Bachelor
Thesis
The influence of
deflation on
consumer confidence
By: Isabelle Dorhout, 10528938
Under supervision of: Singh Swapnil
Faculty of business and economics
School of economics
Statement of originality
This statement is written by UvA student Isabelle Dorhout, who takes full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references were used in creating it. The Economics and Business faculty is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
This paper explores the effect of deflation on consumer confidence. Research has been conducted on the influence of inflation on consumer confidence, but not on the influence of deflation. Deflation is the other side of the coin and is less common than inflation.
Nonetheless it is interesting to assess the effect of deflation, because it may have different implications than theory suggests. This paper looks at the influence of deflation on consumer confidence in Japan, using the Netherlands to verify the results. The time span is from 2001 until 2016. Moreover, because quantitative easing is related to deflation and affects
consumer confidence in Japan, its influence is also taken into account. The models are predicted using an ordinary least squares (OLS) regression and a fixed effects model. Both models are insignificant. This might be due to a lack of variables, bad model design, failure to properly take care of the time lags within the variables. Or the hypothesis is fulfilled and deflation has indeed no influence on consumer confidence. Further research on this topic is recommended.
Table of Contents
0. Introduction ... 3
1. Literature review ... 5
Deflation and inflation ... 6
Consumer confidence ... 6 Irrational behavior ... 7 Quantitative easing ... 8 2. Research design ... 9 Data ... 9 Hypothesis ... 11 Models... 11 OLS assumptions ... 12 3. Results ... 13 OLS assumptions ... 13 CPI ... 16 QE ... 16 Model 2 ... 17 4. Conclusion ... 18 5. Discussion ... 19 6. References ... 21
0. Introduction
Recently, the Netherlands and other European countries have been struggling with the zero bound of interest rates and a fear of deflation (CBS, 2016). Deflation can be hard to fight, as is seen in the case of Japan. Central banks try to fight it even with unconventional tools, such as quantitative easing, because deflation is assumed to be bad for the economy. Neoclassical theory predicts that because of deflation people postpone their expenditures, which results in a downwards spiral (see Figure 1 in the literature review). This theory suggests that deflation causes falling consumer confidence, but is this true? Postponing of expenditures is not only explained by falling consumer confidence, but also by people who think that their goods will be cheaper tomorrow, because they expect prices to drop. Expectancy about the deflation period is therefore also an important feature of postponing expenditures, on which consumer confidence might have an influence as well. Therefore, it is important to know whether deflation has an influence on consumer confidence.
Currently, in the Netherlands there has been anxiety of (and in June 2016 real) deflation and still consumer confidence has risen (CBS-a, 2016). In this paper I am
investigating whether this is normal or an exception, what the cause is of this and whether the theory mentioned above fails and whether consumer confidence is influenced by deflation. The question whether deflation is related to consumer confidence intrigues me.
Japan is used to investigate deflation, since it has been struggling with deflation for decades. How has Japanese consumer confidence changed through the long periods of deflation? How did quantitative easing, that began in March 2001 (Spiegel, 2001), affect consumer confidence? Because inflation and deflation are two sides of the same coin, the Netherlands is also used in this research, even though it has had no real deflation (except for June 2016). Consumer confidence differs as an indicator of consumer expenditures from deflation, because consumer confidence takes irrational thought into account as it is based on a survey that includes aspects as feelings and believes of people about the economy. Deflation does in principle not take irrational behavior into account, because it is mainly based on price levels.
Central bankers worry about the extremely low inflation rate in Europe, but is this concern justified? They worry that deflation causes less expenditures. Is quantitative easing the answer to lift inflation? Can consumer confidence save the economy from the negative results of deflation, which are caused by? If consumer confidence is not much influenced by deflation, more attention should be paid to consumer confidence rather than trying to rise inflation by monetary policies. Consumer confidence is a key indicator for expenditures (Acemoglu and Scott, 1994, Carroll et al., 1994, Eppright et al., 1998, Gelper et al., 2007, Huth et al., 1994, Kwan and Cotsomitis, 2006, Ludvigson, 2004 and Slacalek, 2004).
Therefore, it is important to know more about the influence of deflation on consumer confidence. This paper investigates how deflation influences consumer confidence. The research question is: What is the influence of deflation on consumer confidence in the
Netherlands and Japan looking at monthly data from 2001 until 2016?
To be able to measure this influence, this research uses an OLS regression from deflation on consumer confidence in Japan and in the Netherlands run by STATA with monthly data for 15 years, from 2001 to 2016, so it also includes the financial crisis of 2008. This regression will measure the influence of deflation on consumer confidence. This time span also includes the Japan’s quantitative easing program, which began in a first round in 2001 and ended in March 2006 (Voutsinas and Werner, 2011), then a second round in 2010 followed by a higher-level third round in 2013. The quantitative easing program will be taken into account in the models by a dummy variable. The way consumer confidence is measured in Japan differs from the Netherlands. To be able to compare this data, it has to be
standardized with index numbers, with 2015 as the baseline year. By doing this, it is not possible to look at level effects, but only to look at changes in consumer confidence.
Changes of consumer confidence are interesting, because they can help to predict the future state of the economy.
The rest of the paper is as follows. Section 1 presents a literature review. Section 2 follows with the research design and the econometric methodology by which the main
findings of this paper can be researched. The results are presented and discussed in section 3 and the conclusion of the main findings follows in section 4. Section 5 contains a
1. Literature review
This section provides an overview of the existing literature on the relationship between inflation and consumer confidence. It is not clear what the influence of economic variables on consumer confidence is and research on this topic are not always consistent, but overall there are good reasons to assume that real economic variables affect consumer confidence.
Economic theory suggests that deflation has an influence on consumer confidence in the following way:
Figure 1 (http://www.321gold.com/editorials/schoon/schoon010914.html)
First, deflation causes companies to have lower revenues and profits. Because of the lower demand, companies have to order less inputs. There is less production. Because of the rigid wages, companies have to fire people. This makes unemployment rise. More unemployment causes less spending on goods and services, which is induced by less income on one hand and less consumer confidence on the other. This causes an oversupply of goods and services and makes prices even lower, so deflation is a self-reinforcing
process.
Second, deflation causes prices of goods and services fall. People postpone their expenditures, because they assume prices will keep on falling and goods and services will be cheaper later. Expectations are important here. One aspect in the surveys of measuring consumer confidence is about postponing expenditures (CBS-b, 2005). For these reasons, deflation should influence consumer confidence negatively according to economic theory.
One way inflation might influence consumer confidence is higher employment. According to the Phillips curve on short term higher inflation gives higher output and higher employment (Matthews et al., 2013) and employment influences consumer confidence (Lemmon, 2006).
Deflation and inflation
Deflation and inflation are well-researched topics and there are a lot of theories on these topics. What drives inflation matters in general, but does not matter in this research, because inflation is given and thus is an exogenous variable in this research. This paper uses the word deflation with the definition of price deflation and falling prices and not the Austrian definition (Shostak, 2002) of monetary deflation, which is falling money growth (Bryan, 1997). These definitions are in the long run related, except when the inflation is low, in which case this effect of falling money growth is less on price deflation (Benati, 2009). But this paper looks at short-run, monthly, consumer price indexes (CPI) and therefore it is relevant to distinguish the two definitions of deflation.
Consumer confidence
Consumer confidence, however, is a lesser researched indicator than both deflation and inflation. Consumer confidence is more difficult to measure, because it is measured by surveys. It has therefore more risk of a selective bias than CPI measures. Nonetheless it is very important to assess the future state of the economy (Greenspan, 2002). The Dutch central bank (the DNB), knows that the Dutch consumer index has a high correlation (0,7 between 1983-2006 0,7 and nowadays above 0,8) with consumer expenditures and, therefore, explains a good part of consumer expenditures. Additionally, the DNB points out that consumer confidence predicts consumer expenditures with a time lag of six months. This is why the DNB uses consumer confidence as one of the key DNB business cycle indicators. One of the reasons for that is that consumer confidence can be taken into account for non-rational consumer behavior, especially in the short-term (van Oest & Franses, 2008). Consumer confidence gets more attention in times of recession, because positive changes may indicate an approaching economic recovery (Vuchelen, 2004). It also contains useful information about the probability of entering a recession (Howrey, 2001). Therefore, consumer confidence is an important indicator of the future state of the economy.
Besides expenditures, consumer confidence also has an influence on GDP. According to Utaka (2010), consumer confidence index (CCI) explains 9% to 11% of the variance in GDP eight months ahead in Japan. Matsusaka and Sbordone (1995) show that in the US economy, the variance of CCI explains about 13% to 26% of forecast variance of GNP eight quarters ahead. Therefore, CCI variance explains less of the output fluctuations
in Japan than in the US. This might be due to the bigger social security benefits in Japan, because if the social security benefits in a country are bigger, people worry less and have less insecurities about their income than with a smaller social security benefits or no benefits at all. Individual consumer confidence will react stronger to people’s personal situation and it may depend more on the perceived future economic situation (Van Oest and Franses, 2008). Larger social security benefits countries have less variance in GDP explained by CCI than a smaller social security benefits countries, because confidence becomes more
important for spending and predicting GDP. The social security benefits in Japan look more like those in the Netherlands than in the US. In percentage of GDP, social expenditures are 23% in Japan, 23% in the Netherlands and 19% in the US (OECD, 2013). However, the influence of CCI on GDP is not significant, but the reverse is. GDP has a positive influence on CCI (Utaka, 2010). Consumer confidence has a significant effect on economic
fluctuations in the short term (quarterly and monthly). CCI has no effect on economic fluctuations over longer time spans. One might conclude that confidence effects short-term economic fluctuations, but in the long term, it does not affect the real economy.
There are disagreements about the measurement and reporting of consumer
confidence as well as its relationship with the real economy (e.g., Ludvigson, 2004). Jansen (2003) confirms that the relationship between consumer confidence and the real economy is weak, contradicting Djerf and Taklaka (1985), who found that economic variables explain consumer confidence. They, conclude that the most popular surveys help predict future consumer expenditure, but they admit that the extra predictive power besides other economic indicators is modest. According to Ludvigson (2004), “The relation between consumer confidence and the real economy is still not completely understood and appears to be weak”.
Irrational behavior
Alsem et al. (2008) found a significant influence in the Netherlands of bias in newspapers reporting of economic variables on consumer confidence. This effect is, however, short-term. This is a good example of irrational behavior is influencing consumer confidence.
According to Acemoglu and Scott (1994), inflation has a significant effect on consumer confidence. Still, deflation might have a smaller or no effect, in contradiction to inflation, because of loss aversion. Loss aversion is an example of irrational behavior and means that losses have more impact on people than gains “losses loom larger than gains” (Kahneman & Tversky, 1979). Inflation can be seen as a loss of value of money and deflation as a gain.
Quantitative easing
Quantitative easing is a monetary policy of the central bank, where the bank buys a broad set of assets, such as government bonds, public assets and private assets, to increase the money supply in order to increase the inflation rate (Bank of England, 2011). Japan has experienced deflation and low interest rates for decades. Applying conventional monetary policies in Japan would decrease the interest rates further, which is not the desired effect. Therefore the central bank implemented the unconventional monetary policy of
quantitative easing. The policy of quantitative easing is rather new. Japan is the first country to use this policy in 2001 (Bordo, 2014). The effects of quantitate easing are not proven. Critics say quantitate easing is not effective. Still, because central banks trend to be more transparent (Eijffinger et all, 2006), the public knows about the quantitative easing policies and reacts to it. When Draghi, head of the ECB, stated he was implementing this policy, the economic agents reacted (Klaassen, 2016). So whether or not quantitative easing is
effective, it influences the beliefs of economic agents.
Van Oest and Franses (2008), who investigated changes in consumer confidence in the US and the Netherlands, noted that the consumer confidence explained more of the variance in spending in the US than in the Netherlands, likely due to higher social security in the Netherlands. To be able to compare the Netherlands with Japan, social security in those countries should be comparable. According to the OECD, both countries in 2013 spent 23% of their GDP on social expenditures.
2. Research design
This section describes the data used for this research and how it is collected,
presents the hypothesis, and discusses the models used in the research and their variables.
Data
Inflation
The monthly publications of annual inflation (CPI numbers) are obtained from CBS (the statistical office of the Netherlands) and the Statistics Bureau of Japan using
DataStream. The price indexes for both countries are not seasonally adjusted and 2015 is used as base year. The CPI is the most frequently used indicator of inflation and reflects changes in the cost of acquiring a fixed basket of goods and services by the average consumer. CPI is used for deflation rates, because it does not over- or understate the effect of deflation in the economy and CPI is assumed to be a good measure of changing prices and thus measuring inflation (Cagan and Moore, 1981). Food and energy prices are taken into account by the CPI data. These sectors drive deflation and create short term changes (Smaghi, 2009). Food and energy prices influence consumers a lot, because they have to buy these and cannot live without them, so food and energy prices are an important part of inflation that influences consumer confidence, because they have a direct impact.
The base year of the CPI is 2015 in which the CPI is thus 100. For Japan the most of the CPI data is below the hundred. The mean of the CPI is here 97,6 and the range is small, between 95,7 and 100,4. Despite all the efforts of the government there are no big changes in the inflation index. There is a falling in CPI from January 2001 until November 2003. After that it falls and rises from month to month, but not with big changes and no patterns can be found.
For the Netherlands the mean is lower namely 90,1 and the range is bigger, between 76,3 and 101. This might be due to the introduction of the euro in 2002. The Netherlands has no influence over its monetary policy. The ECB controls the money supply. Which might cause bigger changes. For the Netherlands you see a clear upward going trend in the CPI, which means they had a constant inflation.
Consumer confidence
The consumer confidence data from Japan is reported by the Japan Cabinet Office. Consumer confidence data in Netherlands is reported by the Statistics Netherlands (CBS). Both are available on DataStream. For consumer confidence, index rates are used with 2015 as the base year to make the data from Japan and the Netherlands comparable. Still the levels are different, so the betas of the regression models cannot be compared. This problem can be solved by looking at the growth rates. Unfortunately the CCI in the
Netherlands has an index number, so that it can be a negative value. Therefore, logs do not work and therefore a growth rate of the Dutch CCI cannot be taken and a comparison
between Japan and the Netherlands still cannot be made. Here also the rates are seasonally adjusted. The data from Japan covers 6,720 households that were sampled using a three-level stratified random sampling method of city (or town or village), local unit and household. The survey variables are (1) consumer perceptions, (2) price expectations and (3) state of the household. The data from Japan’s March 2011 survey is seasonally adjusted data of the monthly consumer confidence index and its four compositions, consumer perception
indexes, are released instead of the quarterly seasonally adjusted consumer confidence index (Japan Cabinet Office, 2016). So some data from Japan is missing, because monthly data of consumer confidence is only available from 2011.
The Netherlands bases its numbers on 1000 household telephone surveys. The survey variables are comparable with Japan’s regarding (1)
t
he general economic situation in the next twelve months, (2) the general economic situation in the previous twelve months, (3) the households’ own financial situation in the next twelve months, (4) the household’s own financial situation in the previous twelve months and (5) whether it is a favorable time to purchase expensive items (e.g., washing machines) (CBS-c).In the data of Japan you see that the base of the index is higher, so the CCI will not be a negative value. The mean is 41.1 and the range is between 27.5 and 50.1. The biggest change in CCI is between 2006, where the highest point of CCI is reached and 2009 in the recession where the lowest point of 27.5 is reached. Furthermore, the CCI is quite stable around 41.
As already mentioned the CCI in the Netherlands has a different level. It becomes negative from time to time. The mean is -14.7 and the range is from -44 to 22. The highest point is end 2000 with the internet bubble and the burst is probably one of the causes of a big drop in consumer confidence. Also the economic crisis had a big negative influence on the CCI. In 2013 there is again a big drop of consumer confidence, probably because of the falling and nationalization of two big Dutch banks ABN AMBRO and SNS real. Dutch
Quantitative easing dummy variable
To look at the effect of quantitative easing in Japan on consumer confidence, an extra dummy variable was created, which is set at 1 if there was an active QE program and a 0 if there was not. The dummy variable is 1 from March 2001 to June 2006 and again from October 2010 until December 2016, the end of the dataset. As mentioned before, the level effects of quantitative easing programs are not taken into account, although the last round of QE, which started in April 2013, is more aggressive than the previous ones (Andolfatto and Li Li, 2014).
Hypothesis
The literature shows that inflation affects consumer confidence, however, I expect that deflation has no effect on consumer confidence, because consumers are not that
forward-looking and rational as assumed by neoclassical economic theory and deflation has, therefore, less influence on consumers than the theory suggests. Additionally, deflation occurs less frequently than inflation, and so is less known (by the public) and might, therefore, have a weaker influence. Also, loss aversion can have influence here on consumer confidence.
𝐻0 : deflation (and quantitative easing) has no effect on consumer confidence adjustments (𝛽{1}= 𝛽{2} = 𝛽{3}= 𝛽{4} = 0).
𝐻1: one or more restrictions under 𝐻0 do not hold.
Models
The standard model 1: 𝐘𝐭 = 𝛂 + 𝛃{𝟏}𝐗{𝐭}+ 𝛃{𝟐}𝐗{𝐭−𝟏}+ 𝛃{𝟑}𝐗{𝐭−𝟐}+ 𝛜 that will be estimated for both countries. To ensure that all variables are significant and do not correlate, the model is tested in steps, adding an extra β in every step. So the first test of model 1 is 𝑌𝑡 = 𝛼 + 𝛽{1}𝑋{𝑡} + 𝜀 the second is 𝑌𝑡 = 𝛼 + 𝛽{1}𝑋{𝑡}+ 𝛽{2}𝑋{𝑡−1}+ 𝜖 and the third 𝑌𝑡 = 𝛼 + 𝛽{1}𝑋{𝑡}+ 𝛽{2}𝑋{𝑡−1}+ 𝛽{3}𝑋{𝑡−2}+ 𝜖. This is done, because β{1}, β{2} and β{3} are highly correlated, because inflation at time moment t is related to the inflation during the month before etc, so there will be multi-correlation. Additionally, every model is tested for omitted variables, to see whether adding a variable will take away an omitted variable bias.
Variables in the models
Y is the standardized consumer confidence, X{t} is the inflation at time t, X{t−1} is the inflation in the month before time t; X{t−2} is the inflation at two months before time t, and 𝞮 is the stochastic error.
Japan began the first round of quantitative easing in March 2001 and ended it in 2006 and began a second round in October 2010, which led to a bigger volume policy in 2013. Quantitative easing might have an influence on consumer confidence. An extra dummy variable is added to control for the effect of quantitative easing on consumer confidence in these months.
The extra Japan, model 1-QE: 𝑌𝑡 = 𝛼 + 𝛽{1}𝑋{𝑡}+ 𝛽{2}𝑋{𝑡−1}+ 𝛽{3}𝑋{𝑡−2}+ 𝛽{4}𝑄𝐸 + 𝜖, where QE is added to model 1, the dummy variable for the quantitative easing program of the central bank of Japan, which is 1 for the months this unconventional monetary policy tool is used and 0 for the months it is not.
The problem with this model is that as previously mentioned, β{1}, β{2} and β{3} are very correlated. Additionally, the dummy variable QE is related to the betas 1, 2 and 3 or and to inflation, because imperfect multi-collinearity means that two or more of the regressors are highly correlated in the sense that there is a linear function of the regressors that is highly correlated with another regressor.
Model 2:[ 𝒀{𝒕}− 𝒀{𝒕−𝟏}] = 𝜶 + 𝜷{𝟏}[𝑿{𝒕}− 𝑿{𝒕−𝟏}] + 𝜷{𝟐}𝑸𝑬 + 𝝐 ,
is used in the effort to overcome the multi-collinearity bias. Again, these tests were done step by step and multi-collinearity was tested for to see whether model 2 can improve on model 1. Model 2 is a fixed time model and does not take into account the changes in time.
OLS assumptions
To perform an OLS regression, the OLS assumptions have to be tested as well. Therefore, the normal distribution of the data and the heteroskedasticity of the errors were tested. Normality is tested because it cannot be assumed in this research. Furthermore, the condition of independently drawn observations, is questionable in this research, because it uses a time span and although the time span is randomly chosen, the data is from this time span and correlates with each other for especially the inflation data. The other conditions for the OLS estimator is that data is that large outliers are unlikely: X{i} (CPI) and Y{i} (CCI) and have nonzero finite fourth moments (Stock and Watson, 2015-a). The fourth moment has to be finite, because if it is infinite there is really much weight in the tails and much of the variance of Y arises from extreme values or outliers. In this is the case with the data in this research, because the data are mostly standardized by index numbers and there are no large outliers for that reason. A further condition is that the error term 𝟄 has a conditional mean zero given X, but with observational data, which is the case in this experiment, the issue is whether E ( ε|X = x) = 0 holds. That way the OLS estimator is consistent, efficient and linear. Adding the condition of homogeneity of the errors makes the OLS estimator BLUE, that is, Best Linear Unbiased Estimator (Stock and Watson, 2015-b).
3. Results
In this section, the results of this research are presented and discussed.
To do an OLS regression, the assumptions of an OLS have to be fulfilled. Therefore, tested is whether distribution of the data is normal. Furthermore, homoscedasticity of the error terms and whether the models have omitted variables is tested also.
OLS assumptions
Table 5 — Heterogeneity, normality and omitted variable (Japan)
Variable (1a) (b) (c) (QE) (2) (2QE)
Heterogeneity 0.7708 0.7371 0.5417 0.7297 0.0370 0.8264 Omitted variable 0.0001 0.0171 0.0078 0.3101 0.0000 0.8451 Normality Shapiro-Wilk test – – - 0.0004 - - 0.11558 JB test − – - - - 0.0662 Observations 164 164 164 164 151 151
Notes: The heterogeneity test is performed by a Breusch-Pagan/Cook-Weisberg test for heteroskedasticity with H0: constant
variance or homoskedasticity. The omitted variable test is performed with a Ramsey Reset test using powers of the fitted values of CCIJAP and in column (5) and (6) for diffCCIJAP. The normality Shapiro-Wilk test for normal data with a H0: normal
distributed data is performed. However, this test is biased by sample size. Therefore, a Jarque–Bera test (also known as a goodness-of-fit test) of whether sample data have the skewness and kurtosis matching a normal distribution is performed. The null hypothesis is a joint hypothesis of the skewness being zero and the excess kurtosis being zero. For small samples, the chi-squared approximation is overly sensitive, often rejecting the null hypothesis when it is true. Furthermore, the distribution of p-values departs from a uniform distribution and becomes a right-skewed uni-modal distribution, especially for small p-values. This leads to a large Type I error rate. So here, the alpha has to be corrected and 0.05 is with a sample size of 100 is now 0.062. So the data is normally distributed.
Table 6 — Heterogeneity, normality & omitted variable results (the Netherlands)
Model (1a) (b) (c) (2) Heterogeneity 0,9860 0.7871 0.7685 0.6179 Omitted variable 0.0000 0.0039 0.2247 0.5682 Normality Shapiro-Wilk test – – - 0.0004 0.18380 JB test - - - 0.293 Observations 191 191 191 191
Notes: the heterogeneity test is performed by a Breusch-Pagan/Cook-Weisberg test for heteroskedasticity with H0: constant
variance or homoskedasity. The omitted variable test is done with a Ramsey Reset test using powers of the fitted values of CCINL and in column (4) for diffCCINL. The normality Shapiro-Wilk test for normal data with a H0: normal distributed data.
However this test is biased by sample size. Therefore there is also a Jarque–Bera test is a goodness-of-fit test of whether sample data have the skewness and kurtosis matching a normal distribution. he null hypothesis is a joint hypothesis of the skewness being zero and the excess kurtosis being zero. For small samples the chi-squared approximation is overly sensitive, often rejecting the null hypothesis when it is true. Furthermore, the distribution of p-values departs from a uniform distribution and becomes a right-skewed uni-modal distribution, especially for small p-values. This leads to a large Type I error rate. So here the alpha has to be corrected and 0.05 is with a sample size of 100 is now 0.062. So the data is normal distributed.
Homoscedasticity
Stata uses homoscedastic errors as a standard. There is no reason to assume that the error terms are homoscedastic. The results in (Table 5&6 ) show that the most of the models have homoscedasticity error terms. To test this, the Breusch-Pagan / Cook-Weisberg test for heteroscedasticity is used with a H0: Constant variance. Most models show a p-value over 0.05 (5% significance level), which means H0 cannot be rejected. A
constant variance occurs if the errors are homoscedastic. If the H0 is rejected, as in model 1-QE of Japan, with the dummy variable of 1-QE and does not make usage of a time difference, this model has heteroscedastic errors and, therefore, is tested with a robust error. This test gives the same results. However, in this model there is an omitted variable, which can be the reason for the heteroscedasticity.
Omitted variable
Furthermore, the models are tested for omitted variables using a Ramsey Reset test with powers of the fitted values. This showed that model 1a, 1b for both countries and 1c for Japan have omitted variables. By adding variables (CPINL1, CPIJAP1 and CPIJAP2), this p-value of not having an omitted variable increased, with an exception that when adding CPIJAP2 and QE to the model for Japan, the p-value decreased, which is strange, it goes against expectations, because adding correct variables should give less chance for omitted variables, not more. This indicates that the variables that were added are not right ones, because they do not improve the model. When model 2 was used, there was no significant omitted variable for both the Netherlands and Japan, which makes sense, because there is no problem of the CPI correlating with the CPI in a prior period. Omitted variable bias make the estimators of the betas biased and inconsistent, because the assumption of the error term being uncorrelated with the regressor is violated. The bias increases when the correlation is larger, because the error term will be more correlated with the regressor. Normally a beta 1 is estimated with: Cov{x,y}
σ{x}2 =β{1}
When there is an omitted variable the beta is predicted by: β{1}+ β{2}E(s(X2,X1)s2(X1) |X) ≠ β{1}
Where beta 2 is the omitted variable and s2is the sample covariance and s is the sample covariance.
Tables 3 and 4 (see page 16) show that the correlation between the different periods of inflation are large (CPI, CPI1 and CPI2) in both countries. So model 1 has is biased and is inconsistent. For Japan when the variable of QE was added to model 2, there was again an omitted variable, which might be due to the heteroskedasticity of the errors in this model.
Normality
Normality is an important assumption for performing an OLS regression (Stock and Watson, 2015-c). This is tested by the Shapiro-Wilk test for normal data with a H0: normal distributed data (Shapiro and Wilk, 1965). However, this test is biased by the sample size (Field, 2009). When the sample size is too small, the test is not reliable. As the sample size in this research is small, a Jarque–Bera test (also called a goodness-of-fit test), which tests
whether the sample data is symmetric and has “heavy tails”, is used. This is measured by the skewness and kurtosis (Jarque, 1987). The skewness and kurtosis are unit free. Matching a normal distribution gives a skewness of zero and a kurtosis smaller than three. The Jarque-Bera test uses a joint null hypothesis with the skewness and the excess kurtosis being zero. For small samples, this approximation of the test gets easily a type 1 error. So here, the alpha has to be corrected for the small sample size used in this research (151 at least) and a 0.05 alpha gets to be with a sample size of one hundred is 0.062. So, the data is of both countries is normally distributed as seen in Tables 5 and 6, because the p-values are bigger than 0,062.
After discussing and testing the assumptions of the model, the variables and coefficients are analyzed. The OLS estimator, might not be the best way to do this regression due to some problems with the assumptions, like the independently drawn
assumption, but the tested assumptions like normality and homoscedastic errors are fulfilled. Furthermore, the omitted variables can cause a bias within the models and coefficients.
Table 1—Main Results (Japan)
Variable (1a) (b) (c) (QE) (2) (2QE)
CPI -0.7606** 0.8469 0.6539 0.4901 (0.2902) (1.2333) (1.2768) (1.2555) CPI1 – -1.6772 −0.7243 -0.7781 – (1.2508) (2.0221) (1.9860) CPI2 – – -0.7879 0.0004 -0.6031 − – (1.3122) (1.2906) QE – - 2.1082** -0.2192 – - (0.8031) (0.2513) CPIdiff − – − − -0.5864 -0.5695 – - - (0.3843) (0.3851) Observations 164 164 164 164 151 151
Notes: Coefficients are reported for OLS models with consumer confidence as the dependent variable. For model 1, Column 1 is the basic specification, Column adds B2 with CPI1, or inflation from the previous month, Column 3 adds the inflation of 2 months ago, Column 4 adds the dummy variable of quantitative easing. Column 5 is for model 2 with CPIdiff or differences model. Same goes for Column 6, where only QE, the dummy for quantitate easing, is added. ** indicates that a variable is significant at the 5% level.
Table 2—Main Results (the Netherlands) Variable (1a) (b) (c) (2) CPI 0.1245 0.6321 0.9127 (0.1592) (2.5735) (2.6776) CPI1 – -0.5060 −1.8234 − – (2.5607) (4.2409) CPI2 – – 1.0376 – − – (2.6578) − – – difCPI − – − -0.3981 – (0.4028) Observations 191 191 191 191
CPI
In none of the models or regressions is CPI significantly different from zero and a predictor for CCI. Not for the Netherlands where there is inflation, neither for Japan with deflation. From the literature review in previous researches, inflation was a predictor for consumer confidence in previous research, which mostly involved inflation and no deflation. This might cause the difference, but more likely, most research used more variables in the models and, therefore, have less chance of omitted variable bias and explain more overall. Also, a more experienced form, like a vector autoregressive (VAR) model, might be the cause for significant results. It was hypothesized that in Japan deflation would not have an influence on consumer confidence, considering the results there is no prove to accept this hypothesis. It is worth mentioning that the deflation in Japan explained more of the variance in consumer confidence than the inflation in the Netherlands, the models for Japan have in general a higher r2 than the models for the Netherlands. This might be because deflation is
rarer than inflation. Still, the correlation in Tables 3 and 4 show that CCI is negatively correlated with CPI in Japan and positively in the Netherlands. This indicates that the neoclassical economic theory of deflation being bad for consumer confidence might be right after all.
3. Correlations (Japan)
CCIJAP CPIJAP CPIJAP1 CPIJAP2 diffCCIJAP diffCPIJAP QE CCIJAP 1 CPIJAP -0.3312 1 CPIJAP1 -0.3702 0.9394 1 CPIJAP2 -0.390 0.8588 0.9419 1 difCCIJAP -0.1310 0.1475 0.1512 0.1738 1 difCPIJAP 0.0131 -0.1334 -0.0582 -0.0098 -0.1077 1 QE 0.2178 -0.4711 -0.471 -0.4710 0.0321 0.0808 1
4. Correlations (the Netherlands)
CCINL CPINL CPINL CPINL1 diffCPINL diffCCINL CCINL 1 CPINL 0.0568 1 CPINL1 0.9981 0.0558 1 CPINL2 0.9951 0.0571 0.9981 1 difCCINL -0.0376 -0.0079 -0.0869 -0.1360 1 difCPINL 0.1510 -0.1536 0.1574 0.1568 -0.0717 1 QE
The dummy variable for quantitative easing in Japan has a significant influence on consumer confidence in model 1, but not in model 2. This is strange, because looking at the correlations, the dummy variable for quantitative easing seems to have a negative relation with the CPI in Japan in the recent periods, one month ago, and two months ago, which is
not the purpose of quantitative easing. Rather, the idea of quantitative easing is that putting more money into the economy would increase inflation, so QE should have a positive correlation with inflation, which is measured by the variables of CPI, CPI1, CPI2 in Japan. The mean of QE is 0,7, which means that during 70% of the time span used in this research (2001–2016), Japan had a quantitative easing program. This makes it questionable whether quantitative easing has an influence on deflation, because if quantitative easing was used for most of the period during which Japan experienced deflation, it is not effective for fighting deflation (Voutsinas and Werner, 2011).
And it should not influence consumer confidence much, because people will see that it does not work and would not be influenced by this. Thereby it gets normal in this period, because most of the time within the time span the quantitative easing program was active. On the other hand, influencing consumer confidence might be the only reason to implement QE anyway, so the influence does not have to be a coincidence. Correlation between CCI and QE is only 0.21, so the influence that the betas represent are probably a coincidence after all.
Model 2
Model 2 is an improvement for the Netherlands and gives a higher explained variance (r2-see table 7 and 8 below) but is still insignificant. For Japan, model 2 is a
deterioration. The model becomes insignificant (r2 of 0,8%), while in the best model -1QE r2
is 9%. This might be because of the time lag, as model 2 does not account for changes in time. This model does not have or at least has less have biased and inconsistent estimators, because fewer variables are omitted than in model 1.
Table 7— (Japan)
Variable (1a) (b) (c) (QE) (2) (2QE)
r2 0.0407 0.0513 0.0534 0.0927 0.0154 0.0204
Table 8— (the Netherlands)
Variable (1a) (b) (c) (2)
4. Conclusions
This section will summarize the paper and give the conclusions of the research. Starting with the effect of deflation on consumer confidence in Japan and then the Netherlands and ending with the overall conclusion of this research.
This thesis investigates the influence of deflation on consumer confidence in the case of Japan. The results were insignificant, meaning that as hypothesized, deflation did not have an influence on consumer confidence. Although quantitative easing had a negative effect on the inflation rate, which is the opposite of the purpose of this policy, it has a positive influence on consumer confidence. This might be the only purpose of quantitative easing, because it has proven to not be effective at combating deflation. However, this result might be a coincidence, because quantitative easing was measured in this research with a dummy variable. This means, the volume did not have an impact in the regression, which might be different in reality. Japan had an active quantitative easing policy during 70% of the time span used in this research, which makes the chance of a coincidence large. When model 2, in which time changes did not matter, was used, the model became heteroskedastic, there was an omitted variable and the dummy QE became insignificant, so the effect of
quantitative easing on consumer confidence is also questionable.
The Netherlands was used as a comparison, because previous research found that inflation influences consumer confidence. The Netherlands had only one month (June, 2016) of deflation during the investigated time period. In all the models inflation had no significant influence on consumer confidence. This might be due by a poor model design or incorrectly used data.
However, correlations between deflation and consumer confidence in table 3 show that the deflation in Japan had a negative relation with consumer confidence and the inflation of the Netherlands had a positive relation with consumer confidence, so the neoclassical economic theory might be correct. To confirm this, further research is recommended.
5. Discussion
This section explains the meaning of the results and how they relate to previous research in inflation and deflation. Additionally, it discusses the limitations of this research and provides suggestions for future research.
As seen in the literature review there has been done some research to the influence of inflation on consumer confidence. There was found a positive relation. To go further this research also looked at deflation in Japan and took the Netherlands with inflation as control country. In this research all the regressions were insignificant, meaning deflation and inflation might not have an impact on consumer confidence, which is contradicting previous research which stated that inflation at least has influence on consumer confidence. The results from this research are therefore questionable in more perspectives.
Generalizability
Comparing the effects of inflation and deflation by looking at Japan and the
Netherlands might be not representative for all cases, because both countries are different from each other and other countries, there has not been made a correction for this.
Additionally, although both countries spent the same percentage of GDP on social welfare in 2013, this does not mean that this is the same for the entire time span. Another difference is that Japan has a more aging population than the Netherlands. All these things can cause serious differences as to why inflation in Japan for instance of deflation in the Netherlands or somewhere else might have a whole different effect on consumer confidence. Thereby this research did not take level effects into account, but this might be interesting, because it could be that the levels can have a big effect as well and might be also consistently different. For instance, it might be the case that in this time span Japan had a consistent lower level of consumer confidence than the Netherlands, because of the periods of deflation, but this is not taken into account. Also, because few countries have experienced decades of deflation like Japan, it is difficult to find enough data to have a generalizable research to the influence of deflation on consumer confidence.
Model design
Another possible cause for the lack of significant findings is the time lags between deflation and consumer confidence. Both are measured at the same time, but deflation might be experienced later by consumers and, therefore, also reacted to later as well as in
consumer confidence, so there might be a time lack in the reaction of consumers of deflation in their consumer confidence, so there might be a delayed effect. The time lags might not
have been handled correctly in this research. A way around this would have been to use a VAR modeling approach, but due to practical constraints, this paper does not do so.
The problem of the different measurements of CCI could be solved by taking the growth rates, but this was not possible within this research. Because the Netherlands has negative values of CCI so taking a log of CCI is not possible. Thereby van Oest and Franses (2008) mention that changes in consumer confidence are hard to measure, according to their research of Dutch and American consumer confidence the changes are mostly not significantly different than zero. Looking at changes might therefore not be the best method.
Finally, there might be a simultaneous causation, because if consumer confidence is low and people buy less, this might cause deflation. It is not clear which way the
6. References
Acemoglu, D. and Scott, A. (1994) Consumer confidence and rational
expectations: Are agents’ beliefs consistent with the theory? The Economic Journal, 104 (January) (1994), p. 1–19
Andolfatto, D and Li Li. (2014) Quantitative Easing in Japan: Past and Present,
Economic SYNOPSES, short essays and reports on the economic issues of the day,
January 2014, Number 1
Alsem, K.-J., Brakman S., Hoogduin L., Kuper G. (2008) The impact of
newspapers on consumer confidence: Does spin bias exist? Applied Economics, 40 (2008), pp. 531–539
Bank of England. (2011) Quantitative Easing Explained. London.
Benati, L. (2009), Long Run Evidence on Money Growth and Inflation, ECB Working
Paper, No 1027.
Bordo, M. D. (2014) Exiting from Low Interest Rates to Normality: An Historical
Perspective, Economics Working Paper 14110, Hoover Institution, November 2014 Retrieved 10 March 2015.
Bryan, M. F. (1997) Federal Reserve Bank of Cleveland On the Origin and Evolution
of the Word "Inflation"
Cagan, P., & Moore, G. (1981). Some proposals to improve the Consumer Price
Index. Monthly Labor Review, 104(9), 20-25. Retrieved from: http://www.jstor.org/stable/41830814
CBS. a) (2016). news articles via:
https://www.cbs.nl/nl-nl/nieuws/2016/16/consumentenvertrouwen-weer-positief https://www.cbs.nl/nl-nl/nieuws/2016/31/lange-tijd-lage-inflatie
b) About measuring consumer confidence in the Netherlands: (2005).
Consumentenvertrouwen via:
c) About measuring consumer confidence comparions in the Netherlands: (2005)
Consumenten- en producentenvertrouwen vergeleken via:
https://www.cbs.nl/nl-nl/onze-diensten/methoden/onderzoeksomschrijvingen/aanvullende%20onderzoeksbeschrijvingen/c onsumenten-en-producentenvertrouwen-vergeleken
Carroll, C.D. (1992) The buffer-stock theory of saving: Some macroeconomic
evidence, Brookings Papers on Economic Activity, 2 (1992), pp. 61–135
DNB (2007) de ups en downs in het consumentenvertrouwen ontrafeld
kwartaalbericht juni 2007 via:
http://www.dnb.nl/binaries/De%20ups%20en%20downs%20in%20consumentenvertrouwen %20ontrafeld_tcm46-156861.pdf
Djerf, K. and Takala, K. (1997). Macroeconomy and consumer sentiment:
Performance of Finnish consumer barometer after ten years. Bank of Finland discussion
papers (p. 20)
Eijffinger, S.C.W., Geraats P.M., (2006). How transparent are central banks?
European Journal of Political Economy, 22 (2006), pp. 1–21
Eppright, D.R., Arguea N.M.and Huth, W.L. (1998), Aggregate consumer expectation indexes as indicators of future consumer expenditures, Journal of Economic
Psychology, 19 (1998), pp. 215–235
Field, A. (2009). Discovering statistics using SPSS (3rd ed.). Los Angeles SAGE
Publications. p. 143. ISBN 978-1-84787-906-6.
Friedman, M. (1970) The Counter-Revolution in Monetary Theory, IEA Occasional
Paper, no. 33 Institute of Economic Affairs. First published by the Institute of Economic Affairs, London, 1970. page 11
Gelper, S., Lemmens A. and Croux, C. (2007) Consumer sentiment and consumer
spending: Decomposing the Granger causal relationship in the time domain, Applied
Economics, 39 (January) (2007), pp. 1–11
Greenspan, A. (2002) “Remarks to the Bay Area Council Conference.” San
Available at: http://www. federalreserve.gov/boarddocs/speeches/2002/ 20020111/default.htm
Huth, W.L., Eppright, D.R. and Taube, P.M (1994). The indexes of consumer
sentiment and confidence: leading or misleading guides to future buying behavior Journal of
Business Research, 29 (1994), pp. 199–206
Jarque, C., & Bera, A. (1987). A Test for Normality of Observations and Regression
Residuals. International Statistical Review / Revue Internationale De Statistique, 55(2), 163-172.
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, Voorburg.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision
under risk. Econometrica, 47, 263-291.
Klaassen, F. (2016) during a lecture of the course Internationale monetaire
betrekkingen at the uva at slide 7 of chapter 7 part 1
Kwan A.C.C. and Cotsomitis, J.A. (2006) The usefulness of consumer confidence
in forecasting household spending in Canada: A national and regional analysis Economic Inquiry, 44 (January) (2006), pp. 185–197
Lemmon, M. (2006) Consumer Confidence and Asset Prices: Some Empirical
Evidence, review of Financial Studies (Winter 2006) 19 (4): 1499-1529 first published online March 15, 2006
Ludvigson, (2004) Consumer confidence and consumer spending Journal of
Economic Perspectives, 18 (2) (2004), pp. 29–50
Matsusaka, J. and Sbordone, A. (1995) Consumer confidence and economic
Matthews, K., Giuliodori, M. and Mishkin, F. S. (2013), The Economics of Money,
Banking and Financial Markets, European edition, p. 370.
OECD (2013) social expenditures, checked out at 28 December 2016
http://stats.oecd.org/Index.aspx?DataSetCode=SOCX_AGG
van Oest, R. & Franses, P. H. (2008) Journal of Economic Psychology
Volume 29, Issue 3, June 2008, Pages 255–275
Shapiro S. S., Wilk M. B., An analysis of variance test for normality (complete
samples). Biometrika 1965; 52 (3-4): 591-611.
Shostak, F.(2002) Defining Inflation, Mises Institute, 6 maart.
Slacalek, J. (2004). Forecasting consumption, working paper. German Institute for
Economic Research.
Smaghi, L. B., (2009), Speech, Member of the Executive Board of the ECB at XXI
Villa Mondragone International Economic Seminar “Global Crisis and Long Term Growth: A New Capitalism Ahead?” Tor Vergata Economic Foundation and CEIS – Centre for
Economic and International Studies University of Rome, Rome, 24 June 2009
Spiegel, M. (2001) Quantitative Easing by the Bank of Japan FRBSF ECONOMIC
LETTER Number 2001-31, November 2, 2001
Stock, J. H.and Watson, M. W. (2015), introduction to econometrics, global edition,
Pearson, p.175-a, p. 221-b and p. 173-c
Trevithick, J.A. (1975) The Economic Journal Vol. 85, No. 337 (March), pp. 101-113
Vuchelen, J. Consumer sentiment and macroeconomic forecasts Journal of
Economic Psychology, 25 (2004). pp. 493–506
Voutsinas, K. and Werner, R.A. (2011) New Evidence on the Effectiveness of
‘Quantitative Easing’ in Japan CFS paper No. 2011/30 (2011)
