Transportation Network Density, Domestic Market Integration and Excess Sensitivity of Household Consumption

2.1 Research on the Causes of Excess Sensitivity of Consumption

The rational expections version of the permanent income hypothesis, which holds that household consumption does not respond to changes of anticipated income, are challenged by empirical studies. A large number of studies find that changes of anticipated income can help predict changes of consumption, which is known as the phenomenon of excess sensitivity of consumption (Flavin, 1981). Existing research on the issue of excess sensitivity focus on explaining the reasons for the existence of excess sensitivity of household consumption. There are two main aspects of literature on the causes of excess sensitivity of consumption at the macro or micro level: First, uncertainty leads to the enhancement of precautionary savings motivation, which is an important factor causing excess sensitivity of consumption. From the perspective of uncertainty, the decline of income increases uncertainty, and the precautionary saving motivation increases, leading to a decrease in current consumption, so uncertainty makes consumption excessively sensitive to anticipated income, and the sensitivity coefficient is an increasing function of uncertainty (Shen and Liu, 2003; Deng and Jin, 2008). Second, liquidity constraints affect the availability of household credit, which is also a key factor leading to excess sensitivity of consumption. The liquidity constraint perspective argues that when faced with a temporary decline of income, households cannot or can only borrow less money from the credit market to smooth consumption, and can only decide their spending based on anticipated income, so liquidity constraints make income more important to consumption than the permanent income hypothesis predicts (Flavin, 1985; Campbell and Mankiw, 1990; Song et al., 2003; Einian and Nili, 2020). There is also a category of literature that focuses on the excess sensitivity of consumption caused by closed-economy constraints, which provides ideas for the subsequent analysis of transportation infrastructure to alleviate excess sensitivity of consumption through domestic market integration.

2.2 The Mechanism of Transportation Infrastructure Influencing Consumption Excess Sensitivity through Market Integration Channel

The cause of consumption excess sensitivity due to closed economic constraints suggests that the integration process of commodity markets can significantly alleviate consumption excess sensitivity. Ostergaard et al. (2002) find that in the United States and Canada, the degree of excess sensitivity of household consumption at the national level is much greater than that of household consumption at the state level, mainly due to closed economic constraints.

Dejuan and Luengo-Prado (2006) show that in addition to United States and Canada, closed economic constraints are also an important reason for the excess sensitivity of United Kingdom, Italy, West Germany, Spain, and Japan residents. Therefore, the higher the degree of market integration, the lower the degree of consumption excess sensitivity. However, China’s inter-provincial trade barriers are more serious than those among regions within United States or Canada, and the inter-provincial trade barriers are closer to those among EU countries or between developed countries such as United States and Canada (Liu and Hu, 2011), the domestic market is still severely fragmented, and the unified domestic market has not yet been established.

Although there are no differences in institutions, languages, currencies and cultures, inter-provincial borders continue to hinder trade with the outside world like a country’s borders (Poncet, 2003). Xu (2009) and Chen (2015) use China’s provincial macro data to show that the higher the degree of integration of commodity markets, the less the excess sensitivity of household consumption. These studies provide an important theoretical and empirical basis for explaining the determinant mechanism of commodity market integration in consumption excess sensitivity.

From the above domestic and foreign research, it can be seen that the segmentation of the commodity market reduces the convenience of obtaining goods, and in the face of a sudden increase in consumption demand, the price of commodities will inevitably rise significantly, and only the increased consumption demand of residents with higher incomes can achieve a new equilibrium, that is, consumption is excessively sensitive to income (Chen, 2015), and the improvement of transportation infrastructure can reduce the cost of transportation of goods among domestic regions, promote inter-provincial trade, break the geographical segmentation among provinces, and reduce the degree of segmentation of the domestic commodity market (Liu and Hu, 2011; Fan et al., 2017), and the increased integration of the domestic commodity market can help alleviate the excess sensitivity of household consumption (Ostergaard et al., 2002; Dejuan and Luengo-Prado, 2006; Xu, 2009). Based on the above analysis, the following hypotheses are proposed:

Hypothesis 1: Improvement of transportation infrastructure has a significant alleviating effect on the excess sensitivity of Chinese household consumption.

Hypothesis 2: Transportation infrastructure alleviates excess sensitivity of consumption through mechanism of increasing the degree of integration of the domestic market.

2.3 Heterogeneity of the Impact of Different Type of Transportation Infrastructure on Excess Sensitivity of Consumption

Furthermore, we focus on distinguishing the types of transportation infrastructure and exploring the heterogeneous effects of different types of transportation modes on excess sensitivity of consumption. Due to the constraints of long-distance and short-distance transportation capacity, the impact of transportation infrastructure varies among different types of transportation infrastructure. The improvement of transportation infrastructure can reduce the excessively high transportation costs among provinces, promote the process of regional economic integration, and promote the increase of regional trade volume can promote the increase of inter-provincial trade (Liu and Hu, 2011). If transportation infrastructure affect excess sensitivity of consumption through the mechanism of integrating the domestic commodity market, then the domestic commodity market segmentation mainly corresponds to the inter-provincial market segmentation. In general, railway transport is more suitable for long-distance freight trade than road transport, and road transport has an obvious substitution role for railway transport in short-distance freight trade, among which highway transport not only undertakes long-distance freight trade, but also bears a considerable share of short-distance freight trade. Specifically, the transportation infrastructure that inter-provincial long-distance transportation depends on railways first, highways second, and then substandard roads that solve the “1 km before and after” problem of inter-provincial transportation, and the dependence on other types of transportation modes is small. Therefore, long-distance transportation infrastructure may play a greater role in the integration of inter-provincial commodity markets. At this time, it should be found that the alleviating effect of transportation infrastructure on excess sensitivity of consumption is the largest in railways, highways second, and substandard roads third, and the impact of waterways and other types of highways with relatively fixed transportation networks is limited. Based on this, it can be seen that hypothesis 1 can also be verified from the perspective of heterogeneity of transportation infrastructure types. If hypothesis 1 is true, the following hypothesis should also be presented.

Hypothesis 3: There is heterogeneity in the impact of different types of transportation infrastructure on excess sensitivity of consumption, among which the alleviation effect of railways is the largest, the second is highway, the third is substandard road, and the impact of waterway and other grades of highway is limited.

3.1 Model Building

This section uses the China Family Panel Survey (CFPS) micro-household data and provincial-level data to further validate the relationship between the level of transportation infrastructure and the excess sensitivity of Chinese household consumption. In order to alleviate the potential interference of reverse causality on the empirical results, considering the time lag of each variable affecting household consumption, and in view of the biennial update of the CFPS database, the core explanatory variables and control variables are treated with two-year lag (one period of sample lag) in the model. Referring to the practices of Pozzi et al. (2004) and Dejuan and Luengo-Prado (2006), the benchmark empirical model is constructed as follows:

Among them, i, j and t denote for family, province, and year. The explained variable c_ijt represents the growth rate of non-durable goods consumption of household i in province j between t and t-2 years. The core explanatory variable y_ij,t-2 represents the income growth rate of household i in province j between t-2 and t-4 years, and α₁ is its coefficient. Referring to the practice of Ostergaard et al. (2002), this paper uses the lagged income growth rate to characterize the anticipated income change, and determines whether household consumption is excessively sensitive according to whether its estimation coefficient is significantly different from zero. The core explanatory variable rrw_j,t-2 represents the level of transportation infrastructure in province j where the family residence address is located in t-2 year, and α₃ is its coefficient. α₂ is the parameter to be evaluated for the interaction term rrw_j,t-2×y_ij,t-2, whose sign is expected to be significantly negative. CV_ij,t-2 is the control variable vector of the households, the heads of households, and the provincial-level characteristics that may affect the growth of household consumption, and α is the coefficient vector of the control variable vector, α₀ is a constant term, η_j, δ_t and μ_i are regional, temporal, and household fixed effects, and ε_jjt is a random error term.

3.2 Setting of the Main Variables

1. Level of transportation infrastructure. For the level of transportation infrastructure, a common measurement is to calculate the density of the transportation network. This paper follows Demurger’s (2001) practice of adding up the three types of transportation infrastructure: railways, roads, and inland waterways, and then dividing them by the administrative area of each province. The unit is km/km². However, according to statistics, the lack of navigable mileage data of inland waterways at the provincial level in China is relatively serious. Therefore, in addition to the above measurement methods, this paper also refers to the practice of Luo et al. (2018), which uses the total mileage of provincial railways and highways divided by the regional administrative area as another measure to test the robustness.

2. Household consumption growth and income growth. Referring to the method of Zang and Zhang (2018), the growth of household consumption as the explained variable is measured by the logarithmic difference of household non-durable goods consumption, while household non-durable goods consumption is composed of clothing expenditure, recreation and leisure expenditure, housing expenditure, miscellaneous goods and services expenditure, and other expenditure. The growth of household income as the core explanatory variable is measured by the logarithmic difference of household income, and the adjusted household net income is selected as household income, including operating income, wage income, property income, transfer income and other income. In fact, many factors may have an impact on the consumption behavior of Chinese residents. In order to alleviate the endogeneity problem that may be caused by the omission of variables, this paper refers to the existing research practices (Zang and Zhang, 2018) to further control the factors that may affect the growth of household consumption, such as household economic characteristics, household demographic characteristics, and macroeconomic environment of household location, in the regression equation. Table 1 lists the names, symbols, and calculation methods of the selected variables.

   Table 1

   Main Variable Definition

   	Variable name	Variable symbol	Calculation method
   Explained variables	Consumption growth rate	c	The first-order difference in the logarithm of household non-durable goods consumption
   Explanatory variables	Income growth rate	y	The first-order difference in the logarithm of household income
   	Transportation network density	rrw	The total mileage of railways, highway, and inland waterways divided by the administrative area
   	Rail and highway density	rr	The total mileage of railways and highway divided by the administrative area
   	Rail density	rail	The total mileage of the railroad divided by the administrative area
   	Highway density	road	The total length of the highway divided by the administrative area

   	Variable name	Variable symbol	Calculation method
   Explanatory variables	Waterway density	water	The total length of inland waterways divided by the administrative area
   	Expressway density	high	Expressway mileage divided by administrative area
   	Non-expressway density	nonhigh	Highway mileage minus expressway mileage divided by administrative area
   	Density of other grades of roads	othergrade	Other grade highway mileage minus expressway mileage divided by administrative area
   	Density of substandard road	outgrade	Substandard road mileage divided by administrative area
   Control variables	Family size	size	Number of family members
   	Household assets	asset	The logarithm of the sum of cash, savings, stocks, funds, and money owed to the family
   	Household debt	debt	The logarithm of the sum of bank loans, loans from relatives and friends, and private loans
   	Gender	gender	1 for males and 0 for females
   	Age	age	The age of the year of the survey
   	Age squared term	ages	Age squared
   	Level of education	edu	Years of schooling
   	Type of household registration	hukou	The non-agricultural household registration is 1, and the agricultural household registration is 0
   	Marital status	married	1 for married, 0 for otherwise
   	Type of work	jobtype	The work in the government, public institutions or state-owned enterprises is 1, otherwise it is 0
   	Political outlook	member	1 for CCP membership, 0 otherwise
   	GDP per capita	gdp	The logarithm of GDP per capita in the area where the household is located
   	Price level	cpi	Consumer Price Index (CPI) for the area where the household is located
   	Real interest rates	interest	The one-year savings deposit rate minus the inflation rate in the area where the household is located
   	Level of financial development	credit	The balance of urban and rural RMB deposits of financial institutions in the area where the household is located divided by GDP

4.1 Benchmark Regression Results

The full-sample baseline regression results of transportation infrastructure affecting excess sensitivity of household are shown in Table 2. Table 2 column (1) shows the estimation results that control for household economic characteristics and demographic characteristics, and the coefficient of household income growth (y) is significantly positive at the level of 1%, and income growth has a positive effect on consumption growth, indicating that household consumption is excessively sensitive to changes of anticipated income. The coefficient of the interaction term (rrw×y) of transportation infrastructure and household income growth, which is the core explanatory variable, is significantly negative at the 1% level, indicating that compared with the areas with low level of transportation infrastructure development, the degree of excess sensitivity of household consumption is significantly smaller in areas with high level of transportation infrastructure development, that is, the level of transportation infrastructure development has a negative impact on excess sensitivity of consumption. Considering that there may be missing variables, column (2) controls for the characteristics related to the external economic environment in the area where the household is located, which is consistent with the results of column (1). By comparing the results of the two columns, it can be found that the numerical magnitude and significance of the interaction coefficients are relatively stable. The above regression results consistently show that the excess sensitivity of households in the developed areas of transportation infrastructure is lower, indicating that the improvement of transportation infrastructure conditions can help alleviate the excess sensitivity of households in the areas along the route. This is consistent with the conclusion of the theoretical analysis, and the research hypothesis 1 is valid.

The results in Table 2 show that the regression coefficients of the interactive term (rrw×y) of transportation infrastructure and income growth are between –0.0109 and –0.0167, which is of significant economic significance, which means that the level of transportation infrastructure development has a very important impact on the excess sensitivity of household consumption.

The reason is that the domestic inter-provincial commodity circulation is mainly transported from the domestic origin to the local consumer market through transportation infrastructure such as railways, highways, and substandard roads, and the improvement of transportation infrastructure can reduce the cost of goods transportation among different regions in the country, weaken the degree of domestic market segmentation, promote the integration of the domestic market, and ultimately reduce the dependence of household consumption on income, and alleviate the excessive sensitivity of consumption. In conclusion, the estimates in Table 2 support the hypothesis 1 that improved transportation infrastructure is an important cause of alleviating excess sensitivity of household consumption.

4.2 Endogenous Problem Handling: Multidimensional Instrumental Variables

The benchmark regression results confirm the important role of transportation infrastructure in alleviating excess sensitivity of household consumption. Potential endogeneity issues in the model can affect the accuracy of the regression results. This paper makes a multi-dimensional attempt to solve the problem of endogeneity.

First, the geographical slope. Referring to the practice of Sun et al. (2019), this paper selects the geographic slope that satisfies both correlation and exogenity assumptions as the instrumental variable. On the one hand, the higher the geographical slope, the more difficult it is to construct transportation infrastructure, and the less likely it is to build. On the other hand, the geographic slope with natural exogenous characteristics is a geographic information variable that characterizes the topographic differences of the sample provinces, which is difficult to directly affect the consumption behavior of micro households, so it can be used as a good instrumental variable for transportation facilities.

Considering that the geographic slope data is an inherent natural feature of the sample provinces and does not change over time, this paper refers to the treatment method of time-invariant instrumental variables (Duranton and Turner, 2012), and takes the interaction term of the original variable and the time trend as the instrumental variable.Besides, we also use policy preference index, the historical level of transportation infrastructure, the mortality rate during the “Great Famine”, and the method of shift-share, as instrumental variables to overcome the endogeneity problem. All the results are robust with the benchmark regression results.

4.3 Robustness Test

First, indicator selection problem. Considering the relative importance of railways and highways among the types of transportation infrastructure in China and the lack of navigable mileage data of inland waterways, this paper uses other methods to re-evaluate the transportation infrastructure to test the robustness of the research conclusions. Referring to the practice of Luo et al. (2018), this paper uses the sum of the mileage of railways and highways divided by the administrative area of the province to calculate the density of transportation infrastructure. Column (1) of Table 3 reports the estimates of the explanatory variables for the replacement of transport infrastructure. It can be seen that the estimation coefficients of the interaction terms of transportation infrastructure and income growth are all significantly negative, which once again proves that the conclusion is robust.

In addition, there is a lack of consistency in how anticipated income growth is measured, and it is common to use current income growth to characterize anticipated income growth. To ensure the robustness of the core conclusions, this paper replaces the lagging income growth with the current income growth and to re-estimate. As shown in the regression results in column (2) of Table 3, the change in measure indicator of the anticipated income growth do not have a material impact on the core conclusions. Furthermore, considering that the different lag periods of the explanatory variables may affect the core conclusions, this paper further conducts robustness test by replacing the lag periods of the explanatory variables. The regression results are shown in column (3) of Table 3, and it can be found that the results of the core explanatory variables and control variables have not changed significantly after a four-year lag (the sample lags two periods), so the estimates are still robust.

Second, sample selection problem. The distribution of transportation infrastructure network is often not exogenous, and the more developed the political center and the more economically developed the area, the more complete the development of transportation infrastructure, resulting in the non-random location of transportation line construction. In order to increase the randomness of the sample data and avoid the endogeneity problem that may be caused by reverse causality, all the household samples of the four municipalities were excluded from the whole sample and re-estimated, and the results are shown in column (4) of Table 3. It can be seen that the estimation results are robust for excluding the sample of municipalities, so the core conclusions have high reliability.

Third, re-examination based on provincial macro household consumption data. In order to examine the robustness of the conclusions based on micro data, we re-examine the provincial household consumption data from 1978 to 2017. The construction method of the Transport Infrastructure Index (rrw) is consistent with the above, and the calculation steps for urban household consumption growth (c) and income growth (y) are as follows: The month-on-month consumer price index of urban residents is converted into a fixed-base ratio consumer price index (100 in 1978), and then the nominal per capita consumption (income) of urban residents is deflated by the fixed- basis ratio consumer price index to obtain the actual value, and then the logarithm is taken for differential. Referring to the existing literature (Wang and Guo, 2011), in addition to the real interest rate and GDP per capita, other control variables are added to the regression equation. The total dependency coefficient (depen) is depicted by “(population under 15 years old + over 64 years old) / (population under 15~64 years old) × 100”. The level of urbanization (urban) is measured by using the ratio of the urban population to the total population. The scale of fiscal expenditure (gov) is expressed as “general public budget expenditure/GDP”. The proportion of industrial output (industry) is expressed as “industrial value added/GDP”. Housing system reform (house) is set as a dummy variable, taking 1 after 1998 and 0 in other years. In addition, during the sample period from 1978 to 2017, the global financial crisis has an adverse impact on the international economy and has a negative impact on China’s economy. Considering the possible impact of this event, in order to ensure that the conclusion is not caused by some particularities of the global financial crisis, this paper sets the period from 2008 to 2010 as the period of the global financial crisis, and re-estimates based on the subsample after excluding the period of the global financial crisis. The results, as shown in columns (5) and (6) of Table 3, show that the core conclusions still hold.

4.4 Heterogeneity Analysis: Different Types of Transportation Infrastructure

Combined with the actual background of China’s cargo transportation, the type of transportation infrastructure selected for cargo transportation will be heterogeneous due to the difference in the length of the geographical distance of freight, so the impact of transportation infrastructure on excess sensitivity of consumption may be related to the type of transportation infrastructure. From the perspective of heterogeneity of transportation modes, this paper examines the asymmetric impact of different types of transportation infrastructure on excess sensitivity of consumption, that is, distinguishing different types of transportation infrastructure and grouping regression to different types of transportation infrastructure and excess sensitivity of consumption. In this paper, the transportation infrastructure is divided into three types of infrastructure: railway, waterway and highway, and the highway is divided into expressway and non-expressway, and the non-expressway is further divided into other grades roads and substandard roads, and the ratio of the operating mileage of various types of transportation infrastructure to the administrative area is used as the transportation network density index, and the benchmark model (1) is used to regress, and the specific results are shown in Table 4. Columns (1)-(7) of Table 4 report the heterogeneous effects of railways, waterways, highways, expressways, non-expressways, other grade roads, and substandard roads on excess sensitivity of consumption, respectively. The results show that after controlling for the main characteristic variables at the household, household head and provincial levels, railways, highways, substandard roads, waterways and other grade roads can significantly alleviate the excess sensitivity of consumption. The mitigation effects are –1.8531, –1.5909, –0.1517, –0.1445 and –0.0067, respectively. From an economic point of view, there are obvious differences in the alleviating effect of different types of transportation infrastructure on excess sensitivity of consumption, with railway having the greatest alleviating effect on excess sensitivity of consumption, expressways second, substandard roads third, waterways and other grades roads playing a limited role, which verifies the hypothesis 3 of this paper.

5.1 Mediator Effect Test

Based on the analysis of the influencing mechanism above, we select the degree of domestic commodity market integration (m) as the mediating variable and use the domestic commodity market segmentation index, the most commonly used indicator, to measure it. The smaller the market segmentation index value, the higher the level of integration of the domestic commodity market. The mediating effect model for testing the channel of the domestic commodity market integration is set as follows:

Among them, The mediator variable m_it represents the degree of the integration of the domestic commodity market of province i in t year, which is depicted by the domestic commodity market segmentation index calculated by the relative price method. The steps of the mediating effect are as follows: first, testing the impact of transportation infrastructure on the excess sensitivity of household consumption, according to the previous theoretical analysis, it is significantly negative, and the model setting is consistent with the model (1) above. Second, referring to the practice of Fan et al. (2017), model (2) introduces the scale of fiscal expenditure (gov) measured by the share of local general public budget expenditure of the provincial level to GDP, the openness of international trade (open) and its squared term (opens) control variables measured by imports and exports to GDP.

γ₂ indicates the impact of transport infrastructure on the mediating variable, and if γ₂ is consistently significant, it indicates that the intermediary effect exists, i.e., transport infrastructure does significantly affect the excess sensitivity of household consumption by affecting the degree of integration of domestic commodity markets. Thirdly, in model (3), test whether μ₂ and μ₅ are significant, if μ₂ is not significant and μ₅ is significant, it means that there is a complete mediating effect on the mediating variable, and if both μ₂ and μ₅ are significant, it means that there is a partial mediating effect.

Table 5 reports the estimates of the mediating effect model. Column (1) of Table 5 corresponds to the regression results of model (1) and is the same as column (2) of Table 2. Column (2) of Table 5 corresponds to the regression results of model (2). Column (3) of Table 5 shows the estimation results after the addition of the mediating variable m. Column (2) shows that the estimated coefficient of transportation infrastructure is significantly negative at the 1% level, indicating that the improvement of transportation infrastructure significantly contribute to the integration of the domestic market. The estimation results in column (3) show that compared with the baseline regression results in column (1), the absolute value of the estimation coefficient of the interaction terms of transportation infrastructure and income growth decreases after the introduction of the degree of domestic market integration in column (3), indicating the existence of partial mediating effects on the level of domestic market integration, that is, weakening the degree of segmentation of the domestic commodity market is an important channel for transportation infrastructure to affect excess sensitivity of consumption, thus verifying the hypothesis 2 of this paper.

5.2 Robustness Test of Influencing Channels

First, replace the market integration indicator. Considering that the prominent characteristics of the process of domestic market integration are: trade barriers and border effects between regions are gradually weakening, and cross-regional trade of commodities is more frequent. Referring to the method of Wang and Cen (2022), we also depict commodity market integration through the ratio of freight turnover of the regional logistics linkage index to GDP, and the results of the re-empirical estimation based on the new indicators are shown in columns (4) and (5) of Table 5. It can be seen that the estimation coefficients of all core explanatory variables are consistent with the previous theoretical expectations, indicating that the conclusion of the mechanism analysis in this paper is robust.

Second, change the analysis method of mechanism. Another method of mechanism analysis is step-by-step regression. Specifically, the effects of transportation infrastructure, domestic market segmentation, and excess sensitivity of consumption are regressed separately. Firstly, the domestic market segmentation is used to regress on the transportation infrastructure, and then the domestic market segmentation is regressed on by the excess sensitivity of consumption, and the impact of the transportation infrastructure on the excess sensitivity of consumption through the market integration channel can be revealed based on the estimation results. If the regression coefficients of all steps are significant, this influence channel is confirmed. If the regression coefficients of all steps are not significant or at least one of the estimated coefficients is not significant, this influence channel is not valid. Columns (2) and (3) of Table 5 show that the regression coefficient of the core explanatory variable is significant at least 5%, indicating that domestic market integration is an important channel for transportation infrastructure to alleviate excess sensitivity of consumption.

Another method of mechanism analysis is the addition of the interaction term method. Specifically, firstly, testing whether the core explanatory variables act on the domestic market segmentation, and secondly, the interaction terms of the core explanatory variables and the domestic market segmentation are constructed to test the action channel. This paper further refers to the method of Ma and Zhang (2017) to test the robustness of the influencing channels: firstly, testing whether the transportation infrastructure weakens the degree of segmentation of the domestic commodity market, and the model setting is consistent with the previous model (2), and secondly, examining the effect of transportation infrastructure construction on the excess sensitivity of household consumption by affecting the degree of domestic commodity market integration. The model is set to:

Among them, rrw_j,t-2 × y_ij,t-2 × m_j,t-2 represents the triple interaction term of transport infrastructure network density, household income growth, and domestic commodity market segmentation index, and other variables are consistent with the previous section. Column (6) of Table 5 shows that the estimation coefficient of the additional triple interaction term is significantly negative at the level of 1%, indicating that the improvement of the degree of the domestic commodity market integration is indeed a channel for transportation infrastructure to alleviate the excess sensitivity of household consumption, and thus fully verifies the robustness of the impact channel.