of the confidence indicators improves the forecasting performance of this prediction equation since 1999:I. (19) For the out-of-sample tests, we employ recursive regressions to reestimate Equation 2, adding one quarter at a time and computing a series of one-step-ahead forecasts for the period 1999:I-2002:III.

The out-of-sample test is carried out as follows: (i) The forecasting equation is first estimated using data from late 1980s to 1998:IV. (ii) A one-step-ahead forecast is then carried out for the subsequent period 1999:I-2002:III. The forecasts are evaluated by calculating the mean-squared error (MSE) from the set of one-step-ahead forecasts. (iii) Last, steps (i) and (ii) are repeated, except that the forecasting equation is now estimated without the confidence measure.

Table 6 summarizes the out-of-sample predictive performance of various measures of consumer confidence. The numbers reported in this table are the Clark and McCracken (2002), hereinafter CM, out-of-sample MSE-T test statistics. The CM test is a test of equal forecast accuracy. Under the null hypothesis that lagged values of consumer confidence have no predictive power for current consumption growth, the restricted model, which excludes the four lags of St, has a MSE that is less than or equal to that of the unrestricted model (i.e., Eqn. 2), which includes the four lags of St. Under the alternative, the MSE generated from the unrestricted model should be smaller than that of the restricted model. Hence, the CM test is described as one-sided to the right. (20) In order to use their test in empirical applications, Clark and McCracken propose employing a bootstrap approach to conduct inferences. (21)

From Table 6, we see that our confidence indicators, which include the CCI, the aggregate CCI, and the ESI, display weak out-of-sample predictive capability for the forecast period 1999:I-2002:III. (22) We find that the CM test value is positive in only 4 out of 27 cases examined. For the remaining 23 cases, the CM test statistics are negative, indicating that the addition of past values of these confidence indicators actually lowers the quality of the out-of-sample forecast relative to when it is absent from the prediction equation. As for the forecasting performance of the individual questions, we find that the CM test statistics are positive and statistically significant at the 10% level in only 2 out of 36 cases. As regards the other 34 cases, the CM test statistics are negative. These results indicate very limited out-of-sample predictive capability on the part of our measures of consumer confidence.

6. Concluding Remarks

The objective of this paper has been to examine the usefulness of various measures of consumer confidence in forecasting household spending on the basis of ECBCS data that have a reasonable degree of comparability over time and across a number of EU countries. Our empirical results indicate that there is much variability in the in-sample incremental forecasting performance of the CCI and ESI for the EU countries canvassed. Further, the results of our out-of-sample tests indicate that these confidence indices provide limited information about the future path of household spending. At best, consumer sentiment in these countries would appear to foreshadow the movement of other major macroeconomic variables. European economic forecasters and government policy makers should, therefore, be careful when using the CCI and ESI to predict consumption growth in EU countries.

Appendix A

Industrial Confidence Indicator

The industrial confidence indicator is the arithmetic average of the balances (in percentage points) of the answers to the questions on production expectations, order books, and stocks of finished products.

Question 1. Orderbooks?

--more than sufficent

--sufficient

--not sufficient

Question 2. Stocks of finished products?

--too large

--adequate (normal for the season)

--too small

Question 3. Production expectations for the months ahead?

--up

--unchanged

--down

Construction Confidence Indicator

The construction confidence indicator is the arithmetic average of the balances (in percentage points) of the answers to the questions on order book and employment expectations.

Question 1. Evaluation of order-books (or production schedules). We consider that our present order-book (production schedule) is:

--above normal

--normal

--below normal

Question 2. Employment outlook. We reckon that over the next three or four months the numbers we employ will:

--go up

--remain unchanged

--go down

Retail Trade Confidence Indicator

The retail trade confidence indicator is the arithmetic average of the balances (in percentage points) of the answers to the questions on the present and future business situations, and on stocks.

Question 1. We consider our present business (sales) position to be:

--good

--satisfactory (normal for the season)

--bad

Question 2. We consider our present stocks to be:

--too large

--adequate (normal for the season)

--too small

Question 3. Our business trend over the next six months, excluding purely seasonal variations, will:

--improve

--remain unchanged

--deteriorate

Appendix B: The MSE-T Test Statistic

In this appendix, we briefly review the asymptotic distribution of the MSE-T test statistic as proposed by CM (2002).

Suppose the total numbers of in-sample observations and out-of-sample observations to be R and P, respectively. Then, the number of [tau]-step ahead forecasts is P - [tau] + 1, as the total out-of-sample observations span R + [tau] through R + P. Thus, the total number of observations in the sample is R + P = T.

Let [[??].sub.1,t+[tau]] be the forecast error generated from the restricted model. Also, let [[??].sub.2,t+[tau] be the forecast error generated from the unrestricted model. Diebold and Mariano (1995) propose a test for equal MSE based on the sequence of loss differentials [[??].sub.t+[tau]] = [[??].sup.2.sub.1,t+[tau] - [[??].sup.2.sub.2,t+[tau]]. If we define

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where [MSE.sub.1] is the mean squared error obtained from the restricted model, [MSE.sub.2] is the mean squared error obtained from the unrestricted model, M = [1.5[tau]] for [tau] > 1, [*] is the nearest integer function, K(j/M) = 1 - j/(M + 1), [[??].sub.dd] = [[??].sub.dd](0) for [tau] = 1, the test statistic takes the following form:

MSE - T = [(p-[tau]+1).sup.1/2] [bar.d] / [square root of [[??].sub.dd]]. (3)

Under the null hypothesis of equal forecasting accuracy, [MSE.sub.1] = [MSE.sub.2], so that at and MSE-T are equal to zero. We test this hypothesis against the one-sided alternative hypothesis that the MSE for the unrestricted model forecasts is smaller than the MSE for the restricted model forecasts ([MSE.sub.1] > [MSE.sub.2]), so that MSE-T > 0 under the alternative hypothesis. Clark and McCracken (2002) show that the MSE-T statistic has a nonstandard asymptotic distribution when the restricted model is nested within the unrestricted model. In this situation, Clark and McCracken (2002) recommend basing inferences for the MSE-T statistic on a bootstrap procedure.

We would like to thank two anonymous referees and Professor Dek Terrell, the Co-Editor of this Journal, for constructive comments and suggestions. We are also grateful to the European Commission for supplying us the data on the European Commission Business and Consumer Survey. This research is supported by a grant (number 2020736) from the Chinese University of Hong Kong to the second author.

Received February 2004; accepted May 2005.

References

Acemoglu, Daron, and Andrew Scott. 1994. Consumer confidence and rational expectations: Are agents' beliefs consistent with the theory? Economic Journal 104:154-94.

Akaike, Hirogutu. 1973. Information theory and extension of the maximum likelihood principle. In Proceedings of the 2nd International Symposium on Information Theory, Asilomar, California. pp. 267-81.

Akaike, Hirogutu. 1974. A new look at the statistical model identification. IEEE Transaction on Automatic Control AC-19:716-23.

Belessiotis, Tassos. 1996. Consumer confidence and consumer spending in France. Economic Papers No. 116, European Commission.

Bram, Jason, and Sydney Ludvigson. 1998. Does consumer confidence forecast household expenditure? A sentiment index horse race. Economic Policy Review, Federal Reserve Bank of New York 4(2):59-78.

Carroll, Christopher D., Jeffrey C. Fuhrer, and David W. Wilcox. 1994. Does consumer sentiment forecast household spending? If so, why? American Economic Review 84:1397 1408.

Chopin, Marc, and Ali Darrat. 2000. Can consumer attitudes forecast the macroeconomy? The American Economist 44:34-42.

Clark, Todd, and Michael McCracken. 2002. Evaluating long-horizon forecasts. Federal Reserve Bank of Kansas City, Working Paper.

Curtin, Richard T. 2000. Psychology and macroeconomics: Fifty years of the surveys of consumers. Special Report, Surveys of Consumers, University of Michigan. Available at http://www.sca.isr.umich.edu/main.php.

Delorme, Charles D., David R. Kamerschen, and Lisa F. Voeks. 2002. Consumer confidence and rational expectations in the United States compared with the United Kingdom. Applied Economics 33:863-9.

Diebold, Francis X., and Roberto Mariano. 1995. Comparing predictive accuracy. Journal of Business and Economic Statistics 13:253-65.

Fan, Simon, and Phoebe Wong. 1998. Does consumer sentiment forecast household spending? The Hong Kong case. Economics Letters 58:77-84.

Hall, Robert E. 1978. Stochastic implications of the life cycle-permanent income hypothesis: Theory and evidence. Journal of Political Economy 86:971-87.

Klein, Lawrence R., and John B. Lansing. 1955. Decision to purchase consumer durable goods. Journal of Marketing Research 20:109-32.

Kwan, Andy C. C., and John A. Cotsomitis. 2004. Can consumer attitudes forecast household spending in the United States? Further evidence from the Michigan Survey of Consumers. Southern Economic Journal 71 :136-44.

Kwan, Andy C. C., and John A. Cotsomitis. 2005. The usefulness of consumer attitudes in forecasting household spending in Canada: A national and regional analysis. Economic Inquiry. Advance access published August 10, 2005, http:// oxford journals.org.

Ludvigson, Sydney. 2004. Consumer confidence and consumer spending. Journal of Economic Perspectives 18:29-50.

Mishkin, Frederic. 1978. Consumer sentiment and spending on durable goods. Brooking Papers on Economic Activity 1:38-48.

Mueller, Eva. 1963. Ten years of consumer attitude surveys: Their forecasting record. Journal of the American Statistical Association 53:899-917.

Tobin, James. 1959. On the predictive value of consumer intentions

and attitudes. Review of Economics and Statistics 41:1-11.

(1) Early studies for the United States include Klein and Lansing (1955), Tobin (1959), and Mueller (1963). For further details on these pioneering works, see Curtin (2000).

(2) The CCI is published by the European Commission for 14 EU countries, namely Austria, Belgium, Denmark, Finland, France, Ireland, Italy, Germany, Greece, Portugal, Spain, Sweden, the Netherlands, and the United Kingdom. The CCI survey for Luxemburg is not published.

(3) According to a recent report released by the European Commission, the confidence indicator for the service sector is not included in the construction of the ESI because of its inability to lead the business cycle.

(4) The questions used to derive the other three confidence indicators are provided in Appendix A.

(5) This estimation procedure amounts to a simple test of Hall's (1978) random-walk hypothesis, namely, if the [beta]'s are significantly different from zero, the said hypothesis is rejected.

(6) The close economic relationships among EU member countries imply that changes in one country's consumer spending could affect the exports of other EU nations to this country. We would like to thank an anonymous referee for bringing this point to our attention.

(7) As they are the largest economies in the EU, we include the CCI of Germany, France, and the United Kingdom in the prediction equations of each of the other six EU countries examined to control for intercountry effects. As for Germany, France, and the United Kingdom, we include the CCI of the other two trade partners in each country's respective prediction equation. For example, we include the CCI of Germany and France in the prediction equation of the United Kingdom.

(8) According to the information provided by the European Commission, the consumer surveys conducted for Austria, Finland, and Sweden only begin in the mid-1990s. As for Greece and Ireland, data on the total PCE of these two countries are only available in annual form.

(9) Because these confidence data are available only in monthly form, we convert them into quarterly observations by taking the average of the monthly observations.

(10) Denmark, Germany, and the Netherlands are the only three countries that fail to exhibit a contemporaneous relationship between real total PCE growth and the ESI.

(11) Like the lag structure of the control variables, we include four lags of the confidence indicators in Equation 1. The results of the Akaike (1973, 1974) information criterion (AIC) suggest that there is no need to include more than four quarterly lags in the model. These results are available on request.

(12) It is important to mention that the validity of our prediction equations is subject to vigorous diagnostic testing. The results of our diagnostic tests, which include the Godfrey LM test for first- and fourth-order serial correlation, the Engle test for first- and fourth-order autoregressive conditional heteroscedasticity, the White test for heteroscedasticity, the Bera-Jacque test for normality, and the Ramsey test for model misspecification, indicate that the reduced-form equation does not reveal obvious model inadequacy. This set of results is available on request.

(13) We note that the results obtained in the case of France and the United Kingdom concur with those of single-country studies conducted by Acemoglu and Scott (1994), Belessiotis (1996), and Delorme, Kamerschen, and Voeks (2002).

(14) Although QFPE is able to track future changes in Portugal's real total PCE, its self-predictive value, as measured by the incremental [[bar.R].sup.2], is relatively small (2.3%).

(15) For example, in the case of Portugal, QFPE exhibits the highest incremental [[bar.R].sup.2]. However, its predictive value is merely 1.1%.

(16) The technical description of the aggregate CCI can be found in the User Guide of the Harmonized Business and Consumer Surveys (2004).

(17) We would like to thank two anonymous referees for directing our attention to this point.

(18) These four countries are Belgium, France, Spain, and the United Kingdom (see Table 4).

(19) We choose 1999:I as the starting point of the out-of-sample period because it coincides with the birth of EURO in that year.

(20) A detailed description of the MSE-T test statistic is given in Appendix B.

(21) The simulation results presented in Clark and McCracken's study indicate that in commonly used sample sizes, the "bootstrap" CM test possesses proper size and good power behavior.

(22) Following CM's suggestion, we use the bootstrap approach to conduct statistical inferences.

John A. Cotsomitis, Canadian Center for Research in Economics, Toronto, Ontario, M4J 2X8 Canada; E-mail john.cotsomitis@sympatico.ca.

Andy C. C. Kwan, Department of Economics, The Chinese University of Hong Kong, Shatin, New Territories, Hong Kong, SAR; E-mail kwan1882@cuhk.edu.hk; corresponding author.

Table 1. Component Questions of the CCI and Component Indices of the ESI Questions Answers (a) CCI and 4 Component Questions 1. QFPE (Financial Position A lot better (PP), a little Expectation) How do you think the better (P), the same (E), a financial position of your little worse (N), a lot worse household will change over the (NN), and don't know (NA). next 12 months? 2. QESE (Economic Situation A lot better (PP), a little Expectation) How do you think the better (P), the same (E), a general economic situation in little worse (N), a lot worse this country will develop over (NN), and don't know (NA). the next 12 months? 3. QUE (Unemployment Expectation) Increase sharply (NN), increase How do you think the level of slightly (N), remain the same unemployment in the country will (E), fall slightly (P), fall change over the next 12 months? sharply (PP), and don't know (NA). 4. QSE (Savings Expectation). Very likely (PP), fairly likely Over the next 12 months, how (P), fairly unlikely (N), very likely are you to be able to unlikely (NN), and don't know save any money? (NA). The CCI is the arithmetic average of the scores on Questions 1-4 (b) ESI ESI = 40% X Industrial Confidence Indicator + 20% X (Consumer Confidence Indicator + Construction Confidence Indicator + Retail Trade Confidence Indicator) Table 2. Contemporaneous Correlation between Growth in Total PCE and Confidence Indicators Row Country CCI ESI 1 Belgium 0.225# (0.095) 0.380# (0.004) 2 Denmark -0.023 (0.867) -0.039 (0.781) 3 France 0.430# (0.001) 0.421# (0.001) 4 Germany 0.131 (0.336) -0.039 (0.781) 5 Italy 0.293# (0.029) 0.391# (0.003) 6 The Netherlands 0.035 (0.801) 0.063 (0.650) 7 Portugal 0.301# (0.027) 0.287# (0.035) 8 Spain 0.218 (0.107) 0.244# (0.070) 9 UK 0.476# (0.000) 0.491# (0.000) Numbers in parentheses are p-values. Figures in boldface indicate significance at least at 10% level. Note: Figures in boldface indicated with #. Table 3. Predictive Power of Various Confidence Indicators (Incremental [[bar.R].sup.2]'s from Equation 1) Row Country CO QFPE QESE 1 Belgium -0.001 0.005 -0.006 (0.316) (0.305) (0.205) 2 Denmark -0.063 -0.030 0.013 (0.788) (0.163) (0.134) 3 France 0.145# 0.106# 0.143# (0.000) (0.000) (0.000) 4 Germany -0.042 -0.035 -0.028 (0.634) (0.189) (0.106) 5 Italy 0.022 0.060 -0.005 (0.203) (0.253) (0.396) 6 The Netherlands -0.075 -0.055 -0.060 (0.883) (0.903) (0.626) 7 Portugal -0.001 0.023# 0.008# (0.110) (0.002) (0.013) 8 Spain 0.019# 0.107# 0.056# (0.070) (0.001) (0.021) 9 UK 0.192# 0.224# 0.157# (0.054) (0.003) (0.043) Row Country QUE QSE ESI 1 Belgium -0.018 0.003# 0.080# (0.301) (0.044) (0.090) 2 Denmark -0.075 -0.080 0.105# (0.897) (0.965) (0.000) 3 France 0.111# 0.033# 0.117# (0.000) (0.007) (0.000) 4 Germany -0.035 -0.032 0.142 (0.191) (0.157) (0.143) 5 Italy -0.004 0.086 0.185# (0.264) (0.120) (0.095) 6 The Netherlands -0.023 -0.050 -0.066 (0.558) (0.926) (0.866) 7 Portugal -0.041 0.136 -0.019 (0.452) (0.136) (0.375) 8 Spain 0.027# 0.125# 0.011 (0.046) (0.010) (0.291) 9 UK 0.136 0.150# 0.199# (0.118) (0.088) (0.073) The numbers in parentheses are the p-values for the joint significance of the lags of the confidence indicator. Figures in boldface indicate significance at least at 10% level. Hypothesis tests are conducted using a heteroscedasticity and serial correlation robust covariance matrix (allowing serial correlation at lags up to 4). Note: Figures in boldface indicated with #. Table 4. Predictive Power of Various Confidence Indicators (Incremental [[bar.R].sup.2]'s from Equation 2) Row Country CCI QFPE QESE 1 Belgium -0.071 -0.129 0.032# (0.112) (0.123) (0.000) 2 Denmark 0.002 -0.011 0.014# (0.117) (0.124) (0.000) 3 France -0.061 -0.065 0.064# (0.141) (0.156) (0.064) 4 Germany -0.152 0.071# -0.137 (0.289) (0.001) (0.127) 5 Italy -0.027 -0.025 -0.029 (0.123) (0.160) (0.109) 6 The Netherlands 0.004# 0.009# 0.003# (0.000) (0.000) (0.000) 7 Portugal 0.014# 0.011# 0.009# (0.000) (0.000) (0.000) 8 Spain 0.128# -0.029 -0.038 (0.000) (0.111) (0.112) 9 UK 0.118# 0.144# 0.127# (0.000) (0.000) (0.000) Row Country QUE QSE ESI 1 Belgium -0.05 -0.012 -0.047 (0.126) (0.102) (0.152) 2 Denmark -0.002 -0.045 0.006# (0.104) (0.201) (0.000) 3 France -0.052 0.006# -0.070 (0.104) (0.001) (0.167) 4 Germany 0.075# (0.100# 0.283# (0.030) (0.002) (0.000) 5 Italy 0.002# -0.039 -0.028 (0.000) (0.136) (0.128) 6 The Netherlands 0.003# 0.013# 0.001# (0.000) (0.000) (0.000) 7 Portugal -0.051 -0.051 0.017# (0.119) (0.128) (0.000) 8 Spain 0.036# 0.305# 0.146# (0.000) (0.000) (0.000) 9 UK 0.038 0.164# 0.000# (0.000) (0.000) (0.011) The numbers in parentheses are the p-values for the joint significance of the lags of the confidence indicator. Figures in boldface indicate significance at least at 10% level. Hypothesis tests are conducted using a heteroscedasticity and serial correlation robust covariance matrix (allowing serial correlation at lags up to 4). Note: Figures in boldface indicated with #. Table 5. Incremental Predictive Power of Aggregate CCI (Incremental [[bar.R].sup.2]'s from Equation 2) Row Country [CCI.sup.EU] 1 Belgium 0.158# (0.000) 2 Denmark -0.012 (0.102) 3 France 0.103# (0.000) 4 Germany -0.087 (0.146) 5 Italy -0.051 (0.156) 6 The Netherlands 0.000# (0.000) 7 Portugal 0.150# (0.000) 8 Spain 0.078# (0.000) 9 UK 0.213# (0.000) The numbers in parentheses are the p-values for the joint significance of the lags of the confidence indicator. Figures in boldface indicate significance at least at 10% level. Hypothesis test are conducted using a heteroscedasticity and serial correlation robust covariance matrix (allowing serial correlation at lags up to 4). Note: Figures in boldface indicates with #. Table 6. Out-of-Sample Predictive Power of One-Step-Ahead Forecasts for Real Total PCE Row Country CO [CCI.sup.EU] ESI QFPE 1 Belgium -0.172 -1.134 -0.517 -2.173 2 Denmark -1.038 -0.969 -1.355 -2.139 3 France -2.100 -1.030 -1.429 -1.121 4 Germany -1.969 -1.860 0.608# 1.696# (0.060) (0.008) 5 Italy -1.078 -1.874 -1.360 -1.047 6 The Netherlands -0.030 -0.671 -1.094 -0.611 7 Portugal -0.724 0.598# -0.768 -0.748 (0.037) 8 Spain -0.493 -1.194 0.363# -1.938 (0.043) 9 UK -1.289 0.752# -1.993 -1.197 (0.026) Row Country QESE QUE QSE 1 Belgium 0.285# -0.286 -1.779 (0.077) 2 Denmark -1.307 -1.947 -1.256 3 France -2.441 -2.096 -1.484 4 Germany -2.481 -1.471 -1.590 5 Italy -1.010 -1.454 -1.766 6 The Netherlands -0.269 -1.005 -0.199 7 Portugal -0.645 -1.822 -1.120 8 Spain -1.682 -1.255 -0.003 9 UK -0.631 -1.632 -1.345 The numbers in parentheses give the significance level estimated with the bootstrapped simulations. Figure in boldface indicate significance at least at 10% level. Note: Figure in boldface indicated with #.

Printer friendly Cite/link Email Feedback | |

Author: | Kwan, Andy C.C. |
---|---|

Publication: | Southern Economic Journal |

Geographic Code: | 1USA |

Date: | Jan 1, 2006 |

Words: | 7413 |

Previous Article: | Finance and income inequality: What do the data tell us? |

Next Article: | Export promotion through exchange rate changes: exchange rate depreciation or stabilization? |

Topics: |