Using consumer confidence to forecast private consumption : does this relationship hold for the Dutch economy for the period 1995-2014?

Using consumer confidence to forecast private consumption

Does this relationship hold for the Dutch economy for the period 1995-2014?

Yu-Cheng Yang 10339892

29th _{of June 2016}

Thesis BSc Business Economics Supervisor: Dr. K. Vermeylen

2

Statement of Originality

This document is written by Student Yu-Cheng Yang who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

3

Abstract

This thesis investigates the ability of the consumer confidence index to forecast private consumption in The Netherlands for the period 1995-2014. Consumption is regressed on lagged values of consumer confidence indices and lagged values of consumption itself. The results indicate that consumer confidence index does help forecast private consumption in The Netherlands for the period 1995-2014 and can provide useful information in making assessments of current economic conditions.

4 Index 1. Introduction 4 2. Theoretical framework 6 2.1 Consumer confidence 6 2.2 Literature review 7

3. Data and methodology 9

3.1 Data 9 3.2 Variables 9 3.3 Descriptive statistics 9 3.4 Methodology 10 4. Results 12 5. Conclusion 15 References 16 Appendices 18

5

1. Introduction

Consumer confidence indices receive much attention by many agents, including

governments at the national level, central banks and entities such as the European Union. Consumer confidence indices indicate the level to which households think that the economy is doing better or worse and are seen as crucial information for the current and future state of the economy. It is based on the financial condition of households and their sentiments about the economic climate in general. During recessions these indices are closely tracked as a significant positive change may indicate economic recovery in the near future

(Vuchelen, 2004).

Previous research has studied the link between consumer confidence and consumer spending. However this has proven to be inconsistent. Bram and Ludvigson (1998) and Carroll, Fuhrer and Wilcox (1994) have found evidence of explanatory power in forecasting consumption using consumer confidence indices in the United States. On the other hand, Fan and Wong (1998) have suggested with their results that consumer confidence contain no predictive information about household spending in Hong Kong.

As there has been limited research done on this area for the Dutch economy and private consumption in The Netherlands accounted for 44.81 per cent of GDP in The

Netherlands in 2014 (CBS), it is therefore interesting to investigate whether the relationship between consumer confidence and consumer spending also holds for the Dutch economy. Moreover, most research has been done in the earlier years of this century and therefore does not take in account the impact of the financial crisis. This has led to the research question: ‘Does consumer confidence forecast household spending in The Netherlands for the period 1995-2014?’.

To answer the research question, an empirical study will be conducted using a dataset of consumer confidence index and private consumption conducted by CBS. The dataset covers the consumer confidence indices and private consumption for the period from 1995Q3-2014Q4. To test whether consumer confidence indices help predict household spending in The Netherlands between the period 1995-2014, the Granger causality test will be conducted.

6 This thesis proceeds as follows. In section 2 some main concepts are explained and previous literature will be discussed regarding the predictive power of consumer confidence indices. Next in section 3 the data and methodology will be discussed. The results of the data are presented in section 4. And finally, the conclusions from this thesis will be drawn in section 5.

7

2. Theoretical framework Consumer confidence

A theory regarding the consumer confidence is the life-cycle and permanent income hypothesis. This theory implies that consumers’ decisions depend on the expectations of their future incomes. When consumers are positive about the future, they will consume more and save less than when they are less positive about their future. Thus measurement of consumer confidence are quite useful in forecasting private consumption (Mankiw, 2013).

Katona (1974) claims that consumer behavior was viewed as an insignificant factor of income fluctuations in the late 1950’s. Private spending were assumed to be dependent on income of activities from business and government sectors that are responsible for changes in the economy. Nowadays consumption is not only dependent on the ability to buy but also on the consumer confidence or the willingness to buy. After the World War II, the first survey were introduced and measures of both consumer behavior and expectations were then developed. Other countries used this method to measure the consumer confidence in their country.

The measurement of the consumer confidence index in The Netherlands are computed by the Central Bureau of Statistics (CBS). Since 1972, CBS measures consumers’ expectations on the general economic development and the financial situation of their household with the aim to obtain an early indication on trend changes in private

consumption. In the period 1972-1983 the surveys were taken three to four times annually with the use of questionnaires. From January 1984 data were collected each quarter by interviews by phone. As of April 1986 onwards these surveys are taken on a monthly basis by one thousand households in The Netherlands and published (Jansen, 2013).

Since the start of the survey in 1972 Dutch households have consistently been asked the same questions. The survey includes questions about consumers’ expectations with respect to saving behavior, inflation, unemployment and prospective consumption of Dutch households. The first three questions give an indication on the willingness to consume while the last two questions give an indication about the households’ sentiments on the economic climate (Jansen, 2013). Appendix A shows the questions asked in the survey.

8 These questions can be answered with positively indicating the situation has

improved, negatively i.e. the situation has deteriorated or neutrally signifying that the situation has remained unchanged. Each question is classified in two parts: percentage of positive answers and percentage of negative answers. The value of each question can be calculated by subtracting the percentage of negative answers from the percentage of positive answers. As each question is given the same weight, the weighted average of the willingness to buy and the economic climate is the consumer confidence index (Jansen, 2013).

Literature review

There have been several researchers that investigated whether consumer confidence indices contain predictive information to forecast consumption of a nation. The results however were rather divided.

Bram and Ludvigson (1998) tested whether consumer confidence indices in the United States have any power on its own to predict future changes in consumption spending. They compare two widely followed measures of consumer confidence in the United States, the University of Michigan’s Consumer Sentiment Index and the Conference Board’s Consumer Confidence Index. Using a dataset covering the period 1967Q1-1996Q3, they find that measures from the Conference Board have both statistically and economically significant predictive power for several spending categories, including total personal

consumption expenditures, when other economic variables are known. By including the last four quarter of data from the Conference Board’s Consumer Confidence Index in their equation has resulted in an increase of nine percent in the variation predicted in next period’s consumption growth. While the measures from the University of Michigan’s Survey Research turned out to have a weaker forecasting power for consumer spending. By adding four quarterly lags of the Michigan expectations component resulted in an increase of just three percent. They state that the key differences between these two indices are explained by the specific questions asked in the survey, the sample size and the survey methodology. This is however out of scope for this thesis.

9 The methodology used by Bram and Ludvigson (1998) are inspired by the work of Carroll, Fuhrer and Wilcox (1994). They investigated the predictive power of one of the consumer confidence index in the United States, the University of Michigan’s Index of Consumer Sentiment. Using quarterly data of this index with four lags together with quarterly real private consumption growth for the period 1978-1993, they find that lagged values of this index explain about fourteen percent of one-quarter-ahead real private consumption growth. Thus concluding that consumer sentiment index forecast future changes in household spending. How much the consumer confidence index help predict future changes in household spending proved however to be ambiguous.

Researchers in countries outside the United States, such as Kwan and Cotsomitis (2006) and Berg and Bergström (1996) ought to investigate this relationship whether consumer confidence is a dependable predictor of household spending in Canada and Sweden respectively. Kwan and Costomitis (2006) used a sample covering the period 1979Q4–2001Q4 and the methodology of Carroll et al. (1994). Berg and Bergström (1996) have analyzed this for the period 1975-1994. To investigate this link, they applied the Granger causality test where four lags were added into their equation. The results of Kwan and Cotsomitis (2006) and Berg and Bergström (1996) are consistent with those of Carroll et al. (1994) and Bram and Ludvigson (1998), both researches indicate that the consumer confidence index in their country is a reliable predictor of total household spending.

There has also been researchers such as Fan and Wong (1998) and Chopin and Darrat (2000) that have found little or no evidence that consumer confidence indices are helpful to forecast private consumption. Fan and Wong (1998) examined this using

consumer confidence indices in Hong Kong while Chopin and Darrat (2000) studied this for the United States. They suggest that the value of consumer sentiment adds little value in forecasting consumption spending that is not already captured by economic fundamentals.

10

3. Data and methodology Data

The variables, private consumption and consumer confidence, are used to test whether consumer confidence forecast private consumption. Data of consumer confidence index and private consumption are derived from the database of CBS, Statline. These data are

quarterly data spanning over the period from the third quarter of 1995 to the fourth quarter of 2014. The sample data consist of 78 periods.

Variables

Consumer confidence index (hereby 𝐶𝐶)

Data of consumer confidence index conducted by CBS are seasonally adjusted. By carrying out seasonal adjustments, trend-related fluctuations are more evident in the data.

Consumers are relatively more optimistic in spring and summer than in the rest of the year. Adjusting the data seasonally will make it easier to compare the results between

consecutive months (Van Velzen, Wekker and Ouwehand (2011)).

Real private consumption growth (hereby 𝐶)

Data of real private consumption are reflected in the account households and non-profit institutions serving households (NPISHs). These data are seasonally and calendar adjusted and are expressed in millions of euros. By carrying out seasonal and calendar adjustments, trend-related fluctuations are more evident in the data. (Van Velzen et al. (2011)). Moreover the growth (relative to the previous quarter) of real private consumption is taken to make the real private consumption a stationary variable. The growth of real private consumption is expressed in percentage.

Descriptive statistics

Before analyzing whether consumer confidence forecast private consumption, it is

interesting to first analyze some descriptive data. Appendix B shows the graphs of quarterly real private consumption growth in percentage and the quarterly consumer confidence index in The Netherlands for the period 1995Q3-2015Q3. As in line with Carroll et al (1994),

11 the correlation between the growth of spending and the sentiment is positive. For this sample a correlation of 0.65 is found. Carroll et al. (1994) state that when economic

conditions are poor, households restrict their spending and give less positive answers to the interview questions.

In the second quarter of 2009, total household spending relative to the previous quarter decreased by 1.6 per cent, an all-time low level for this sample data. The first quarter of 2000 shows an increase in growth of household spending relative to the previous quarter of 3.5 per cent, a record high for this sample data. CBS (2000) indicates that this record high increase is explained due to a large part of household purchasing a car.

In the fourth quarter of 2012 consumers were least optimistic about their financial condition of households and sentiments towards the economic climate with a consumer confidence index of -40. For the first and second quarter consumers were quite optimistic indicating a consumer confidence index of 26.

Methodology

As previously mentioned, to test whether consumer confidence indices contain any

predictive information about household spending in The Netherlands the Granger causality test will be conducted.

Before setting up the model, it is needed to determine how many lags should be included in the model. This can be done with Akaike information criterion (AIC). With this criterion measurement among time series data can be made to determine how many lagged values of the dependent variable should be included in the model to find the best measure of fit. The smallest value of the AIC will determine the number of lags to be included in the model to give the best measure of fit (Stock and Watson, 2012).

After finding the amount of lags to be included in the model, ordinary least squares (OLS) can be used to determine the values of the coefficients of the model. After this the Granger causality test can be applied. An important assumption for the Granger causality test is that the dependent variable (i.e.

𝐶

) and independentvariable (i.e.

𝐶𝐶

) are stationary variables. This is the why the growth of real private consumption have been taken instead of the real private consumption. Furthermore according to Granger (1969), variable CC is said to Granger-cause variable C, if CC can provide statistically significant information about

12 future values of C. The reversed granger causality test is also applied to determine whether variable C is said to Granger-cause variable CC. The following models are applicable to the needs for this thesis:

Granger causality model:

𝐶

_𝑡

= 𝛼 + 𝛽 ∑

𝑛_𝑡=1

𝐶

_𝑡−𝑖

+

𝛾 ∑

𝑛_𝑡=1

𝐶𝐶

_𝑡−𝑖

+ 𝜀

_𝑡

(1)

Reversed Granger causality model:

𝐶𝐶

_𝑡

= 𝛿 + 𝜎 ∑

𝑛_𝑡=1

𝐶𝐶

_𝑡−𝑖

+

𝜇 ∑

𝑛_𝑡=1

𝐶

_𝑡−𝑖

+ 𝜑

_𝑡

(2)

where

𝐶

_𝑡 is the real private consumption growth at time t,

𝐶𝐶

_𝑡 the consumer confidence index at time t and

𝜀

_𝑡

and 𝜑

𝑡the error terms at time t. Parameters

𝛼

and

𝛿

are

13

4. Results

With the use of the Akaike information criterion, the number of lags to be included in the model is found. Table 1 shows that three lags should be included in the model for the best measure of fit for both model (1) and (2). The lowest value of the AIC is found in the third lag. Lags AIC 0 2.597 1 2.338 2 2.254 3 2.238 4 2.266 5 2.295 6 2.321 7 2.311 8 2.337 Number of observations 70

Table 1: AIC values.

After determining the amount of lags, the values of the coefficient of model (1) can be found using OLS. Table 2 presents the values of the coefficients for model (1). Model (1) accounts for 46.70 per cent of the variance as shown in table 2.

Dependent variable: Consumption

Table 2: Coefficient values of model (1).

Variable Coefficient Standard error

𝛼

0.582 0.203 𝐶_𝑡−1 0.105 0.113 𝐶_𝑡−2 0.161 0.111 𝐶𝑡−3 0.178 0.110 𝐶𝐶𝑡−1 0.020 0.012 𝐶𝐶_𝑡−2 0.007 0.018 𝐶𝐶_𝑡−3 -0.012 0.012 Number of observations R-squared 75 0.4670

14 Now that the values of coefficients of model (1) are determined, the Granger

causality can be applied. The F-statistic testing the joint significance of the lags of the consumer confidence indices are jointly significant at a confidence level of five per cent as reflected in the p-value. From the Granger causality test, a p-value of 0.007 is found for the lags of consumer confidence index (

𝐶𝐶).

This indicating that the results show that

consumer confidence indices help forecast private consumption. However, the F-statistic testing the joint significance of the lags of private consumption is shown to be statistically insignificant at a significant level of five per cent. A p-value of 0.778 is found.

Table 3: P-values of joint significance of model (1).

The values of the coefficients for model (2) are conducted in the same way as that of model (1). This is shown in table 4. The regression model (2) accounts for 89.60 per cent of the variance as shown in table 4.

Dependent variable: Consumer confidence index

Table 4: Coefficient values of model (2).

Dependable variable Lags CC Lags C

𝑪 0.007 0.778

Variable Coefficient Standard error

𝛿 -2.735 2.001 𝐶𝐶𝑡−1 1.151 0.115 𝐶𝐶_𝑡−2 -0.076 0.176 𝐶𝐶_𝑡−3 -0.223 0.118 𝐶_𝑡−1 0.746 1.116 𝐶𝑡−2 0.721 1.093 𝐶_𝑡−3 0.123 1.091 Number of observations R-squared 75 0.8960

15 The joint significance of model (2) is shown in table 5. The F-statistic testing the joint significance of the lags of consumption is shown to be statistically insignificant at a

significant level of five per cent. A p-value of 0.778 is found, indicating that the reverse Granger causality does not hold.

Table 5: P-values of joint significance of model (2).

The results shows that consumer confidence indices in The Netherlands help forecast household spending in The Netherlands for the period 1995-2014. Consumer confidence can thus provide useful information in making assessments of current economic conditions.This is in line with the work of Carroll, Fuhrer and Wilcox (1994), Bram and Ludvigson (1998) and researchers in countries outside the United States, such as Kwan and Cotsomitis (2006) and Berg and Bergström (1996).

Dependable variable Lags C Lags CC

16

5. Conclusion

This thesis ought to answer the question whether consumer confidence forecast household spending in The Netherlands for the period 1995-2014. Consumer confidence indices

receive much attention by many agents, including governments at the national level, central banks and entities such as the European Union. Consumer confidence indices indicate the level to which households think that the economy is doing better or worse and are seen as crucial information for the current and future state of the economy (Vuchelen, 2004).

There have been several researchers that investigated whether consumer confidence indices contain predictive information to forecast consumption of a nation. The results however were rather divided. This thesis focus on whether consumer confidence indices help predict household spending in The Netherlands between the period 1995-2014, with the help of Granger causality test. The measurement of the consumer confidence index and private consumption in The Netherlands are computed by the Central Bureau of

Statistics (CBS). A dataset consisting of quarterly data spanning over the period from the third quarter of 1995 to the fourth quarter of 2014 has been used.

Consumption is regressed on lagged values of consumer confidence and lagged values of consumption itself similar to the work of Carroll et al (1994). After applying the Granger causality test, the results provide evidence that consumer confidence indices help forecast the change in consumption for the Dutch economy in the period 1995-2014. This is in line with the work of Carroll, Fuhrer and Wilcox (1994), Bram and Ludvigson (1998) and researchers in countries outside the United States, such as Kwan and Cotsomitis (2006) and Berg and Bergström (1996). They all investigated the predictive power of the consumer confidence index in their countries and found that consumer sentiment index forecast future changes in household spending. The reverse Granger causality however does not hold indicating that consumption does not help forecast consumer confidence.

17

Literature

Berg, L., & Bergstrom, R. (1996). Consumer confidence and consumption in Sweden. Dept. of Econ., Uppsala University WP, 7.

Bram, J., & Ludvigson, S. C. (1998). Does consumer confidence forecast household expenditure? A sentiment index horse race. Economic Policy Review, 4(2).

Carroll, C. D., Fuhrer, J. C., & Wilcox, D. W. (1994). Does consumer sentiment forecast household spending? If so, why?. The American Economic Review, 84(5), 1397-1408.

CBS. (25 mei 2000). Groei consumptie trekt aan. (CBS). Retrieved from: https://www.cbs.nl/NR/rdonlyres/A5E3F4EA-FAF7-42F8-BDA1 30A99E8C52CC/0/pb00n119.pdf.

Chopin, M. C., & Darrat, A. F. (2000). Can consumer attitudes forecast the macroeconomy?. The American Economist, 44(1), 34-42.

Fan, C. S., & Wong, P. (1998). Does consumer sentiment forecast household spending?: The Hong Kong case. Economics Letters, 58(1), 77-84.

Granger, C. W. J. (1969). Investigating causal relations by econometric models and cross spectral methods. Econometrica, 37, 424–438.

Jansen, M. (2003). Consumentenvertrouwen als indicatie voor de toekomstige particuliere consumptie (Consumer Confidence as an Indicator of Future Private Consumption), CBS (Statistics Netherlands). Divisie Macro-Economische Statistieken en Publicaties.

18 Kwan, A. C., & Cotsomitis, J. A. (2006). The usefulness of consumer confidence in forecasting

household spending in Canada: A national and regional analysis. Economic Inquiry, 44(1), 185-197.

Mankiw, G. (2013). Macroeconomics. 8th edition. Houndmills: Palgrave Macmillan.

Stock, J.H., & Watson, M.M. (2012). Introduction to Econometrics. 3th edition. Harlow: Pearson Education Limited.

Van Velzen, M., Wekker, R., & Ouwehand, P. (2011). Method series, seasonal adjustment. CBS (Statistics Netherlands). Divisie Macro-Economische Statistieken en Publicaties.

Vuchelen, J. (2004). Consumer sentiment and macroeconomic forecasts. Journal of economic psychology, 25(4), 493-506.

19

Appendix A

1. In your belief, has the financial situation of your household improved, deteriorated or remained the same in the past twelve months?

2. How do you appraise the financial situation of your own household in the next twelve months?

3. Do you think it is the right or wrong time to buy costly items like furniture, a washing machine or a television set?

4. Do you think that the general economic situation in The Netherlands has improved, deteriorated or remained the same in the past twelve months? 5. What is your opinion on the general economic situation for the next twelve months?

20 Appendix B

Graph 1. Quarterly consumer confidence indices in The Netherlands (1995Q3-2015Q3).

Graph 2. Quarterly real private consumption growth in The Netherlands (1995Q3-2015Q3). -50 -40 -30 -20 -10 0 10 20 30 1995Q 3 1996Q 3 1997Q 3 1998Q 3 1999Q 3 2000Q 3 2001Q 3 2002Q 3 2003Q 3 20 04 Q 3 2005Q 3 2006Q 3 2007Q 3 2008Q 3 2009Q 3 2010Q 3 2011Q 3 2012Q 3 2013Q 3 2014Q 3 20 15 Q 3

Consumer confidence index

-2.0% -1.0% 0.0% 1.0% 2.0% 3.0% 4.0% 1995Q 3 1996Q 3 1997Q 3 1998Q 3 1999Q 3 2000Q 3 2001Q 3 2002Q 3 2003Q 3 20 04 Q 3 2005Q 3 2006Q 3 2007Q 3 2008Q 3 2009Q 3 2010Q 3 2011Q 3 2012Q 3 2013Q 3 20 14 Q 3 2015Q 3