Chapter 2 - Literature Review
2.3 Treatment of the Indicators
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
17
2.3 Treatment of the Indicators
This section introduces the indicators used in the calculations followed by the reasons for their use.
Most of the studies about consumption use the aggregate consumption expenditure as an independent variable, since they want to understand the direct impact of the independent variables over the consumption. As this research aims to understand what are the reasons that leads the Chinese Households to consume less proportionally to the income along the time, a different, more comprehensive model must be adopted. This study uses the proportion of this consumption over the income rather than the consumption expenditure. Variations of the study of consumption through other dependent variables can be seen for example in Barnett
& Brooks (2010), which used the total savings instead of consumption expenditures.
So, this is a reasonable way to understand what factors lead the behavior of this proportion (independently of inflation), other than just the variation of the nominal consumption.
Since the purchasing power of consumers and the value of currency varies according to the development of the economy, when household consumption in China is studied the obtained results are an average of the impact of a certain dependent variable over the independent one; but this does not consider the effect of inflation, the real evaluation of the income and its effect over the other expenses, such education and health. An exception in studies applied to China is the research lead by Modigliani & Cao (2004) in which inflation was considered an independent variable.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
18
In that sense, the variables education, health and housing will have their indicator divided by the Household Total Income. This is also because the share of these expenses over the Household Total Income translates not just into the ability of the family to pay such expenses, but also tracks down the real variation of these expenses in relation to the family`s income – since the analysis is made by level of income. For that, in order to find the income and expenditures (consumption, health and education) per household, the data obtained through CEIC database of income and expenses per capita was multiplied by the number of people in a household, which generated the Household Total Consumption Expenditure, Household Total Income and Consumption Exp. per Household on: (1) Recreation, Educational &
Cultural Service ; and on (2) Medicine & Medical Service.
In particular, for the housing variable, the indicator PTI was obtained from the rate between Commodity Bldg. Selling Price: Residential and Household Total Income for each income level. It is understood that the impact of the housing price on the less wealthy families might be underestimated; and for the wealthiest ones, overestimated. This, in a last analysis, might alter the coefficients obtained at the extremes of the income levels. It is also understood that the housing price might vary by region and by income level, since for each level, the families would choose their real estate according to its affordability. But the insistence of this indicator is to test if and how the burden assumed by the families of buying a house would impact on the consumption, otherwise, this analysis would not be possible due to restrictions on available data.
For the variable “Dependence”, the used indicator was obtained from the difference between the No of persons per household and the No of employees per household. The use of this indicator is to check if and how the number of
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
19
dependents in a household affects the APC. As said, it is still not completely clear whether this variable impacts the propensity to consume of the families, specifically in China.
And finally, as defended by Modigliani and the Life-Cycle Hypothesis, the growth of the households` income has a negative relationship with consumption.
So, to measure this variable, the indicator used is obtained simply by the difference between the income of the previous year and the current one. It is known that others methods to capture the variation of income exist, such as those proposed by Friedman with the Permanent Income Hypothesis (Friedman, 1957) but in order to maintain the length of the analyzed period and minimize interventions on data, this study opted to use the simple variation of income.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
20
3. Methodology and Data Information
3.1 Methodology
Basically the approach used in this research, classified as quantitative, follow two steps. The main and first one consists of running the Quantile Regression, shared into quintiles, and test the effects of the independent variables over the APC. It is followed by the results of the regressions of each income level and their interpretation. The results and analysis are shown firstly for the National level and it is followed by the results for each of the income level. The second step consists of a qualitative analysis attempting to track down the possible reasons for the main downturns (drastic variations) in the National APC during the studied period.
3.1.1 Quantile Regression
The OLS Regression, also called Ordinary Least Squares, is the most used method to test and run models throughout most of the literature about theories of consumption (Chamon & Prasad, 2010; Jin, Li, & Wu, 2011; Meng, 2003;
Modigliani & Cao, 2004 etc). But, running the ARCH test (Autoregressive Conditional Heteroskedasticity), the results show the presence of heteroskedasticity, which justifies the use of Quantile Regression for the analysis of this study, instead the OLS Regression.
Following Jin, Li, & Wu (2011), in order to identify the effect of the dependent variables over the consumption behavior, we will estimate an empirical model, as follow:
APC = α + βE + γH + δD + ⱷPTI + πΔY + ε (1)
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
21
Where APCis the Average Propensity to Consume, E the Indicator for the variable Education; H, for health; D, for Dependence; PTI, for Housing; and ΔY, for the variation on the income. The coefficients β, γ, δ, ⱷ and π are their respective coefficients. α is the intercept coefficient, ε is the coefficient for standard error – values considered as zero, for simplification.
In Eq. (1), each of the coefficients shows the impact their respective variables over the APC. Which means that the coefficient β and γ measure the impact over the APC for each variation on Consumption Expenditure per Household on Recreation, Educational & Cultural Service, and on Consumption Exp. per Household on Medicine & Medical Service over Household Total Income, respectively. For the variable D, δ expresses the relationship between APC and the dependency rate for each household. And for PTI, ⱷ represents the impact of the burden on a family of purchasing a house on the APC. The value of π, in turn, shows the impact of variation of income on the APC.
Following this method, it is expected to better understand: (1) the relationship of all these variables for each level of income, based on the findings of previous studies; and, (2) which variables really have influence over the APCof each level of income. Besides, it is also expected to understand why different authors could get different results even if studied the same variables. The answer for this puzzle might be in an analysis of the Chinese households from the perspective of each income level.
In addition, the use of quantile regression opens a new perspective on understanding the relationship between the variables according to the variations in the APC along the past 20 years in the Chinese economy by income and national
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
22
level. In other words, how do the variables behave and what are their interactions with the propensity to consume according to its own variations?
In the presence of heteroskedasticity, the simple OLS regression is unable to capture a better view of the real relationship between the dependent and the independent variables. To cover this gap, the Quantile Regressions (QR) is an excellent tool since it focuses not only on the conditional mean, but also on the whole conditional distribution of the variables, along different locations; which means that the QR works better with variables that are not evenly distributed. This is why the QR offers a better overview of the relationship between the dependent and independent variables (Davino, Furno & Vistocco, 2014). Therefore, Quantile Regression makes it possible to understand the role of each independent variable over the dependent variable, for each quantile.
In that sense, this research will divide the analysis into five quantiles, for each income level and for the national level. Which means that, each quantile possesses 20 percent of the Average Propensity to Consume. Therefore, it is expected to understand the role of these variables over the APC according to its scale of values.
3.1.2 Analysis of the Turning Points: 1995-1996 and 2000-2002
In order to give a more comprehensive analysis of the results found through the previous steps, a qualitative analysis is made aiming to track down the possible reasons for the main variations (main upside-downs) of the APC during the studied period. This analysis consists of associated politic-economical events that might have a high impact over the APC. The starting point of this analysis begins is with the dependent variables, i.e., investigate relevant politic-economical facts on education,
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
23
health, housing price, dependence ratio and variation of income; and if no possible relationship is found between the working variables, then the scope of the investigation is widened and other facts are analyzed, such as economic crisis and general political milestones.
As observed on the Graph 1 – National APC, the most abrupt variations in the National APC happened in 1995-1996 and on the years 2000, 2001 and 2002.
Therefore, this qualitative phase of the analysis for the national level remains just over these years.
The sources for this qualitative analysis are considered secondary sources and vary from official statements/laws issued by Chinese government to academic and journalistic articles.
3.2 Data Information
All the data used in this research are secondary data from the National Bureau of Statistics of China through the National Household Survey and compiled/obtained from the CEIC Data Base by Euromoney Institutional Investor Company. The intertemporal analysis goes from 1992 to 2012, due to limitations in the data available.
The Table 3 – “Data Description” shows the name of each data used in this work, as well as the used abbreviation code and the length time of each data.
In order to expand the horizon of the analysis, the data “Commodity Bldg. Selling Price: Residential” had to be projected for the years 1992, 1993 and 1994. A linear progression was used to do this projection taking as base for the calculation the first eight years of the data range (from 1995 to 2002). Since for
‧
Data** Abbreviation Time Period
National Level
Household Income per capita Ypc Dec./1952 - Dec/2013
Household Consumption Expenditure per Capita Cpc Dec./1952 - Dec/2015 Consumption Exp. per Capita: Recreation,
Educational & Cultural Service
Epc Dec./1992 - Dec/2014
Consumption Exp. per Capita: Medicine &
Medical Service
Mpc Dec./1992 - Dec/2015
No of Employee per Household EpH Dec./1985 - Dec/2012
No of Person Per Household PpH Dec./1985 - Dec/2012
Commodity Bldg. Selling Price: Residential RP Dec./1995 - Dec/2014 by Income Level*
Household Income per capita Ypc* Dec./1985 - Dec/2012
Household Consumption Expenditure per Capita Cpc* Dec./1985 - Dec/2012 Consumption Exp. per Capita: Recreation,
Educational & Cultural Service+
Epc* Dec./1985 - Dec/2012
Consumption Exp. per Capita: Medicine &
Medical Service+
Mpc* Dec./1985 - Dec/2012
No of Employee per Household EpH* Dec./1985 - Dec/2012
No of Person Per Household PpH* Dec./1985 - Dec/2012
*All the abbreviations by Income level will have a *, and after that, the specification to which level of income that code belongs.
**The yearly data had to be converted to quartiles.
+Although it is specified that these data runs from 1985 to 2012, but the data for the years 1986-1989 are not available.
Another important observation is that in order to increase the number of samples, the software EViews was used to convert the yearly data to quarterly data.
For that, the frequency of conversion was the average of observations (as high to low frequency method) and constant-match average (as low to high frequency method). In addition, Eviews is also the software used to run the regressions used in this study.
‧
About the definition of the main variables, the National Bureau Statistics of China (National Bureau Statistics of China, 2002), the Total Income of Urban Household can be defined as:
“[…] the total actual income of the sample households, including regular or fixed income and occasional income. The income of a circulating nature such as withdrawal from bank deposits, loans borrowed from relatives or friends, repayment of loans received and various temporary collection of money is excluded.” (National Bureau Statistics of China, 2002)
Another important concept defined by the same institution is the so-called Expenditure for Consumption of Urban Households:
“[…] refers to total expenditure of the sample households for consumption in daily life, including expenditure for various commodities and expenses for non-commodity items such as culture and service, etc., but excluding fines and confiscation, loss, tax payments (such as income tax, license tax, real estate tax, etc.) and various expenses by individual laborers for business purposes.” (National Bureau Statistics of China, 2002)
And finally, as defined by the China Statistical Yearbook of 2012 (National Bureau Statistics, 2012), Urban Households by Income Group is understood as11:
“All households in the sample are grouped, by per capita disposable income of the household, into groups of lowest income, low income, lower middle income, middle income, upper middle income, high income and highest income, each group consisting of 10%, 10%, 20%, 20%, 20%, 10%
and 10% of all households respectively” (National Bureau Statistics, 2012).
11 For more the definition of the others variables and for more details about the National Household Survey, see APPENDEX I.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
26
4. Results & Discussions
This Chapter is divided into three sections. The first section “Empirical Results”
reports the results and the findings obtained from the empirical model used. This section firstly analyzes the results at the National Level and, then, the results by Income Level. The second section, called “Overall Interpretation of the Results”
brings the main points and their respective interpretation besides those already presented in the previous section. The third section is “Analysis of the Turning Points: 1995-1996 and 2000-2002” and, it shows an economic and political analysis of the possible reasons for the main variations the APC during the studied period.
4.1 Empirical Results
In this section the results are presented first at the national level, and then, by Income Level.
4.1.1 National Level
The starting-point of the analysis begins with a comparison of the results of the Quantile Regression at the National Level and the Graph 1, presented in the first chapter. As shown on the Table 4, the simple OLS regression reveals that the variables Health, Housing Price and the Variation of Total Income has a negative impact over the Average Propensity to Consume (APC) of the urban Chinese households. Their respective coefficients are approximately (-11.1), (-0.5) and (-0.2), which means that during the period analyzed (1992-2012), the relative increase of the household expenses on health-care per total household income had the strongest impact over the National APC. These results are followed by the relative increase of the housing price over the household total income; and the variation of the income.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
27
In addition, as observed, the impact of PTI has a little gap between the results computed by the OLS model and those from the QR model, which means a value of (-0.72), in average – obtained from all the significant values of this variable.
Therefore, it is important to notice that, considering the presence of heteroskedasticity, the values that should be observed are those obtained from the QR model – this highlight the importance of the use of QR model, since the results obtained from the OLS model would not frame the most precise scenario.
On the other hand, the variable Education and Dependence have a positive relationship with the APC, which shows that the greater the increase of the expenses of a household on education (relative to the total income), the greater is the propensity to consume of this family. The variable with the highest positive impact is education, with the coefficient value of (6.4); followed by the variable Dependence, with the coefficient value of (0.7). For this regression, the Adjusted R-squared value is 0.96, which means that the model can explain in great part the variation of the APC at the national level.
With the results of the Quantile Regression and with the help of Graph1, it is possible to conclude that for the years in which the APC is bigger (quantile 0.8, for example), the negative impact of the variable health is smaller than the negative impact caused by this variable for the years in which the APC is smaller (quantile 0.2, for example). This means that, considering that the years with smaller APC are coincidently the more recent ones, then it is possible to affirm that the negative impact caused by the expenditure with health-care over the APC are having a stronger effect in recent years if compared to the beginning of the time period. In other words, the burden of a family with health-care has a heavier weight for the years with smaller APC.
‧
Table 4 - Results of the OLS and Quantile Regression at the National Level
OLS Regression
Variables Coefficient Std. Error t-Statistic Prob.
E 6.399234 0.220714 28.993330 0.0000
H -11.083800 0.406933 -27.237390 0.0000
PTI -0.493703 0.105879 -4.662918 0.0000
D 0.703351 0.017325 40.597590 0.0000
ΔY -0.228674 0.030222 -7.566451 0.0000
R-squared 0.964683
0.400 6.334534 0.271588 23.324060 0.0000 0.500 6.370054 0.299960 21.236360 0.0000 0.600 6.396478 0.315640 20.265130 0.0000
0.800 6.466688 0.437929 14.766530 0.0000
H 0.200 -12.077520 0.643559 -18.766760 0.0000
0.400 -12.223200 0.357839 -34.158340 0.0000 0.500 -12.016400 0.353234 -34.018280 0.0000 0.600 -11.913090 0.335582 -35.499780 0.0000 0.800 -11.273760 0.335738 -33.579080 0.0000
PTI 0.200 -0.619100 0.113971 -5.432103 0.0000
0.400 -0.776100 0.079457 -9.767494 0.0000 0.500 -0.768756 0.081053 -9.484576 0.0000 0.600 -0.760946 0.077945 -9.762551 0.0000
0.800 -0.681834 0.085303 -7.993119 0.0000
D 0.200 0.721133 0.022905 31.484350 0.0000
0.400 0.760770 0.017909 42.479060 0.0000 0.500 0.755173 0.020122 37.530630 0.0000 0.600 0.751456 0.020285 37.044170 0.0000 0.800 0.728061 0.025145 28.954920 0.0000
ΔY 0.200 -0.239092 0.027954 -8.553094 0.0000
0.400 -0.256857 0.022656 -1.133730 0.0000 0.500 -0.251811 0.027748 -9.075062 0.0000 0.600 -0.248358 0.028280 -8.782112 0.0000
0.800 -0.232729 0.032895 -7.074792 0.0000
Source: Calculated by the author
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
29
Therefore, findings related to the variables H, PTI and ΔY confirm the raised hypotheses 1, 3 and 5, considering the results at the national level. However, it is not possible to confirm the Hypothesis 2 since the variable “E” has a positive impact on APC.
As cited, a positive impact was found between “D” and APC, which means that the greater the number of dependents in a household, the greater is the APC. This is possible simply because independently of the income of a household, the number of people has a positive relationship on its consumption – it would be impossible for a household to save if the number of dependents increased since each individual has a minimum required level of consumption.
Therefore, the results at the national level denies the Hypothesis 4 stated as: “the more the increase of the number of dependents in a household, the more this household would have incentives to save, and thus, decrease its consumption”.
4.1.2 Level of Income
The analysis by income level is presented in ascendant order of income, starting, thus, from the lowest household income level until the highest one.
Therefore, the Table 5 shows the results of the empirical model using the OLS and Quantile Regression for the Lowest Income of the households in urban China. Of course in the presence of heteroskedasticity, the results that have to be considered are those from the Quantile Regression, but the results from the OLS regression are presented because it gives us an idea of the behavior of each variable throughout the studied period as a whole; and again, in case of incompatibility of the
Therefore, the Table 5 shows the results of the empirical model using the OLS and Quantile Regression for the Lowest Income of the households in urban China. Of course in the presence of heteroskedasticity, the results that have to be considered are those from the Quantile Regression, but the results from the OLS regression are presented because it gives us an idea of the behavior of each variable throughout the studied period as a whole; and again, in case of incompatibility of the