Sei sulla pagina 1di 13

North American Journal of Economics and Finance xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

North American Journal of Economics


and Finance
journal homepage: www.elsevier.com/locate/najef

The asymmetric behavior of household consumption under the


business cycle
Kuang-Ta Loa, , Ta-Sheng Choub, Stephanie Tsuic

a
Department of Public Finance, National Chengchi University, No. 64, Sec. 2, Zhi-nan Rd, Wenshan, 11605 Taipei, Taiwan
b
Department of Business Administration, Chung Yuan Christian University, No. 200, Chung-pei Rd, Chung-li, 32023 Taoyuan, Taiwan
c
Department of Accountancy, City University of Hong Kong, 88 Tat Chee Avenue, Kowloon Tong, Hong Kong

ARTICLE INFO ABSTRACT

Keywords: In this study, we examine whether and to what extent the pattern of household consumption is
Cyclicality asymmetrically sensitive to economic changes. We use a threshold model with error correction to
Asymmetry characterize household consumption under the business cycle. Using data from the 1979–2014
Business cycle Survey of Family Income and Expenditure in Taiwan, we find that household consumption has
Household consumption
pro-cyclical characteristics during the business cycle. More importantly, household consumption
JEL Classifications: responds asymmetrically to economic fluctuations such that changes in consumption tend to be
E32 larger during expansions than during downturns. Furthermore, the nature of this asymmetry
D10
differs across household income quantiles.

1. Introduction

In this study, we examine the asymmetric responses of household consumption to changes in the business cycle. In addition to
being the subject of a large stream of literature, the relationship between economic change and the stages of the business cycle is an
important issue for policy makers. Some economic activities are believed to react asymmetrically to changes in the business cycle. For
example, loan rates and investment levels react quickly to economic downturns and recover slowly as the economic situation im-
proves. Therefore, this non-linear relationship and its consequences need to be incorporated into economic models to capture the
asymmetric disturbances. Moreover, this relationship is crucial for policy makers because it can influence the timing and effectiveness
of policy reforms and government interventions.
Household consumption is an important and frequent economic activity and serves as an important economic indicator. Cutler
and Katz (1992) use both income and consumption to understand changes in the distribution of wealth, and Blundell and Preston
(1995) use consumption to understand permanent differences in income. Household consumption usually fluctuates with the business
cycle and a nation’s economic performance. Although numerous studies have examined the cyclical nature of household con-
sumption, only a few studies have analyzed the links between the changes in household consumption and the state of the economy.
Moreover, this stream of research has provided mixed evidence concerning the relationship between household consumption and the
business cycle. For example, Shea (1995) finds that private consumption responds more strongly to predictable income declines than
to predictable income increases. He further argues that consumption exhibits asymmetric behavior due to the loss aversion in in-
tertemporal preferences. In other words, individuals suffer more when forced to reduce their consumption due to the diminishing
marginal utility of wealth. Similarly, using quarterly U.K. consumption and industrial production data, Öcal and Osborn (2000) find


Corresponding author.
E-mail addresses: vancelo@nccu.edu.tw (K.-T. Lo), tarzan@cycu.edu.tw (T.-S. Chou), Stephanie.Tsui@cityu.edu.hk (S. Tsui).

https://doi.org/10.1016/j.najef.2019.03.017

1062-9408/ © 2019 Elsevier Inc. All rights reserved.

Please cite this article as: Kuang-Ta Lo, Ta-Sheng Chou and Stephanie Tsui, North American Journal of Economics and Finance,
https://doi.org/10.1016/j.najef.2019.03.017
K.-T. Lo, et al. North American Journal of Economics and Finance xxx (xxxx) xxx–xxx

that consumption and industrial production respond to shocks asymmetrically. Conversely, Jappelli and Pistaferri (2000) use the
subjective quantitative income expectations of a sample of Italian households as an indicator of income growth and find no evidence
of excess sensitivity to income increases or declines.
In this study, we examine whether and to what extent changes in household consumption are asymmetrically sensitive to eco-
nomic movements. We use a threshold model with error correction to characterize household consumption under a business cycle. As
the business cycle may influence the pattern of household consumption, this model can prevent the possible bias of ignoring
asymmetric effects and allow a more general partition of the business cycle by setting a flexible threshold value.
Using household consumption expenditure data from the 1979–2014 Survey of Family Income and Expenditure in Taiwan, we
examine the relationship between the business cycle and household consumption over a long period and closely study the compo-
nents of household consumption. We first find that household consumption displays pro-cyclical characteristics during the business
cycle in Taiwan. Second, and more importantly, household consumption responds asymmetrically to economic fluctuations such that
the changes in consumption tend to be larger during expansions than during downturns.
This study makes a number of contributions to the literature. First, it extends the growing and important body of literature on the
volatility of household consumption. We not only consider the cyclicality of household consumption but also demonstrate the upside
and downside risks involved with household consumption. Second, using a general-to-specific modeling approach, we show that the
asymmetric effects of the business cycle on household consumption stem from short-run changes in household income and deviations
from the long-term trends of household expenditure.
In conclusion, we show that the business cycle has a nonlinear effect on household consumption. This asymmetry has some
important implications. First, because the responses of private consumption are asymmetric over different phases of the business
cycle, some of the existing theoretical economic models and empirical econometric models need to be revised. For example, models
with non-linear and asymmetric disturbances may be more appropriate. Second, considering the asymmetric behavioral responses,
the government should design different policies to stimulate household consumption during economic booms and recessions. In
addition, given that the asymmetric responses are observed in households with different levels of income, the relevant policies should
also be designed in accordance with the different income quantiles.
In the next section, we discuss the related literature and the empirical evidence on household consumption and its asymmetric
responses to the business cycle. In Section 3, we describe the data and empirical designs and discuss the findings. Section 4 concludes
the study.

2. Literature review

Although economic fluctuations have been widely found to affect household consumption, a more important and interesting issue
is whether the influence is symmetric during economic upturns and downturns. Patterson (1993) finds that consumption behaves
asymmetrically to wealth shocks and that this asymmetry mainly reflects the presence of imperfect capital markets, such as those
with liquidity constraints. Shea (1995) also finds that consumption responds more strongly to predictable income declines than to
predictable income increases. He further argues that consumption exhibits asymmetric patterns due to the loss aversion in inter-
temporal preferences. In other words, individuals suffer more when forced to reduce their consumption due to the diminishing
marginal utility of wealth. In contrast, using the subjective quantitative income expectations of a sample of Italian households as an
indicator of income growth, Jappelli and Pistaferri (2000) do not find evidence of excess sensitivity to income increases or declines.
Cook (2002) provides evidence of a highly significant asymmetric pattern of consumption, which is mainly attributed to dif-
ferences in consumption and savings behavior. Moreover, Carruth and Dickerson (2003) assess the likelihood that aggregate con-
sumers behave differently under various disequilibrium asymmetric shocks, and their empirical findings provide strong support for
this possibility. Tagkalakis (2008) uses a yearly panel of 19 OECD countries from 1970 to 2002 to investigate the effects of fiscal
policy changes on private consumption during recessions and expansions. He concludes that in the presence of binding liquidity
constraints on households, fiscal policies are more effective in boosting private consumption during recessions than during expan-
sions. This effect is more pronounced in countries characterized by a less developed consumer credit market because individuals who
face binding liquidity constraints during a recession consume the extra income generated by an unanticipated tax cut or government
spending increase. Similarly, using consumption data on Taiwan from 1971 to 1998, Kuo and Chung (2002) show that the asym-
metric sensitivity of consumption to the phases of the business cycle generates asymmetric patterns, and conclude that the con-
sumption patterns of liquidity-constrained consumers are closely related to the business cycle.
Another stream of literature focuses on the determinants of household expenditure. Miles (1997) uses data from the U.K. Family
Expenditure Survey to examine the determinants of household income and consumption. He finds that the estimates of permanent
income and earnings uncertainty have significant effects on spending and that one of the major motives for household savings is the
variability of earned income. Çağlayan and Astar (2012) investigate the determinants of household consumption expenditure in
Turkey and estimate separate models for rural and urban areas to examine the regional gaps in the distribution of consumption
expenditure. Using quantile regressions to examine the distribution of consumption at different points in rural and urban areas, they
find that age is positively related to consumption expenditure in the general and urban estimations and negatively related to con-
sumption expenditure in the rural estimations. In rural areas, only age, income, marital status, insurance, and the size of the
household have significant correlations. They also find that men have lower consumption expenditure than women in both the rural
and urban estimates.
Furthermore, some studies have focused on the intersection of the two streams of research on consumption. For instance, Souleles
(1999) examines how consumption responds to income tax refunds in the U.S. and shows that consumption reacts to preannounced

2
K.-T. Lo, et al. North American Journal of Economics and Finance xxx (xxxx) xxx–xxx

and temporary policy changes in fiscal policy. Specifically, while total consumption increases with the receipt of tax refunds, response
sensitivity varies between consumption types. Relatedly, Huntley and Michelangeli (2014) find that households have different
marginal propensities to consume tax rebates based on their levels of financial assets.
More recently, studies have documented that people in different income brackets have different consumption patterns and make
different decisions. For example, Jan (2006) analyzes the structure of household consumption and food expenditure segmented by
income levels in Taiwan. Using data from the Nationwide Household Revenue and Expenditure Surveys of 1994 and 2000, he
categorizes all families as low-income, medium-low income, or medium-high income according to government poverty guidelines
and uses the demand to estimate household consumption and food expenditure. He finds that demand elasticity differs between
income groups. Specifically, the results show that the food expenditure of low-income households is more responsive to income and
prices than that of medium low and medium high income households in Taiwan. Using a quantile regression approach, Lin and Lo
(2012) show that tax price elasticity and household income elasticity vary substantially across the income distribution. Furthermore,
it has been argued that cyclical downturns and economic crises may have different asymmetric effects on poverty. For instance,
Lustig, Fishlow, and Bourguignon (2000) argue that the fall in real income associated with economic downturns may have an
irreversible impact on the human capital of the poor. They argue that children from poor families (particularly the very poor) are
sometimes taken out of school and put to work in response to large adverse shocks—mitigating the fall in household income—but do
not return to school when the “good times” return. Overall, their findings suggest that adverse shocks to household income adversely
affect investment in schooling (in addition to nutrition and health) and the longer-run ability of the poor to enhance their stock of
human capital, which hinders their ability to escape from poverty. Thus, large recessions have a type of “asymmetric hysteresis”
effect on poverty in the sense that temporary negative shocks have persistent effects. Agenor (2002) uses a vector auto-regression
model to examine whether the output contractions associated with downturns and crises have asymmetric effects on poverty in
Brazil. His empirical results indicate that poverty is less sensitive to output shocks during the initial stages of economic decline.
In this study, our findings complement previous studies by documenting the pro-cyclical nature of household consumption using
time-series data. We further study the dynamics of different components of household consumption and document the asymmetrical
responses to various economic conditions. Moreover, in addition to cyclicality, we show the upside and downside risks associated
with household consumption and document the source of the asymmetric effects of business cycles on household consumption.

3. Empirical design

In this section, we examine the asymmetric effects of the business cycle on the patterns of household expenditure in both the short
run and the long run. Using data from Taiwan, we provide empirical evidence of the relationship between changes in household
expenditure and the business cycle. This relationship has not been well documented in the literature, and our findings are likely to be
of interest to policy makers.
To deal with the effects of serial dependence, all of our testing procedures include a heteroscedasticity and autocorrelation
consistent (HAC) covariance matrix, which ensures that the conclusions are asymptotically robust to heteroscedasticity and serial
correlation.1

3.1. Data source

To investigate how the business cycle affects the patterns of household expenditure in Taiwan, we collected data for three
categories of variables: (1) the amount of household expenditure; (2) the major explanatory variables and auxiliary regressors that
have been shown to be related to household expenditure; and (3) an indicator of the business cycle. Our focus is primarily on the
dynamic behavior of real household expenditure. We divide the total expenditure on household consumption (EXP) into two com-
ponents, namely real consumption expenditure (C-EXP) and real non-consumption expenditure (NC-EXP), and examine how each
component responds to different stages of the business cycle.2 The three variables are related (i.e., EXP = C-EXP + NC-EXP) and are
analyzed in a similar manner. In addition, it is well documented that different income groups respond differently to the business
cycle. Thus, we further investigate the responses of each expenditure component for different income levels.3
Following consumption theory, household real income (INC) is used as the major explanatory variable.4 We use the sex ratio (SEX,
the percentage of female-headed families), the average age of household heads (AGE), the average length of education attainment in

1
Recently, HAC-based bootstrapping has been proposed to further improve the finite sample performance of the asymptotic tests based on the
HAC standard errors (Davidson & Monticini, 2014). Considering its current popularity in empirical studies, we use the HAC-based asymptotic tests
in this study and leave a thorough investigation of the bootstrapping tests to future research.
2
In a traditional Keynesian consumption model, there are two components of consumption, namely autonomous consumption and induced
consumption. It is believed that these two types of consumption react differently to the business cycle. Due to data limitations, we closely proxy for
these two elements by using data on non-consumption expenditure and consumption expenditure. Examples of items included in non-consumption
expenditure are interest payments and various tax and social security payments, and examples of items included in consumption expenditure are
medical expenses, food and beverages, rent, utility bills, and entertainment. We thank a reviewer for this valuable comment and suggestion.
3
Various studies (e.g., Patterson, 1993; Carroll, Slacalek, & Tokuoka, 2014; and Carroll, Slacalek, Tokuoka, & White, 2017) have shown that
households with different income and wealth levels respond to shocks differently.
4
The figures for household real expenditure (and household real income) are constructed by dividing the household nominal expenditure (and
household nominal income) by a CPI (set at 100 for the year 2011). Both real expenditure and real income are transformed into a logarithmic form.

3
K.-T. Lo, et al. North American Journal of Economics and Finance xxx (xxxx) xxx–xxx

years (EDU), the monetary aggregate (M2), and the Gini coefficient (GINI) as the auxiliary regressors. Following Harding and Pagan
(2002), we use real gross domestic product (GDP), which measures aggregate economic activity, to represent the fluctuations in the
business cycle.
The sample observations are annual data from 1979 to 2014, mostly collected from the Report on the Survey of Family Income and
Expenditure published by the Directorate-General of Budget, Accounting and Statistics (DGBAS), Executive Yuan, and the online
statistical database maintained by the DGBAS, which is an official statistical institution in Taiwan.5

3.2. Model specification

To examine how the business cycle affects the patterns of household expenditure, we first included GDP as a regressor to explain
the changes in the various types of household expenditure (EXP, C-EXP, and NC-EXP). Our preliminary results indicate that
household income (INC) completely mediates the relationship between the changes in GDP and the various types of household
expenditure. As a result, the effect of GDP as an explanatory variable for the changes in household expenditure becomes statistically
insignificant once INC is included as a regressor. Thus, our main analysis focuses on the relationship between household expenditure
and income, especially on the asymmetric characteristics during different stages of the business cycle. The upper panel in Fig. 1
depicts the time series for EXP0 and INC0 (“0” is added to denote the sample of all households (no subgrouping)), which show similar
trends for the two variables. The lower panel in Fig. 1 shows the results for EXP1, INC1, EXP5, and INC5, which indicate subgroups of
EXP and INC for the first group of households (i.e., EXP1 and INC1 indicate the group with the lowest level of income) and the fifth
group of households (i.e., EXP5 and INC5 indicate the group with the highest level of income). We also examine the patterns of
household consumption expenditure (C-EXPs) and non-consumption expenditure (NC-EXPs), which are shown in Figs. 2 and 3.
To understand the basic properties of the time series, the commonly used augmented Dickey-Fuller (ADF) test (Said & Dickey,
1984) is applied with a lag length that minimizes the Akaike information criteria (AIC).6 The results of the ADF test indicate that the
variables for household expenditure and income are nonstationary and integrated of order one. Motivated by the graphical re-
lationships depicted in Figs. 1–3, we first explore the cointegration relationships with some structural break dummies. Notably, to
investigate the asymmetric effects, we use a framework motivated by the threshold models for discovering threshold unit roots
(Enders & Granger, 1998) and threshold cointegration (Enders & Siklos, 2001). Using a threshold model can prevent the possible bias
of ignoring asymmetric effects and allow a more general partition of the business cycle by setting a flexible threshold value. To make
our approach coherent, we use the two-step cointegration test pioneered by Engle and Granger (1987). As shown in Table 1(a)–(c),
the ADF test statistics in the second step indicate that the residual series from the first-step regression considering a structural break is
stationary at the 5% level.7 Accordingly, it can be concluded that household expenditure (EXPs, C-EXPs, and NC-EXPs) and household
income (INCs) are cointegrated under various structures in different periods. Thus, when studying the dynamics of household ex-
penditure based on the information in INCs, it is intuitive to include an error correction term (ECTs, ECTcs, and ECTns) that captures
a household’s adjustment to the deviation from the cointegration relationship between household expenditure and income.8
In addition to verifying the positive relationship between the changes in household expenditure (EXPs, C-EXPs, and NC-EXPs) and
the fluctuations in INCs, we are particularly interested in determining the asymmetric effects of the business cycle on the patterns of
household expenditure. Thus, we construct empirical models to capture the potentially unequal effects of household income (INCs) at
different stages of the business cycle. We adopt a commonly used method, namely the signs of the first-order differences of an
indicator, to partition the stages of the business cycle in a manner consistent with the spirit of the classical cycle.9 Specifically, we use
I ( GDP 0) to identify two apparently different stages of the business cycle, namely expansion and recession.10 Accordingly, we
design a threshold model by including INCs I ( GDP 0) and INCs I ( GDP < 0) to allow for the possibly unequal effects of
INCs on the changes in household expenditure at different stages of the business cycle. The estimated value of the coefficient prior
to INCs I ( GDP 0) measures the effect of INCs when GDP is rising. The coefficient prior to INCs I ( GDP < 0) is used to assess
the effect of INCs when GDP is falling.
Following the aforementioned studies, to control our analysis we include as auxiliary regressors the sex ratio (SEX), the average
age of household heads (AGE), the average length of educational attainment (EDU), the monetary aggregate (M2), and the Gini
coefficient (GINI), which are factors cited in the literature as affecting household expenditure. To remove stochastic trends from these
auxiliary regressors with various time series characteristics, we use a unified approach to data transformation to produce the de-
viation cycle, i.e., DC(x t) = x t HP(x t) , whereHP(.) denotes a stochastic trend generated by the Hodrick-Prescott filter (Hodrick &

5
Please refer to the DGBAS at http://statdb.dgbas.gov.tw/pxweb/dialog/statfile9L.asp.
6
As noted by Diebold (2007, p. 301), when selecting the augmentation order in a Dickey-Fuller regression, it may be better to use a less harsh
degree of freedom penalties. Thus, we use the AIC instead of the SIC.
7
The timings of the structural breaks are suggested by the multiple breakpoint tests of Bai and Perron (1998, 2003).
8
This model can be seen as a variant of the modified error correction model in Apergis and Miller (2006).
9
Three definitions are commonly used to distinguish the expansion and contraction of the business cycle: the classical cycle, the growth cycle, and
the growth rate cycle (Proietti, 2005). We follow the classical cycle (which results in a partition equivalent to that of the growth rate cycle) in
separating business cycles into two kinds of apparently different states. When we use a partition from the growth cycle, our main findings remain
almost the same if we replace all of the differenced variables in the models with their corresponding trend-deviation forms and ignore the error
correction term to have a consistent modeling perspective.
10
I(.) denotes an indicator function where: I(A) equals one if event A occurs and zero if event A does not occur. Over the 35-year sample period,
GDP rises every year with the exceptions of 2001 and 2008.

4
K.-T. Lo, et al. North American Journal of Economics and Finance xxx (xxxx) xxx–xxx

Fig. 1. Household income and total expenditure. Note: The EXP and INC values are logarithmic.

Prescott, 1997) with a smoothing parameter of 100.


In summary, we design a full model for the change in household expenditure, which considers the change in household income
( INC) and the mechanism of error correction (ECT) for the deviation from the long-run equilibrium at different stages of the
business cycle:

EXPst = 0 + 1· INCst · I ( GDPt 0) + 2· INCst ·I ( GDPt < 0) + 3·ECTst 1· I ( GDPt 0) + 4 ·ECTst 1·I ( GDPt < 0)
+ 1·SEXt + 2·AGEt + 3·EDUt + 4·M2t + 5·GINIt + t (1.s)

where s = 0, 1, and 5 indicate, respectively, the sample of all households, the first group of households (with the lowest income
level), and the fifth group of households (with the highest income level). We also examine two reduced models with restrictions R1:
1 = 2 = = 5 = 0 and R2: 3 = 4 , respectively.11 The regression results are tabulated in Table 2. In addition, Model (1.s) is used to
examine the changes in household consumption expenditure (C-EXPs) and non-consumption expenditure (NC-EXPs), which are
labeled as Model (2.s) and Model (3.s), respectively. The corresponding regression results are tabulated in Tables 3 and 4.

3.3. Empirical results

In this subsection, we examine how the business cycle affects the patterns of household expenditure, which include the short-run
changes in household income and the deviation from the long-run equilibrium. Here, the cointegration relationships between INCs and
each type of household expenditure (EXPs, C-EXPs, and NC-EXPs) are used to construct an error correction mechanism to study the
correction effects for deviations from the long-run equilibrium. All of the regression specifications in this section passed the assessment
of parameter stability via a sequence of one-step forecast tests based on the recursive residuals; see Diebold (2007, p. 209).12

11
The authors would like to thank a reviewer for suggesting that we properly impose the restriction of R2.
12
The authors would like to thank a reviewer for suggesting that we check the structural stability.

5
K.-T. Lo, et al. North American Journal of Economics and Finance xxx (xxxx) xxx–xxx

Fig. 2. Household income and consumption expenditure. Note: The EXP and INC values are logarithmic.

First, using GDP as the measure of the business cycle, Models (1.0), (1.1), and (1.5) in Table 2 represent the threshold models that
can identify the effects of INCs and ECTs on EXPs for the three groups of households (i.e., all households, lowest-income households,
and highest-income households) during different stages of the business cycle. By examining the full Model (1.0) (and its reduced form
with restriction R1) for all households, at the 5% level, we find that when GDP < 0 , the previous-year error correction term (ECT0)
has significant reverse-correcting effects on the current-year EXP0. When GDP 0 , there is no evidence that a deviation from the
long-run equilibrium has correcting effects. In addition, there is significant evidence that ECT0 has asymmetric effects according to
the HAC-based test (shown with its p-value as Asy-p in the LR-asymmetry). Considering that all of the auxiliary regressors are
insignificant, the conclusion in Model (1.0) with R1 is preferred to the full model. Specifically, the model indicates that changes in
INC0 have positive and significant effects on changes in EXP0, regardless of whether GDP is rising or falling. The estimated elasticity
of EXP0 with respect to INC0 is larger when GDP is rising than when it is falling. Furthermore, at a looser level of 10%, Model (1.0)
with R1 also provides evidence to reject the equality of 1 = 2 , which thus supports a significant difference between the two
aforementioned elasticities, as indicated by the p-value of 0.061 in the SR-asymmetry.
Similarly, the full model of Model (1.1) for the lowest-income households provides significant evidence at the 5% level showing
that the reverse-correcting effects of ECT1 on EXP1 are asymmetric (because Asy-p in the LR-asymmetry equals 0.000). Moreover,
considering that many of the auxiliary regressors have significant effects, the results of the full Model (1.1) are preferred. Specifically,
the model indicates that changes in INC1 have positive and significant effects on the changes in EXP1, regardless of whether GDP is
rising or falling, and that the effects are larger when GDP is rising than when it is falling. Furthermore, at an ordinary level of 5%, the
equality of 1 = 2 can be rejected, thus indicating that the difference is significant. In contrast, even at the 10% level, the results of
the full Model (1.5) for the highest-income households are insufficient to conclude that the effects of ECT5 on EXP5 are asymmetric
(because Asy-p in the LR-asymmetry equals 0.942). This implies that the results of Model (1.5) with R2 are preferred to those of the
full model. Thus, the resulting Model (R2) indicates that INC5 has significant effects on EXP5 when GDP is rising but is insig-
nificant when GDP is falling. In addition, at the 5% level, it can be concluded that the effects of INC5 on EXP5 are significantly
asymmetric during different stages of the business cycle.
Second, we investigate the effects of INCs (and ECTcs) on household consumption expenditure (C-EXPs) in Table 3 and find that
the major conclusions are similar to those for household total expenditure. The full model of Model (2.0) (and its reduced form with

6
K.-T. Lo, et al. North American Journal of Economics and Finance xxx (xxxx) xxx–xxx

Fig. 3. Household income and non-consumption expenditure. Note: The EXP and INC values are logarithmic.

restriction R1) for all households shows that when GDPt < 0 , the previous-year error correction term (ECTc0) has significant re-
verse-correcting effects on the current-year C-EXP0. In addition, there is significant evidence that the effects of ECTc0 are asym-
metric during different stages of the business cycle. As all of the auxiliary regressors are insignificant, the results of Model (2.0) with
R1 are preferred in regard to the model fit. Specifically, the model shows that changes in INC0 have positive and significant effects on
the changes in C-EXP0, regardless of whether GDP is rising or falling. The estimated elasticity of C-EXP0 with respect to INC0 is
larger when GDP is rising than when it is falling. In addition, at a level of 5%, there is sufficient evidence to reject 1 = 2 , thus
indicating that INC0 has asymmetric effects on C-EXP0 during different stages of the business cycle. The results of Model (2.1) for
the lowest-income households show that the correction mechanism adjusting for the deviation from the long-run pattern of household
consumption is significantly asymmetric between the periods when GDP 0 and GDP < 0 . Moreover, the changes in INC1 have
positive and significant effects on C-EXP1, and the effects are asymmetric during different stages of the business cycle. For Model
(2.5) for the highest-income households, the effects of ECTc5 on C-EXP5 are not significantly asymmetric. Thus, Model (2.5) with
R2 is preferred, which shows that INC5 has significant effects on C-EXP5 when GDP is rising. In addition, the effects of INC5 on
C-EXP5 are significantly asymmetric during the different stages of the business cycle.
Lastly, the results for the effects of INCs (and ECTns) on household non-consumption expenditure (NC-EXPs) in Table 4 indicate
that these coefficients have different characteristics than EXPs and C-EXPs. For all households, the best-fitting model is Model (3.0)
with R1. At a level of 5%, the model indicates that the reverse-correcting effects of ECTn0 are significantly asymmetric during the
different stages of the business cycle. However, there are no significant differences for the effects of INC0 on NC-EXPs at different
stages of the business cycle. Model (3.1) for the lowest-income households shows that at the 5% level, there is no evidence of
asymmetric effects of the correction mechanism between GDPt 0 and GDPt < 0 . Thus, according to the results of the selected
model (R2), INC1 has positive and significant effects on NC-EXP1 when GDPt 0 , and the effects are asymmetric during different
stages of the business cycle (Asy-p in the SR-asymmetry equals 0.046). For the highest-income households in Model (3.5), the effects
of ECTn5 and INC5 on NC-EXP5 are not significantly asymmetric during different stages of the business cycle.

7
K.-T. Lo, et al. North American Journal of Economics and Finance xxx (xxxx) xxx–xxx

Table 1(a)
Cointegration tests for total household consumption expenditure.
First-step regression (Eq.1st):

indep. \ dependent var. EXP0 EXP1 EXP5


constant −3.980 *** −0.500 ** 0.504 **
(1.018) (0.241) (0.244)
INC0 1.295 ***
(0.078)
INC1 1.040 ***
(0.019)
INC5 0.947 ***
(0.018)
BK86 4.522 ***
(1.060)
BK95 4.779 *** 5.502 ***
(1.000) (1.808)
BK96 7.931 ***
(0.986)
INC*BK86 −0.346 ***
(0.081)
INC*BK95 −0.337 ***
(0.073)
INC1*BK96 −0.616 ***
(0.078)
INC1*BK09 0.004 ***
(0.001)
INC5*BK95 −0.372 ***
(0.125)
ADF test on the residuals from (Eq.1st):
ADF test statistics = −7.202 *** −4.353 *** −4.748 ***
[0.000] [0.000] [0.000]

Note: *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively. The numbers in (.) and [.] indicate the standard errors and p-
values, respectively. BK86/BK95/BK96/BK09 indicates a dummy variable with a value of one for the periods since 1986/1995/1996/2009, and a
value of zero for other periods. These dummy variables are designed according to the results of multiple breakpoint tests (Bai and Perron, 1998,
2003).

Table 1(b)
Cointegration tests for household consumption expenditure.
First-step regression (Eq.1st):

indep.\dependent var. C-EXP0 C-EXP1 C-EXP5


constant 1.627 *** 1.167 *** 1.866 ***
(0.155) (0.256) (0.219)
INC0 0.855 ***
(0.012)
INC1 0.896 ***
(0.021)
INC5 0.834 ***
(0.016)
BK95 0.074 *** 5.700 ***
(0.008) (1.618)
BK96 9.896 ***
(0.698)
BK02 3.179 **
(1.391)
INC*BK02 −0.228 **
(0.101)
INC1*BK96 −0.771 ***
(0.055)
INC5*BK95 −0.388 ***
(0.112)
ADF test on the residuals from (Eq.1st):
ADF test statistics = −5.368 *** −4.802 *** −5.707 ***
[0.000] [0.000] [0.000]

Note: *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively. The numbers in (.) and [.] indicate the standard errors and p-
values, respectively. BK95/BK96/BK02 indicates a dummy variable with a value of one for the periods after 1995/1996/2002, and a value of zero
for other periods. These dummy variables are designed according to the results of multiple breakpoint tests (Bai and Perron, 1998, 2003).

8
K.-T. Lo, et al. North American Journal of Economics and Finance xxx (xxxx) xxx–xxx

Table 1(c)
Cointegration tests for household non-consumption expenditure.
First-step regression (Eq.1st):

indep.\dependent var. NC-EXP0 NC-EXP1 NC-EXP5


constant −41.719 *** −24.018 *** −40.468 ***
(8.689) (1.602) (8.780)
INC0 4.028 ***
(0.668)
INC1 2.753 ***
(0.130)
INC5 3.814 ***
(0.644)
BK84 34.051 *** 35.701 ***
(8.721) (8.813)
BK91 −0.23826 ***
0.06369
BK95 0.258 *** 0.245 ***
(0.024) (0.026)
BK01 37.650 ***
(4.421)
INC*BK84 −2.602 ***
(0.670)
INC1*BK01 −2.953 ***
(0.349)
INC5*BK84 −2.605 ***
(0.646)
ADF test on the residuals from (Eq.1st):
ADF test statistics = −4.612 *** −3.772 *** −5.630 ***
[0.000] [0.000] [0.000]

Note: *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively. The numbers in (.) and [.] indicate the standard errors and p-
values, respectively. BK84/BK91/BK95/BK01 indicates a dummy variable with a value of one for periods since 1984/1991/1995/2001, and a value
of zero for other periods. These dummy variables are designed according to the results of multiple breakpoint tests (Bai and Perron, 1998, 2003).

4. Discussion and conclusion

Although a number of studies have demonstrated the cyclical nature of household consumption, only a few empirical studies have
analyzed the patterns of household consumption under different economic states. Moreover, our study is one of the first to use time-
series household consumption data to examine the relationship between the patterns of household consumption and economic
changes for different types of household expenditure and different levels of household income.
We use a threshold model with error correction to characterize household consumption during different stages of the business
cycle. Using a general-to-specific modeling approach, we find that the asymmetric effects of the business cycle on household con-
sumption arise from short-run changes in household income and deviations from the long-term trend of household expenditure. In
addition, we use the HAC standard error to investigate whether household consumption displays asymmetric behavior. Using data
from the 1979–2014 Survey of Family Income and Expenditure in Taiwan, we first find that household consumption has pro-cyclical
characteristics during the business cycle in Taiwan, which is consistent with previous studies. Second, and more importantly, we find
that household consumption responds asymmetrically to economic fluctuations such that changes in household consumption tend to
be larger during expansions than during downturns. Moreover, this asymmetry differs across the different household income
quantiles.
In conclusion, in this study, we examine the asymmetric responses of household consumption to different stages of the business
cycle. A large body of literature has examined the relationships between economic variables over the phases of the business cycle,
which is an important topic for policy makers. This asymmetry has a number of implications. First, because the patterns of private
consumption are asymmetric over different phases of the business cycle, some of the theoretical economic models and empirical
econometric models need to be revised. For example, models with non-linear and asymmetric disturbances may be more appropriate.
Second, considering the asymmetric behavioral responses, the government should develop different policies to stimulate household
consumption during economic booms and recessions. In addition, given that the asymmetric effects are observed in households with
different levels of income, policies should also be designed to address the needs of different income quantiles.

9
K.-T. Lo, et al.

Table 2
Regression results for the change in the total household consumption expenditure ( EXPs).
Model(1.0) Model (1.1) Model (1.5)

full R1 R2 full R1 R2 full R1 R2


indep.\dependent var. ΔEXP0 ΔEXP1 ΔEXP5
constant 0.0038 0.0050 0.0038 0.0032 0.0045 * 0.0032 0.0051 0.0062 0.0053
(0.004) (0.004) (0.004) (0.003) (0.003) (0.003) (0.006) (0.006) (0.005)
ΔINCs·I(ΔGDP > 0) 0.9329 *** 0.9182 *** 0.9347 *** 0.9116 *** 0.9168 *** 0.9190 *** 0.8544 *** 0.8535 *** 0.8526 ***
(0.068) (0.064) (0.069) (0.080) (0.065) (0.082) (0.079) (0.083) (0.065)
ΔINCs·I(ΔGDP < 0) 0.2798 0.5935 *** 0.2412 0.4653 *** 0.5556 *** 0.2692 −0.4199 0.3406 −0.3237
(0.309) (0.133) (0.360) (0.148) (0.022) (0.301) (1.380) (0.550) (0.339)
ECTst−1·I(ΔGDP > 0) −0.0107 −0.0673 −0.7597 *** −0.5539 *** −0.3702 −0.3486
(0.537) (0.454) (0.132) (0.159) (0.307) (0.269)
ECTst−1·I(ΔGDP < 0) −0.7821 ** −0.9576 *** −4.5774 *** −4.4902 *** −0.2889 −0.9157 ***
(0.309) (0.099) (0.442) (0.095) (1.010) (0.297)
ECTst−1 −0.0657 −0.7877 *** −0.3676
(0.542) (0.145) (0.285)

10
SEX 0.1644 0.2334 −2.5890 * −2.5210 * 1.4368 1.4422
(1.547) (1.500) (1.455) (1.319) (1.608) (1.550)
AGE −0.0011 −0.0007 −0.0375 * −0.0374 * 0.0121 0.0123
(0.016) (0.015) (0.021) (0.022) (0.020) (0.019)
EDU 0.0199 0.0304 −0.1338 −0.0929 0.0032 0.0035
(0.093) (0.090) (0.088) (0.084) (0.105) (0.101)
M2 −0.0292 −0.0294 −0.0638 *** −0.0705 *** −0.0577 −0.0577
(0.034) (0.032) (0.016) (0.015) (0.038) (0.036)
GINI −0.7685 −0.8865 −1.2116 −1.6488 −0.8025 −0.7424
(0.904) (0.838) (1.372) (1.515) (1.464) (0.917)
R-squared 0.7876 0.7819 0.7834 0.8736 0.8270 0.8401 0.6783 0.6574 0.6783
Adjusted R-squared 0.7111 0.7529 0.7168 0.8281 0.8039 0.7909 0.5625 0.6118 0.5793
Asy-p in SR-asymmetry [0.060] * [0.061] * [0.089] * [0.025] ** [0.000] *** [0.051] * [0.373] [0.398] [0.001] ***
Asp-p in LR-asymmetry [0.028] ** [0.019] ** [0.000] *** [0.000] *** [0.942] [0.277]

Note: The s after INC and ECT equals 0, 1, and 5, corresponding to all households, the first group of households (group with the lowest income), and the fifth group of households (group with the highest
income), respectively. The numbers in (.) indicate the Newey and West (1987) HAC standard errors. The numbers in [.] indicate the corresponding p-values of the test statistic (Asy-p) for asymmetric
effects. In addition, *, **, and *** denote significance for tests at the 10%, 5% and 1% levels, respectively.
North American Journal of Economics and Finance xxx (xxxx) xxx–xxx
K.-T. Lo, et al.

Table 3
Regression results for the change in household consumption expenditure ( C-EXPs).
Model(2.0) Model(2.1) Model(2.5)

full R1 R2 full R1 R2 full R1 R2


indep.\dependent var. ΔC-EXP0 ΔC-EXP1 ΔC-EXP5
constant 0.0022 0.0024 0.0022 0.0023 0.0040 * 0.0023 0.0026 0.0033 0.0027
(0.003) (0.003) (0.003) (0.002) (0.002) (0.002) (0.004) (0.004) (0.004)
ΔINCs·I(ΔGDP > 0) 0.8539 *** 0.8611 *** 0.8547 *** 0.8213 *** 0.8225 *** 0.8240 *** 0.7942 *** 0.7947 *** 0.7929 ***
(0.069) (0.065) (0.070) (0.068) (0.053) (0.068) (0.062) (0.066) (0.061)
ΔINCs·I(ΔGDP < 0) 0.4108 0.5230 *** 0.3392 0.3417 *** 0.5026 *** 0.2758 ** −0.7180 0.9510 −0.5392
(0.261) (0.092) (0.322) (0.104) (0.020) (0.134) (2.533) (0.806) (0.422)
ECTcst−1·I(ΔGDP > 0) −0.4176 −0.3978 * −0.7232 *** −0.5998 *** −0.4458 −0.4513
(0.261) (0.223) (0.108) (0.139) (0.352) (0.303)
ECTcst−1·I(ΔGDP < 0) −1.0838 *** −0.9874 *** −1.3455 *** −1.3555 *** −0.3062 −1.4897 ***
(0.230) (0.078) (0.109) (0.031) (1.856) (0.518)
ECTcst−1 -(0.456) * -(0.792) *** -(0.445)
(0.260) (0.111) (0.342)

11
SEX −0.8972 −0.8890 −1.8505 * −1.9310 ** 0.7325 0.7349
(1.421) (1.378) (0.990) (0.912) (1.765) (1.715)
AGE −0.0093 −0.0091 −0.0302 * −0.0302 * −0.0029 −0.0028
(0.015) (0.015) (0.016) (0.016) (0.021) (0.020)
EDU −0.0691 −0.0587 −0.1321 ** −0.1210 * −0.0789 −0.0788
(0.088) (0.083) (0.062) (0.064) (0.119) (0.115)
M2 −0.0059 −0.0065 −0.0406 *** −0.0441 *** −0.0057 −0.0057
(0.021) (0.020) (0.012) (0.014) (0.031) (0.031)
GINI −0.5271 −0.6084 −1.7638 * −2.0176 * −0.9527 −0.9080
(0.838) (0.785) (1.004) (1.009) (1.332) (0.827)
R-squared 0.8581 0.8517 0.8548 0.9017 0.8646 0.8948 0.7262 0.7159 0.7262
Adjusted R-squared 0.8070 0.8319 0.8101 0.8664 0.8465 0.8624 0.6276 0.6780 0.6419
Asy-p in SR-asymmetry [0.119] [0.015] ** [0.147] [0.001] *** [0.000] *** [0.002] *** [0.552] [0.852] [0.002] ***
Asp-p in LR-asymmetry [0.004] *** [0.001] *** [0.001] *** [0.000] *** [0.943] [0.158]

Note: The s after INC and ECT equals 0, 1, and 5, corresponding to all households, the first group of households (group with the lowest income), and the fifth group of households (group with the highest
income), respectively. The numbers in (.) indicate the Newey and West (1987) HAC standard errors. The numbers in [.] indicate the corresponding p-values of the test statistic (Asy-p) for asymmetric
effects. In addition, *, **, and *** denote significance for tests at the 10%, 5%, and 1% levels, respectively.
North American Journal of Economics and Finance xxx (xxxx) xxx–xxx
K.-T. Lo, et al.

Table 4
Regression results for the change in household non-consumption expenditure ( NC-EXPs).
Model(3.0) Model(3.1) Model(3.5)

full R1 R2 full R1 R2 full R1 R2


indep.\dependent var. ΔNC-EXP0 ΔNC-EXP1 ΔNC-EXP5
constant 0.0168 0.0208 0.0168 0.0085 0.0139 0.0084 0.0173 0.0216 0.0182
(0.017) (0.016) (0.017) (0.010) (0.009) (0.010) (0.016) (0.014) (0.012)
ΔINCs·I(ΔGDP > 0) 1.2938 *** 1.2057 *** 1.2922 *** 1.5532 *** 1.4874 *** 1.5609 *** 1.0632 *** 0.9917 *** 1.0485 ***
(0.259) (0.208) (0.252) (0.417) (0.326) (0.409) (0.221) (0.194) (0.174)
ΔINCs·I(ΔGDP < 0) 0.5295 1.0955 * 0.5021 0.0538 0.3412 *** 0.0043 0.8922 0.4786 1.0075
(1.250) (0.552) (1.288) (0.474) (0.086) (0.495) (0.923) (0.643) (0.604)
ECTnst−1·I(ΔGDP > 0) −0.0890 −0.0633 −0.7294 *** −0.6638 *** −0.4597 −0.3477
(0.281) (0.258) (0.218) (0.198) (0.328) (0.320)
ECTnst−1·I(ΔGDP < 0) 0.0378 −0.5240 ** −1.0117 *** −1.0412 *** −0.2245 −0.8728 ***
(0.431) (0.205) (0.171) (0.038) (0.996) (0.316)
ECTnst−1 −0.0859 −0.7552 *** −0.4451
(0.272) (0.195) (0.292)

12
SEX 4.2117 4.1992 −1.3400 −1.4078 5.8160 5.8075
(2.825) (2.767) (4.563) (4.428) (3.514) (3.462)
AGE 0.0250 0.0250 −0.0357 −0.0364 0.0545 0.0558
(0.037) (0.036) (0.051) (0.051) (0.035) (0.033)
EDU 0.3057 0.3017 −0.1398 −0.1232 0.3408 0.3423
(0.207) (0.196) (0.280) (0.269) (0.211) (0.206)
M2 −0.1039 −0.1029 −0.2925 *** −0.2971 *** −0.1820 −0.1796
(0.119) (0.116) (0.101) (0.097) (0.130) (0.127)
GINI −1.0732 −1.0288 −2.5426 −2.8467 −0.8297 −0.4125
(2.838) (2.690) (4.083) (3.987) (3.083) (1.830)
R-squared 0.4896 0.4465 0.4894 0.6468 0.5739 0.6438 0.4222 0.3450 0.4217
Adjusted R-squared 0.3058 0.3727 0.3324 0.5196 0.5171 0.5342 0.2141 0.2577 0.2438
Asy-p in SR-asymmetry [0.602] [0.881] [0.597] [0.059] * [0.001] *** [0.046] ** [0.876] [0.513] [0.953]
Asp-p in LR-asymmetry [0.764] [0.024] ** [0.295] [0.061] * [0.834] [0.360]

Note: The s after INC and ECT equals 0, 1, and 5, corresponding to all households, the first group of households (group with the lowest income), and the fifth group of households (group with the highest
income), respectively. The numbers in (.) indicate the Newey and West (1987) HAC standard errors. The numbers in [.] indicate the corresponding p-values of the test statistic (Asy-p) for asymmetric
effects. In addition, *, **, and *** denote significance for tests at the 10%, 5%, and 1% levels, respectively
North American Journal of Economics and Finance xxx (xxxx) xxx–xxx
K.-T. Lo, et al. North American Journal of Economics and Finance xxx (xxxx) xxx–xxx

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.najef.2019.03.017.

References

Agenor, P. R. (2002). Business cycles, economic crises, and the poor. The Journal of Policy Reform, 5(3), 145–160.
Apergis, N., & Miller, S. M. (2006). Consumption asymmetry and the stock market: empirical evidence. Economics Letters, 93(3), 337–342.
Bai, J., & Perron, P. (1998). Estimating and testing linear models with multiple structural changes. Econometrica, 66(1), 47–78.
Bai, J., & Perron, P. (2003). Computation and analysis of multiple structural change models. Journal of Applied Econometrics, 18(1), 1–22.
Blundell, R., & Preston, I. (1995). Income, expenditure and the living standards of UK households. Fiscal Studies, 16(3), 40–54.
Çağlayan, E., & Astar, M. (2012). A Microeconometric analysis of household consumption expenditure determinants for both rural and urban areas in Turkey. American
International Journal of Contemporary Research, 2(2), 27–38.
Carroll, C. D., Slacalek, J., & Tokuoka, K. (2014). The distribution of wealth and the MPC: Implications of New European Data. American Economic Review, 104(5),
107–111.
Carroll, C. D., Slacalek, J., Tokuoka, K., & White, M. N. (2017). The distribution of wealth and the MPC: Implications of New European Data. Quantitative Economics,
8(3), 977–1020.
Carruth, A., & Dickerson, A. (2003). An asymmetric error correction model of UK Consumer Spending. Applied Economics, 35(6), 619–630.
Cook, S. (2002). Assymetric mean reversion in the consumption-income ratio: Evidence from OECD economies. Applied Econometrics and International Development,
2(2), 27–34.
Cutler, D. M., & Katz, L. F. (1992). Rising Inequality? Changes in the distribution of income and consumption in the 1980s. American Economic Review, 82(2), 546–551.
Davidson, R., & Monticini, A. (2014). “Heteroskedasticity and Autocorrelation Consistent Bootstrapping.” In Working Paper No.12, Dipartimento di Economia e
Finanza (DISCE), Università Cattolica del Sacro Cuore.
Diebold, F. X. (2007). Elements of forecasting (4th ed.). Independence, KY: Thomson South-Western.
Enders, W., & Granger, C. W. J. (1998). Unit-root tests and asymmetric adjustment with an example using the term structure of interest rates. Journal of Business and
Economic Statistics, 16(3), 304–311.
Enders, W., & Siklos, P. L. (2001). Cointegration and threshold adjustment. Journal of Business and Economic Statistics, 19(2), 166–176.
Engle, R. F., & Granger, C. W. J. (1987). Co-integration and error correction: Representation, estimation, and testing. Econometrica, 55(2), 251–276.
Harding, D., & Pagan, A. (2002). Dissecting the cycle: A methodological investigation. Journal of Monetary Economics, 49(2), 365–381.
Hodrick, R., & Prescott, E. C. (1997). Postwar U.S. business cycles: An empirical investigation. Journal of Money, Credit, and Banking, 29(1), 1–16.
Huntley, J., & Michelangeli, V. (2014). Can tax rebates stimulate consumption spending in a life-cycle model? American Economic Journal: Macroeconomics, 6(1),
162–189.
Jan, M. S. (2006). The structure of household consumption and food demand: A comparison of Taiwan households segmented by income. Agriculture and Economics, 36,
35–75.
Jappelli, T., & Pistaferri, L. (2000). Using subjective income expectations to test the excess sensitivity of consumption to predicted income changes. European Economic
Review, 44(2), 337–358.
Kuo, B. S., & Chung, C. T. (2002). Is consumption liquidity constrained? The asymmetric impact from business cycles. Academia Economic Papers, 30(4), 443–472.
Lin, H. Y., & Lo, K. T. (2012). Tax incentives and charitable contributions: The evidence from quantile regression. Pacific Economic Review, 17(4), 535–558.
Lustig, N., Fishlow, A., & Bourguignon, F. (2000). Presidential address: Crises and the poor: Socially responsible macroeconomics [with Comments]. Economia, 1(1),
1–30.
Miles, D. (1997). A household level study of the determinants of incomes and consumption. The Economic Journal, 107(440), 1–25.
Newey, W. K., & West, K. D. (1987). A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica, 55(3),
703–708.
Öcal, N., & Osborn, D. R. (2000). Business cycle non-linearities in UK consumption and production. Journal of Applied Econometrics, 15, 27–43.
Patterson, K. D. (1993). The impact of credit constraints, interest rates and housing equity withdrawal on the intertemporal pattern of consumption - A diagrammatic
analysis. Scottish Journal of Political Economy, 40(4), 391–407.
Proietti, T. (2005). New algorithms for dating the business cycle. Computational Statistics & Data Analysis, 49(2), 477–498.
Said, S. E., & Dickey, D. A. (1984). Testing for unit roots in autoregressive-moving average models of unknown order. Biometrika, 71(3), 599–607.
Shea, J. (1995). Myopia, liquidity constraints, and aggregate consumption: A simple test. Journal of Money, Credit, and Banking, 27(3), 798–805.
Souleles, N. S. (1999). The response of household consumption to income tax refunds. American Economic Review, 89(4), 947–958.
Tagkalakis, A. (2008). The effects of fiscal policy on consumption in recessions and expansions. Journal of Public Economics, 92(5–6), 1486–1508.

13

Potrebbero piacerti anche