A Study on the Influence of Borrowing on Household Consumption Expenditures: A Layered Comparison from the Perspective of Alleviating Relative Poverty

Abstract

This study investigates how borrowing influences household consumption patterns in low-income and formerly impoverished regions of China, with implications for sustainable development goals (SDGs) such as poverty reduction (SDG 1), inclusive economic growth (SDG 8), and reducing inequalities (SDG 10). Using four rounds of 7516 households of balanced panel data from the China Family Panel Studies (CFPS), a fixed-effects model was employed to analyze the impact of borrowing on total and categorized household consumption. Instrumental variable and treatment-effects models were applied to address endogeneity issues. Results reveal that a 1% increase in borrowing boosts total household consumption by 0.026%, with stronger effects on developmental and investment-oriented consumption, particularly among low-income households. Formerly impoverished areas experienced greater consumption growth, especially in education and training, indicating higher domestic demand potential and faster human capital accumulation. The study concludes that improving financial markets, aligning credit with specific needs, and leveraging local resources are essential for upgrading consumption structures and alleviating relative poverty.

1. Introduction

By the end of 2020, China had historically eliminated absolute poverty. However, this does not signify the end of its poverty governance efforts; relative poverty will persist for a long time due to social stratification. The Fourth Plenary Session of the 19th Central Committee of the Communist Party of China set forth the goal of “resolutely winning the battle against poverty, consolidating the results of poverty alleviation, and establishing a long-term mechanism to address relative poverty”, indicating that after 2020, China’s poverty governance will enter a new phase focused on mitigating relative poverty. According to calculations based on the stratified linear random sampling data (CHIP2019) from the Institute of Income Distribution at Beijing Normal University, the number of people in China with a per capita household disposable income below 1090 yuan per month reached approximately 599.92 million, of whom 75.62% were rural residents. Despite China’s remarkable achievements in eliminating absolute poverty, structural challenges such as weak foundations, insufficient infrastructure, and uneven development still persist; in other words, China remains a developing country characterized by a predominantly low-income population. Consequently, in the new stage of development, consolidating and enhancing the outcomes of poverty alleviation and continuously mitigating relative poverty remain formidable tasks. In the relative poverty governance stage, the principal policy targets expand to low-income groups, primarily in rural areas. Such groups often exhibit a stronger demand for borrowing but are constrained by limited borrowing capacity. They generally can access lower borrowing amounts and face greater difficulty in obtaining loans compared to other segments of the population, frequently restricted by formal lending practices [1,2]. Theoretically, inflows of credit enhance capital availability in impoverished regions and help break the cycle of poverty [3,4], strengthening the self-sustaining capacity of impoverished populations [5], buffering negative income shocks, and stabilizing household welfare [6]. Hence, alleviating relative poverty entails not only improving social security from an income perspective but also meeting the consumption needs of low-income groups from a consumption perspective. By progressively improving the living standards of low-income families, they can more fully share in the fruits of economic and social development. Against this backdrop, this paper explores the influence of borrowing on household consumption from the perspective of relative poverty and provides practical recommendations. Such an approach is of great significance for enhancing the consumption capacity of low-income groups and mitigating relative poverty.

The existing research on the impact of borrowing on household consumption expenditure primarily focuses on four dimensions. First, regarding the direct effects of borrowing on household consumption expenditure. The direct influence manifests through how debt fluctuations affect changes in household spending behavior [7]. Borrowing can alleviate household liquidity constraints, thereby promoting increased consumption expenditure [2,8]. Studies indicate that a 1 percentage point reduction in credit constraint intensity increases rural households’ non-essential consumption expenditure by 55.77 yuan [9]. However, some scholars argue that debt repayment obligations reduce household cash flow and heighten uncertainty about future credit accessibility, thereby forcing households to curtail expenditures [10].

Second, concerning the indirect effects of borrowing on household consumption expenditure. On one hand, households can utilize borrowed funds for productive and operational activities to boost income, subsequently driving consumption growth [6]. Moreover, the impact of actual income growth on consumption is long term and sustainable [11]. On the other hand, financial borrowing may generate psychological wealth effects that stimulate consumption expenditure, with such psychological wealth effects demonstrating stronger consumption promotion than real wealth accumulation [12].

Third, examining borrowing’s impact on household consumption structure. Some scholars posit that borrowing facilitates consumption structure upgrading [11,12], transforming household consumption patterns from “subsistence-oriented” to “enjoyment-oriented” [13]. Conversely, others contend that increased debt levels exacerbate repayment pressures, significantly inhibiting development and leisure-oriented consumption while impeding consumption structure optimization [7]. Additionally, credit’s impact on consumption structure varies across financing channels. Formal financial institutions tend to implement cautious lending practices toward rural households’ improvement-oriented consumption demands for risk control purposes, thereby limiting formal credit constraints’ influence on consumption structure [14]. In contrast, broad credit constraints (including both formal and informal) negatively affect consumption structure optimization in rural households [15].

Fourth, heterogeneity analysis of borrowing’s consumption effects. Regionally, borrowing demonstrates stronger consumption-promoting effects in central and western rural areas compared to eastern regions [16]. Urban–rural disparities reveal that borrowing reduces total urban household consumption while increasing rural households’ overall consumption expenditure and subsistence-oriented spending [17].

Overall, existing research has primarily focused on the effects of borrowing on specific population groups or specific types of lending products and has rarely analyzed, from a relative poverty perspective, how borrowing affects the consumption expenditures of different household types or regions. However, household consumption expenditure, reflecting household well-being and living conditions, manifests notable heterogeneity and typicity. It remains unclear how borrowing products that theoretically can improve household welfare and mitigate relative poverty affect households in various levels of relative poverty and in regions facing differing degrees of relative poverty.

Based on this, from the vantage point of alleviating relative poverty, this paper aims to clarify the mechanism through which household borrowing influences household consumption expenditures, establish a theoretical framework, and empirically examine the group and regional effects of borrowing on households in different degrees of relative poverty and in areas with differing relative poverty conditions. Meanwhile, to address endogeneity issues arising from reverse causality and “self-selection”, we employ the instrumental variable method and a treatment-effects model, followed by robustness tests. Compared with existing research, this paper offers three primary contributions. First, using nationally representative data, it provides a novel perspective by exploring differences in the effects of borrowing on households at various levels of relative poverty and in regions with differing relative poverty statuses. Second, in terms of methodology, the paper not only uses the instrumental variable method to overcome endogeneity caused by reverse causality but also adopts a treatment-effects model to address estimation bias resulting from self-selection, providing a methodological reference for identifying the endogeneity of household borrowing. Finally, the finding that “borrowing more significantly promotes developmental and investment-oriented consumption among low-income households” offers valuable insights for formulating national credit policies and optimizing the consumption structures of low-income families in the relative poverty stage.

2. Theoretical Analysis

2.1. The Direct Mechanism by Which Borrowing Affects Household Expenditure Structure

Based on the consumer intertemporal choice framework, Dynan [18] proposes a precautionary savings model in which the variance of changes in consumption expenditure measures uncertainty in expenditures, thus examining precautionary savings. The model assumes that the utility function is additively separable over time, with $u^{\prime }>0$, $u^{″}<0$, and $u^{‴}>0$, whereas labor income is uncertain. At period $t$, the dynamic optimization problem of consumption for a representative household can be expressed as:

$$\mathit{max}{E}_t\left[{\displaystyle \sum _{j=0}^{T-t}}{\left(1+\delta \right)}^{-j}U\left(C_{i,t+j}\right)\right]$$

(1)

$$s.t. A_{i,t+j+1}=(1+r_i)A_{i,t+j}+Y_{i,t+j}-C_{i,t+j}$$

(2)

In Equation (1), $E_t$ is the expectation based on the information set in period $t$, $T$ denotes the length of the lifespan, and t + j represents a particular period within the lifespan. $\delta$ is the time preference rate (assumed to be a constant), and $C_{i,t}$ is the consumption of a given consumer in period t. In Equation (2), $A_{i,t}$ is the nonhuman wealth of a consumer in period $t$, and $A_{i,t+1}=0$ is known. $r_i$ is the after-tax real interest rate; $Y_{i,t}$ is labor income (which is uncertain). Additionally, the utility function is additive separable over time and concave, and labor income is assumed to be uncertain.

Using the Bellman equation of dynamic optimization to solve this consumption model yields the first-order condition in $j=1$:

$${\displaystyle \frac{1+r_i}{1+\delta }}{E}_t\left[u^{\prime }\left(C_{i,t+1}\right)\right]=u^{\prime }{C}_{i,t}$$

(3)

Applying the second-order Taylor expansion to $u^{\prime }\left(C_{i,t+1}\right)$ gives:

$$u^{\prime }\left(C_{i,t+1}\right)=u^{\prime }{C}_{i,t}+u^{″}\left(C_{i,t}\right)\left(C_{i,t+1}-C_{i,t}\right)+{\displaystyle \frac{1}{2}}{u}^{‴}{C}_{i,t}{\left(C_{i,t+1}-C_{i,t}\right)}^2+o(C_{i,t})$$

(4)

Neglecting higher-order terms in the Taylor expansion and substituting Equation (4) into Equation (3) yields:

$$E_t\left[{\displaystyle \frac{C_{i,t+1}-C_{i,t}}{C_{i,t}}}\right]={\displaystyle \frac{1}{\zeta }}\left[{\displaystyle \frac{r_i-\delta }{1+r_i}}\right]+{\displaystyle \frac{\rho }{2}}{E}_t\left[{\left[{\displaystyle \frac{C_{i,t+1}-C_{i,t}}{C_{i,t}}}\right]}^2\right]$$

(5)

Let $\zeta =-C_{i,t}(u^{″}/u^{\prime })$ be the coefficient of absolute risk aversion, and define −$\rho =-C_{i,t}(u^{‴}/u^{\prime })$ as the coefficient of relative prudence. Equation (5) provides a method to estimate the intensity of precautionary savings using panel data on consumption. Suppose $G{C}_{i,t}$ is the growth of an individual’s consumption of period $t$, $M$ is the total number of periods, and $u_i$ and $v_i$ represent error terms from replacing the sample mean with the expected value and from preference shocks influencing the marginal utility of consumption, respectively. Then:

$$G{C}_{i,t}=\left[{\displaystyle \frac{C_{i,t+1}-C_{i,t}}{C_{i,t}}}\right]$$

(6)

Substituting Equation (6) into Equation (5) yields:

$${\displaystyle \frac{1}{M}}{\displaystyle \sum _{t=1}^M}G{C}_{i,t}+u_i={\displaystyle \frac{1}{\zeta }}\left[{\displaystyle \frac{r_i-\zeta }{1+r_i}}\right]+{\displaystyle \frac{\rho }{2}}\left[{\displaystyle \frac{1}{M}}{\displaystyle \sum _{t=1}^M}G{C_{i,t}}^2\right]+v_i+{\eta }_i$$

(7)

where consolidating the error terms gives:

$$avg(GC{)}_i={\displaystyle \frac{1}{\zeta }}\left[{\displaystyle \frac{r_i-\zeta }{1+r_i}}\right]+{\displaystyle \frac{\rho }{2}}avg(G{C}^2{)}_i+{\epsilon }_i$$

(8)

Given the initial assumptions $u^{\prime }>0$, $u^{″}<0$, and $u^{‴}>0$, we can see theoretically that $\rho$ should be a positive value. According to Equation (5), consumers’ expected uncertainty about future expenditures is positively correlated with the current period’s savings rate. Equation (8) indicates that if consumers in period $t$ anticipate increased uncertainty in period $t+1$, leading to higher expenditures in period $t+1$, they will reduce current consumption (all else being equal) to increase precautionary savings in order to address future uncertainties. This means that when consumers expect tighter borrowing constraints in the next period, they anticipate more difficulty in obtaining loans in the future and thus undertake precautionary savings in response. Conversely, when consumers’ borrowing constraints are relaxed, precautionary savings decline, thereby boosting consumption.

Based on this theoretical reasoning, we propose the first research hypothesis:

H0: 

Borrowing has a positive effect on total household consumption.

2.2. Mechanism of the Indirect Influence of Borrowing on Household Expenditure Structure

Borrowing primarily affects low-income groups by relaxing borrowing constraints, making it easier for these groups to obtain financial support, and indirectly influencing their expenditure structures via intermediary effects. To more clearly illustrate the relationship between borrowing and income growth among low-income groups, Wang [19] introduces a theoretical model that incorporates “financial exclusion of low-income groups” into the original theoretical framework. Meanwhile, it is assumed that households exhibit varying levels of production efficiency, some high, some low, leading to differences in the amount of capital obtained. Consequently, families with lower production efficiency also have lower income levels, thus placing them in a relatively disadvantaged position.

Suppose there are $N$ households in an economy during period $t$. The $i$ household possesses $e_i$ units of capital, and the total capital stock can be written as:

$$K_t={\displaystyle \sum _{i=1}^N}{e}_i$$

(9)

At this point, the production function of the $i$ household can be expressed as:

$$y_t={\tau }_i{k}_i$$

(10)

where ${\tau }_i$ denotes the household’s production efficiency, which depends on factors such as the household’s overall educational attainment and the region’s level of economic development. When both $a$ and $b$ are natural numbers, the following conditions hold:

$${\tau }_a>{\tau }_b$$

(11)

$$s.t.a>b$$

(12)

Let ${\tau }_m$ be the production efficiency of $m$, the marginal production household. Households above $m$ are considered high-income, while those below $m$ are considered low-income. To maximize returns, a rational household prefers to select a capital usage $k_j$. Accordingly, its production behavior can be represented as:

$$\mathit{max}\{{\tau }_i{k}_i-r(k_i-e_i\left)\right\}$$

(13)

$$s.t.k_i<<v{e}_i$$

(14)

In Equation (13), $r$ denotes the local market interest rate. Equation (14) specifies the borrowing constraint for the $i$ household, where $v$ denotes the “financial exclusion” factor and reflects the degree of financial exclusion in the given economy ($1\le v\le +\infty$). The larger the value of $v$, the lower the degree of financial exclusion and, hence, the weaker the borrowing constraints on low-income groups, increasing the likelihood that low-income households can obtain credit. When $v=1$, financial exclusion is severe, indicating high borrowing constraints on low-income groups, making it difficult for them to secure financing. Under this scenario, a single household can only rely on its initial capital to produce. When $v=+\infty$, the economy exhibits no financial exclusion; in other words, even low-income households face no credit constraints.

When the total capital utilized equals the total capital stock $k$, the financial market reaches equilibrium. If the equilibrium interest rate $r$ equals the production efficiency ${\tau }_m$ of the marginal production household $m$, the above equilibrium can be achieved. The total capital usage consists of the capital used by the marginal production household $m$ plus the maximum amount used by the high-income households. Under these conditions, the equilibrium can be expressed as:

$$K_m+v{\displaystyle \sum _0^{m-1}}{e}_i=K_i$$

(15)

and given $0\le K_m\le v{e}_m$, the financial market equilibrium condition is:

$${\displaystyle \sum _0^{m-1}}{e}_i\le K_i/v\le {\displaystyle \sum _0^m}{e}_t$$

(16)

As financial exclusion gradually eases, low-income families can obtain financial services on an equal footing. In other words, as $v$ increases, high-production-efficiency/high-income households will not fully use the maximum capital they could borrow; the remaining capital can be used by households with lower production efficiency. Consequently, some lower-efficiency, lower-income families will continuously replace the previously identified marginal production household $m$, becoming the new marginal production households. The total output at this point can be written as:

$${{\rm Y}}_t={\tau }_m{k}_t+v{\displaystyle \sum _0^{m-1}}({\tau }_i-{\tau }_m)e_i$$

(17)

Because ${\tau }_t>{\tau }_m$, for all rural households $i<m$, then we have:

$${\displaystyle \frac{\partial {{\rm Y}}_t}{\partial v}}{\displaystyle \sum _0^{m-1}}({\tau }_i-{\tau }_m)e_i>0$$

(18)

Equation (18) indicates that as financial exclusion declines, households with lower production efficiency and lower income can acquire more capital through financial channels to increase their incomes. The prerequisite is that the production efficiency exceeds the cost of using financial capital, ${\tau }_j>r$.

From the above theoretical model, we conclude that borrowing aimed at low-income groups can influence household expenditure structures by raising their income levels.

Based on this theoretical reasoning, we propose the following hypotheses:

H1-1: 

Borrowing exerts a more pronounced positive effect on total consumption among low-income households and in areas that have recently overcome poverty; conversely, the effect is less pronounced among higher-income households and non-poverty regions.

H1-2: 

Borrowing positively promotes educational/training expenditures for low-income households and in formerly impoverished areas; conversely, the effect is less pronounced in higher-income households and non-poverty regions.

H1-3: 

Borrowing positively promotes agricultural production expenditures among low-income households and in formerly impoverished areas; conversely, the effect is less pronounced for higher-income households and non-poverty regions.

2.3. Mechanisms Underlying the Differential Impact of Borrowing on Household Expenditure Structures Across Varying Levels of Relative Poverty

Credit constraints affect household expenditure structures not through a single channel but via two distinct pathways: a direct effect and an indirect effect. Because different types of consumption items exhibit varying sensitivities to external factors, the composition of consumption expenditures changes accordingly. From the perspective of a rational household, consumption behavior aims to maximize utility by choosing the kinds and quantities of goods and services, subject to the household’s income and current prices.

For ease of derivation, Li et al. [20] assume households make rational consumption choices and categorize household consumption into essential consumption and non-essential consumption, denoted by $x_1$ and $x_2$, respectively. When the quantity of $x_1$ has not reached a level satisfying basic needs, its marginal utility ($\partial u(x_1,x_2)/\partial x_1$) is significantly larger than that of $x_2$ ($\partial u(x_1,x_2)/\partial x_2$). Hence,

$$m=p_1{x_1}^{*}+p_2{x_2}^{*}$$

(19)

$$s.t.\mathit{max}u(x_1,x_2)$$

(20)

where $x_1$, $x_2$ and $p_1$, $p_2$ represent the consumption quantities and prices of the two types of goods, respectively, and $m$ is the budget constraint. In Equation (20), $u(x_1,x_2)$ denotes the household utility function. Defining the Lagrangian function:

$$L=u(x_1,x_2)-\lambda (p_1{x}_1+p_2{x}_2-m)$$

(21)

where $\lambda$ is the Lagrange multiplier. By the Lagrange theorem, the optimal conditions $({x_1}^{*},{x_2}^{*})$ must satisfy the following three first-order conditions:

$${\displaystyle \frac{\partial L}{\partial x_1}}={\displaystyle \frac{\partial u(x_1,x_2)}{\partial x_1}}-\lambda p_1=0$$

(22)

$${\displaystyle \frac{\partial L}{\partial x_2}}={\displaystyle \frac{\partial u(x_1,x_2)}{\partial x_2}}-\lambda p_2=0$$

(23)

$${\displaystyle \frac{\partial L}{\partial \lambda }}=p_1{x}_1+p_2{x}_2-m=0$$

(24)

Dividing Equation (22) by Equation (23) yields:

$${\displaystyle \frac{\partial u(x_1,x_2)/\partial x_1}{\partial u(x_1,x_2)/\partial x_2}}={\displaystyle \frac{p_1}{p_2}}$$

(25)

Equation (25) indicates that household utility is maximized when the marginal rate of substitution between the two types of goods equals their price ratio. Let $x_1^{\prime }$ be the quantity of essential consumption that meets the household’s basic needs. When $x_1<{x_1}^{\prime }$, the two sides of Equation (25) become:

$${\displaystyle \frac{\partial u(x_1,x_2)/\partial x_1}{\partial u(x_1,x_2)/\partial x_2}}\sim +\infty$$

(26)

$${\displaystyle \frac{p_1}{p_2}}\le k$$

(27)

where (27) indicates real number $k$ that grows without bounds. Under such circumstances, Equation (25) does not hold. To maximize utility, the household allocates its entire budget $m$ to purchasing $x_1$, $m=x_1{p}_1$. Once $x_1>{x_1}^{\prime }$, the marginal utility of essential goods $x_1$ begins to fall below a certain fixed value:

$$\partial u(x_1,x_2)/\partial x_1\le \theta$$

(28)

where $\theta$ is a fixed constant. At this point, the condition for Equation (25) to hold is satisfied. To achieve maximum utility, the household’s consumption of $x_2$ becomes:

$$x_2={\displaystyle \frac{m-x_1{p}_1}{p_2}}\ge 0$$

(29)

From this theoretical derivation, when a household’s budget constraint is tight, the household must first meet its basic survival needs, leaving little for other goods. As the budget constraint loosens, once essential consumption is no longer the sole priority, the household will increase spending on non-essential items.

Based on the above theory, we propose the following research hypotheses:

H2-1: 

Borrowing has a positive effect on health-related (medical) consumption for low-income households and in formerly impoverished areas.

H2-2: 

Borrowing has a positive effect on “enjoyment-oriented” consumption for low-income households and in formerly impoverished areas.

3. Materials and Methods

3.1. Data Sources

This paper uses data from the China Family Panel Studies (CFPS) conducted by Peking University. Based on the research needs, we select household data from 2014, 2016, 2018, and 2020 to construct four-period balanced panel data for empirical analysis. During data processing, we exclude samples that do not appear in all four waves, as well as those with severely missing or abnormal data. The final sample consists of 7516 households across four periods, yielding a total of 30,064 household-year observations.

3.2. Concept Definitions

To investigate the varying impacts of borrowing on households at different levels of relative poverty, this paper references the classification criteria of Wang and Sun [21]. Specifically, we use 40% of the median (The per capita household income indicator used here is the average of the per capita net household income for four periods in 2014, 2016, 2018, and 2020. It differs from the classification standard of per capita disposable household income used by Wang Sangui and Sun Junna. Calculated according to the above-mentioned definition, the rural relative poverty line relied on in this paper is 5719.58 yuan, and the relative poverty line for urban samples is 13,919.90 yuan.) per capita net income of urban and rural residents, respectively, to set the relative income poverty lines for urban and rural areas. Households below the respective poverty lines are defined as low-income households, whereas those above are defined as other households.

To examine the differentiated effects among households in different regions of relative poverty, we define provinces (or municipalities, autonomous regions) with a nonzero ratio of originally state-designated poor counties to total county-level administrative units as formerly poor (poverty-alleviated) areas, and provinces (or municipalities, autonomous regions) with a zero ratio as other areas. (Through calculation, the proportion of the former poverty-stricken counties designated by the state in China to the county-level administrative units is 28.84%. According to the above classification criteria, the poverty-alleviated areas include Chongqing, Shanxi Province, Hunan Province, Sichuan Province, Guizhou Province, Yunnan Province, Shaanxi Province, Gansu Province, Qinghai Province, Inner Mongolia Autonomous Region, Guangxi Zhuang Autonomous Region, Tibet Autonomous Region, Ningxia Hui Autonomous Region, Xinjiang Uygur Autonomous Region, Hebei Province, Jilin Province, Heilongjiang Province, Anhui Province, Jiangxi Province, Henan Province, Hubei Province, and Hainan Province. Other regions include Beijing, Tianjin, Shanghai, Liaoning Province, Jiangsu Province, Zhejiang Province, Fujian Province, Shandong Province, and Guangdong Province.)

3.3. Variable Selection and Settings

Household Consumption Expenditure Indicators. Total consumption: The sum of all annual household expenditures, including education/training expenditures, basic living expenditures, enjoyment-oriented expenditures, medical expenditures, and agricultural production expenditures. Education/Training Expenditures: All education-related expenses in a year, such as school-selection fees, miscellaneous fees, training fees, after-school tutoring fees, and costs of purchasing supplementary educational materials for all family members. Basic Living Expenditures: Consumption aimed at meeting essential household needs, primarily including spending on food, housing, household equipment, and daily necessities. Enjoyment-Oriented Expenditures: Consumption oriented toward enjoyment, which is closely linked to consumption upgrading. This category mainly consists of spending on travel, cultural entertainment, beauty care, and clothing for all household members within the year. Medical Expenditures: The total out-of-pocket medical expenses in a year, excluding amounts that have been or are expected to be reimbursed. Agricultural Production Expenditures: Costs arising from agricultural production activities, mainly including input expenses for planting, forestry, livestock, aquaculture, and related operations.

Household Borrowing Indicators. Formal Borrowing: Outstanding bank loans during the current period, specifically, debts owed to banks (excluding mortgage loans). Informal Borrowing: The total of outstanding private (non-bank) borrowing and borrowing from relatives/friends. Private borrowing refers to debts owed to non-bank entities or individuals (e.g., private lending institutions) for purposes other than home purchase or construction, while borrowing from relatives/friends refers to debts owed to family or friends, also excluding home purchase or construction purposes.

Household Characteristic Indicators. Average Age of Household Members; Average Years of Education in the Household; Average Health Level in the Household (1–7 scale, with higher numbers indicating better health); Household Size (number of family members); Total Household Assets; Indicator of Rural vs. Urban Household; and Total Household Income.

To reduce heteroscedasticity and address missing values, we apply “add one and take the logarithm” to the values for total consumption, each consumption subcategory, borrowing amounts, total assets, and total household income. The average age, average education, and average health level are calculated as the mean values across household members, and the health level ranges from 1 to 7, where a higher value indicates better health.

3.4. Descriptive Statistics of Variables

This section presents the descriptive statistics of household characteristics in three parts: household consumption expenditures, borrowing amounts, and basic household features. See

Table 1. Descriptive statistics of variables.

Variables	Full Sample	Low-Income Households	Other Households	Formerly Impoverished Areas	Other Areas	Unit
Total consumption	56,818.57	39,667.81	68,249.87	51,435.57	66,736.18	yuan/year
	(73,246.41)	(44,971.56)	(85,243.34)	(63,408.73)	(87,696.91)
Education/training expenditures	4427.497	3604.978	4975.721	4405.104	4468.754	yuan/year
	(9823.392)	(6941.419)	(11,311.73)	(8995.277)	(11,190.11)
Basic living expenditures	36,042.82	24,235.15	43,912.86	32,008.23	43,476.13	yuan/year
	(63,380.34)	(36,528.25)	(75,169.63)	(53,656.99)	(77,646.12)
Enjoyment-oriented expenditures	4499.716	2343.827	5936.656	3923.783	5560.809	yuan/year
	(8115.597)	(3650.009)	(9783.75)	(6364.939)	(10,528.03)
Medical expenditures	5393.746	5196.395	5525.284	5355.82	5463.621	yuan/year
	(15,923.75)	(14,704.8)	(16,685.93)	(15,325.16)	(16,971.93)
Agricultural production expenditures	6393.971	9914.38	4047.551	7061.318	5164.454	yuan/year
	(25,990.34)	(34,813.74)	(17,441.1)	(26,092.67)	(25,756.75)
Borrowing	59,786.82	18,773.65	87,122.87	67,843.23	44,943.73	yuan
	(4039,686)	(65,778.41)	(5214,583)	(5015,436)	(202,521.2)
Average age of household	44.629	45.041	44.354	43.283	47.109	Years old
	(14.609)	(14.786)	(14.484)	(14.026)	(15.320)
Average years of education	7.225	5.768	8.196	6.921	7.785	year
	(3.699)	(3.235)	(3.670)	(3.647)	(3.729)
Average health level	3.229	3.338	3.157	3.237	3.214	-
	(0.943)	(0.962)	(0.923)	(0.943)	(0.944)
Household size	3.942	4.304	3.701	4.155	3.549	persons/household
	(1.885)	(2.056)	(1.720)	(1.891)	(1.808)
Total household assets	5.95 × 107	2.32 × 107	8.37 × 107	5.19 × 107	202,521.2	yuan/household
	(1.34 × 109)	(5.21 × 108)	(1.68 × 109)	(8.62 × 108)	(1.94 × 109)
Total household income	61,209.45	31,891.04	80,750.72	52,619.58	77,035.36	yuan/household/yrs
	(102,854.1)	(32,354.32)	(126,403.8)	(80,287.92)	(133,447.1)
Rural household (Yes = 1)	0.528	0.528	0.528	0.584	0.425	-
	(0.499)	(0.499)	(0.499)	(0.493)	(0.494)
Number of household members with public service posts	0.085	0.030	0.122	0.091	0.074	persons/household
	(0.332)	(0.183)	(0.398)	(0.348)	(0.300)
Observations	30,664	12,024	18,040	19,487	10,577	-

for details. Values in parentheses are standard deviations. The means and standard deviations are calculated based on the four-year panel data (2014, 2016, 2018, 2020). Agricultural production expenditures are calculated only for sample households engaged in agricultural activities.

After determining the final sample, we divided it into five categories according to the grouping criteria for low-income households and formerly impoverished areas: full sample, low-income households, other households, formerly impoverished (poverty-alleviated) areas, and other areas. This grouping yielded the descriptive statistics for the main variables used in this paper (see Table 1).

From the perspective of household consumption expenditures, the average total consumption among the sample households is 56,818.57 yuan, with basic living expenditures at 36,042.82 yuan, significantly higher than the other types of consumption. By comparison, low-income households have an average total consumption of 39,667.81 yuan, notably lower than the 68,249.87 yuan average of other households; meanwhile, formerly impoverished areas also show lower total consumption than other areas.

Regarding borrowing, households in the sample have an average borrowing amount of 59,786.82 yuan, with low-income households borrowing less on average than other households, but households in formerly impoverished areas having significantly higher borrowing amounts than those in other regions.

In terms of basic household characteristics, the average age in the sample is 44.629 years, with an average of 7.225 years of education and a mean household size of 3.942 people. The mean annual household income is 61,209.45 yuan, and rural households account for 52.8% of the sample. By comparison, low-income households tend to have a higher average age and a significantly lower average education level than other households, while both the average age and average years of education in formerly impoverished areas are lower than in other areas. Furthermore, regions with higher relative poverty levels display a higher proportion of rural households in the sample.

4. Results

4.1. Model Specification

To investigate the effect of household borrowing on household consumption expenditures, as well as differences in consumption expenditures across various types of households, this paper divides the sample into five categories: the full sample, low-income households, other households, formerly impoverished areas, and other areas. In addition, to eliminate the influence of other inherent characteristics, we include control variables that may affect household consumption. Based on this rationale, the fixed-effects model (The p-value of the Hausman test is 0.000. Therefore, the fixed-effects model is adopted in this paper.) in this paper is specified as follows:

$$y_{it}=c+\alpha \ast fm{l}_{it}+\delta \ast k_{it}+{\epsilon }_{it}$$

(30)

where subscript $i$ denotes the household, $t$ denotes the time period, and $y_{it}$ is the household consumption expenditure of household $i$ in period $t$. This expenditure includes total consumption, education/training expenditures, basic living expenditures, enjoyment-oriented expenditures, medical expenditures, and agricultural production expenditures. $fm{l}_{it}$ is the borrowing amount of household $i$ in period $t$. $k_{it}$ is a set of control variables that may influence household consumption. ${\mathsf{\epsilon }}_{it}$ is the random disturbance term, while $\mathsf{\delta }$ and $\mathsf{\delta }$ are the parameters to be estimated.

4.2. Empirical Results Analysis

To examine the impact of household borrowing on consumption expenditures, we employ a fixed-effects model for regression estimation. The results are as follows:

The Impact of Household Borrowing on Total Consumption. (1) Full sample: Borrowing has a significant positive effect on household total consumption.

Table 2. Effect of household borrowing on total consumption.

Variable	Full Sample	Low-Income Households	Other Households	Formerly Impoverished Areas	Other Areas
Borrowing	0.026 ***	0.032 ***	0.021 ***	0.027 ***	0.024 ***
	(0.002)	(0.003)	(0.002)	(0.002)	(0.003)
Average age	−0.003 **	−0.005 **	−0.003	−0.003 **	−0.002
	(0.001)	(0.002)	(0.002)	(0.002)	(0.002)
Average years of education	0.042 ***	0.045 ***	0.039 ***	0.036 ***	0.052 ***
	(0.005)	(0.010)	(0.006)	(0.007)	(0.009)
Average health level	0.057 ***	0.072 ***	0.048 ***	0.054 ***	0.059 ***
	(0.010)	(0.017)	(0.012)	(0.012)	(0.017)
Household size	0.136 ***	0.124 ***	0.143 ***	0.131 ***	0.151 ***
	(0.009)	(0.014)	(0.011)	(0.010)	(0.017)
Total assets	0.072 ***	0.080 ***	0.068 ***	0.069 ***	0.079 ***
	(0.008)	(0.013)	(0.011)	(0.010)	(0.014)
Rural household (Yes = 1)	−0.057	−0.149	0.018	−0.032	−0.056
	(0.043)	(0.100)	(0.064)	(0.057)	(0.068)
Total household income	0.090 ***	0.061 ***	0.133 ***	0.094 ***	0.081 ***
	(0.009)	(0.011)	(0.016)	(0.011)	(0.016)
Time fixed effects	Controlled	Controlled	Controlled	Controlled	Controlled
Constant	6.338 ***	6.339 ***	6.077 ***	6.379 ***	6.200 ***
	(0.150)	(0.226)	(0.222)	(0.184)	(0.266)
Observations	30064	12024	18040	19487	10577
R-squared	0.079	0.066	0.092	0.080	0.078

Note: Robust standard errors are in parentheses. ** and *** denote significance at the 5% and 1% levels, respectively.

shows that a 1% increase in borrowing raises total household consumption by an average of 0.026%, significant at the 1% level. Han Liyan and Du Chunyue [16] pointed out that each 1% increase in urban borrowing income leads to a 0.0149% rise in consumption, lower than the conclusion here because our sample includes both urban and rural households. Compared with urban households, rural households generally have lower incomes and thus a higher borrowing elasticity of income; consequently, alleviating liquidity constraints via borrowing exerts a greater stimulative effect. In addition, the average years of education, average health level, household size, total assets, and total household income all positively influence total consumption, while the average age of household members exerts a significantly negative effect. (2) Different Income Groups: Borrowing markedly promotes total consumption among both low-income and other households; specifically, a 1% increase in borrowing boosts total consumption of low-income households by 0.032%, which is higher than the 0.021% for other households. This suggests that borrowing is more impactful for low-income households, indicating that reducing their credit constraints can more effectively stimulate consumption. (3) Different Regions: Borrowing significantly increases total consumption among households in both formerly impoverished and other areas. A 1% increase in borrowing raises total consumption in formerly impoverished areas by 0.027% and in other areas by 0.024%. This higher stimulative effect in impoverished areas likely reflects stronger liquidity constraints in those regions. Overall, borrowing exerts a significant positive effect on total consumption across households with varying degrees of relative poverty, but the effect is especially pronounced for low-income households and formerly impoverished areas. This finding further confirms the massive consumption potential in China’s low-income households and poverty-alleviated regions.

The Impact of Household Borrowing on Education/Training Expenditures. The impact of household borrowing on education/training consumption is shown in

Table 3. Effect of household borrowing on education/training expenditures.

Variable	Full Sample	Low-Income Households	Other Households	Formerly Impoverished Areas	Other Areas
Borrowing	0.031 ***	0.042 ***	0.021 ***	0.032 ***	0.027 ***
	(0.005)	(0.008)	(0.007)	(0.006)	(0.010)
Average age	0.004	0.000	0.007	0.003	0.007
	(0.004)	(0.005)	(0.005)	(0.005)	(0.007)
Avg. years of education	0.193 ***	0.146 ***	0.217 ***	0.197 ***	0.184 ***
	(0.016)	(0.026)	(0.020)	(0.019)	(0.028)
Avg. health level	−0.132 ***	−0.063	−0.172 ***	−0.119 ***	−0.162 ***
	(0.026)	(0.039)	(0.035)	(0.033)	(0.045)
Household size	0.734 ***	0.582 ***	0.854 ***	0.680 ***	0.849 ***
	(0.028)	(0.042)	(0.038)	(0.033)	(0.054)
Total household assets	0.083 ***	0.036 *	0.111 ***	0.076 ***	0.102 ***
	(0.014)	(0.022)	(0.019)	(0.018)	(0.023)
Rural household (Yes = 1)	0.148	0.467	0.234	−0.006	0.312
	(0.143)	(0.312)	(0.208)	(0.186)	(0.234)
Total household income	0.063 ***	0.085 ***	0.033	0.066 ***	0.055 **
	(0.017)	(0.021)	(0.029)	(0.022)	(0.026)
Time fixed effects	Controlled	Controlled	Controlled	Controlled	Controlled
Constant	−1.537 ***	−0.522	−2.233 ***	−1.007 **	−2.421 ***
	(0.360)	(0.528)	(0.524)	(0.449)	(0.620)
Observations	30064	12024	18040	19487	10577
R-squared	0.085	0.073	0.096	0.084	0.088

Note: Robust standard errors are in parentheses. *, **, *** indicate significance at the 10%, 5%, and 1% levels, respectively.

. (1) Full sample: Borrowing has a significant positive effect on education/training expenditures. Specifically, a 1% increase in borrowing stimulates a 0.031% rise in education/training spending, suggesting that borrowing, overall, is conducive to building household human capital. (2) Different Income Groups: Borrowing significantly promotes education/training expenditures for both low-income and other households. A 1% increase in borrowing raises education/training spending by 0.042% among low-income households, exceeding the 0.021% increase among other households. This aligns with a series of education-assistance loan policies targeting low-income households; our empirical results confirm the positive role of these policies to some extent. Additionally, compared with other households, each additional year of average education, each additional household member, and each extra yuan of total assets lead to smaller increases in low-income households’ education/training expenditures, reflecting their disadvantage in allocating resources for education investment. (3) Different Regions: Borrowing exerts a significant positive influence on education/training expenditures in both formerly impoverished and other areas. A 1% increase in borrowing raises education/training expenditures by 0.032% in formerly impoverished areas, higher than the 0.027% in other areas, further demonstrating that alleviating credit constraints in poverty-alleviated regions can effectively enhance local human capital accumulation.

Effect of Household Borrowing on basic living expenditures. The impact of household borrowing on basic living expenditures is shown in

Table 4. Effect of household borrowing on basic living expenditures.

Variable	Full Sample	Low-Income Households	Other Households	Formerly Impoverished Areas	Other Areas
Borrowing	0.015 ***	0.012 ***	0.017 ***	0.015 ***	0.016 ***
	(0.001)	(0.002)	(0.002)	(0.002)	(0.002)
Average age	−0.004 ***	−0.005 ***	−0.002 *	−0.004 ***	−0.003
	(0.001)	(0.002)	(0.001)	(0.001)	(0.002)
Avg. years of education	0.014 ***	0.019 **	0.010 *	0.015 ***	0.011
	(0.005)	(0.008)	(0.006)	(0.006)	(0.008)
Avg. health level	0.000	−0.015	0.011	−0.004	0.009
	(0.008)	(0.012)	(0.010)	(0.010)	(0.013)
Household size	0.081 ***	0.076 ***	0.083 ***	0.074 ***	0.094 ***
	(0.007)	(0.012)	(0.008)	(0.008)	(0.013)
Total household assets	0.077 ***	0.086 ***	0.073 ***	0.090 ***	0.057 ***
	(0.008)	(0.013)	(0.010)	(0.010)	(0.012)
Rural household (Yes = 1)	−0.196 ***	−0.199 **	−0.121 **	−0.207 ***	−0.154 ***
	(0.034)	(0.088)	(0.050)	(0.046)	(0.054)
Total household income	0.066 ***	0.047 ***	0.084 ***	0.072 ***	0.053 ***
	(0.007)	(0.008)	(0.011)	(0.009)	(0.010)
Time fixed effects	Controlled	Controlled	Controlled	Controlled	Controlled
Constant	7.966 ***	7.902 ***	7.888 ***	7.698 ***	8.430 ***
	(0.131)	(0.206)	(0.185)	(0.166)	(0.219)
Observations	30064	12024	18040	19487	10577
R-squared	0.080	0.069	0.085	0.085	0.071

Note: Values in parentheses are robust standard errors. *, **, *** indicate significance at the 10%, 5%, and 1% levels, respectively.

. (1) From the perspective of the entire sample, borrowing has a significant positive promoting effect on basic lifestyle consumption. According to Table 4, every 1% increase in borrowing can promote an average increase of 0.015% in basic lifestyle consumption. In addition, the average length of education, population, total assets, and total income of a household have a significant positive impact on basic living consumption, while the average age of the household and whether it is a rural household have a significant negative impact on basic living consumption expenditure. (2) From the perspective of families with different incomes, borrowing has a significant promoting effect on the basic lifestyle consumption of low-income families and other families. An increase of 1% in borrowing can significantly promote a 0.012% increase in basic lifestyle consumption for low-income households and a 0.017% increase for other households. The impact of borrowing on the basic living consumption of low-income families is smaller than that of other families, because low-income families have lower incomes, overall lower living standards, and less elasticity in basic living consumption. (3) From the perspective of different regions, borrowing has a significant positive effect on basic living consumption in poverty-stricken areas and other areas. However, the promotion effect of borrowing on basic living consumption in poverty-stricken areas is slightly smaller than that in other areas. The reason for this is similar to that of low-income families. The development level of poverty-stricken areas is low, and the average income level is not high, which also shows the characteristic of low elasticity of basic living consumption.

Effect of Household Borrowing on enjoyment-oriented expenditures. The impact of household borrowing on enjoyment-oriented expenditures is shown in

Table 5. Effect of household borrowing on enjoyment-oriented expenditures.

Variable	Full Sample	Low-Income Households	Other Households	Formerly Impoverished Areas	Other Areas
Borrowing	0.006 ***	0.008 **	0.004 *	0.006 **	0.008 **
	(0.002)	(0.004)	(0.003)	(0.003)	(0.003)
Average age	−0.012 ***	−0.011 ***	−0.012 ***	−0.012 ***	−0.010 ***
	(0.002)	(0.003)	(0.002)	(0.002)	(0.003)
Avg. years of education	0.035 ***	0.058 ***	0.023 ***	0.031 ***	0.044 ***
	(0.006)	(0.012)	(0.007)	(0.008)	(0.011)
Avg. health level	−0.021 *	−0.028	−0.022	−0.034 **	−0.002
	(0.013)	(0.022)	(0.015)	(0.016)	(0.022)
Household size	0.119 ***	0.124 ***	0.115 ***	0.126 ***	0.100 ***
	(0.011)	(0.017)	(0.014)	(0.013)	(0.019)
Total household assets	0.097 ***	0.110 ***	0.089 ***	0.100 ***	0.091 ***
	(0.008)	(0.014)	(0.010)	(0.010)	(0.014)
Rural household (Yes = 1)	−0.148 ***	−0.309 **	0.023	−0.161 **	−0.133
	(0.050)	(0.132)	(0.070)	(0.066)	(0.084)
Total household income	0.149 ***	0.122 ***	0.182 ***	0.154 ***	0.139 ***
	(0.012)	(0.016)	(0.017)	(0.015)	(0.018)
Time fixed effects	Controlled	Controlled	Controlled	Controlled	Controlled
Constant	4.474 ***	4.170 ***	4.476 ***	4.436 ***	4.515 ***
	(0.174)	(0.274)	(0.237)	(0.215)	(0.300)
Observations	30064	12024	18040	19487	10577
R-squared	0.064	0.054	0.077	0.067	0.061

Note: Values in parentheses are robust standard errors. *, **, *** denote significance at the 10%, 5%, and 1% levels, respectively.

. (1) Full sample: Borrowing positively influences enjoyment-oriented consumption: a 1% increase in borrowing raises such consumption by 0.006%, indicating that relaxing credit constraints helps households allocate more spending to leisure and enjoyment. (2) Different Income Groups: Borrowing stimulates enjoyment-oriented consumption for both low-income and other households. As shown in Table 5, each 1% increase in borrowing raises enjoyment-oriented consumption by 0.008% for low-income households, and by 0.004% for other households. Additionally, average years of education, household size, total assets, and total income all have a significant positive impact on enjoyment-oriented consumption for both groups, whereas being a rural household has a significantly negative effect only for low-income families. (3) Different Regions: Borrowing also significantly promotes enjoyment-oriented consumption in both formerly impoverished and other areas, but the effect in formerly impoverished areas is slightly weaker than in other regions, likely because enjoyment-related expenditures are more suppressed under stronger borrowing constraints in impoverished areas.

Effect of Household Borrowing on Medical Expenditures. (1) Full sample perspective: Borrowing exerts a significantly positive impact on medical expenditures. As shown in

Table 6. Effect of household borrowing on medical expenditures.

Variable	Full Sample	Low-Income Households	Other Households	Formerly Impoverished Areas	Other Areas
Borrowing	0.034 ***	0.036 ***	0.033 ***	0.033 ***	0.036 ***
	(0.004)	(0.007)	(0.006)	(0.005)	(0.008)
Average age	0.009 ***	0.005	0.011 ***	0.005	0.016 ***
	(0.003)	(0.004)	(0.004)	(0.004)	(0.005)
Avg. years of education	−0.010	−0.008	−0.014	−0.025 *	0.015
	(0.012)	(0.019)	(0.015)	(0.014)	(0.021)
Avg. health level	0.296 ***	0.274 ***	0.307 ***	0.287 ***	0.316 ***
	(0.024)	(0.038)	(0.032)	(0.030)	(0.042)
Household size	0.159 ***	0.083 ***	0.223 ***	0.137 ***	0.211 ***
	(0.020)	(0.031)	(0.027)	(0.023)	(0.041)
Total household assets	0.033 **	0.063 ***	0.018	0.020	0.056 **
	(0.013)	(0.019)	(0.018)	(0.016)	(0.022)
Rural household (Yes = 1)	0.048	−0.327	0.031	0.054	0.002
	(0.107)	(0.224)	(0.164)	(0.140)	(0.175)
Total household income	0.070 ***	0.054 **	0.081 ***	0.080 ***	0.054 **
	(0.017)	(0.021)	(0.027)	(0.021)	(0.028)
Time fixed effects	Controlled	Controlled	Controlled	Controlled	Controlled
Constant	3.495 ***	4.101 ***	3.194 ***	3.979 ***	2.511 ***
	(0.308)	(0.448)	(0.443)	(0.379)	(0.542)
Observations	30064	12024	18040	19487	10577
R-squared	0.024	0.021	0.027	0.024	0.024

Note: Robust standard errors are in parentheses. *, **, *** denote significance at the 10%, 5%, and 1% levels, respectively.

, a 1% increase in borrowing leads to a 0.034% rise in medical spending on average, indicating that borrowing relieves household liquidity constraints, thus reducing the suppression of medical consumption. (2) Different Income Groups: Household borrowing significantly boosts medical spending for both low-income and other households, but the effect is more pronounced for low-income households. During the targeted poverty alleviation phase, the government introduced a range of health poverty alleviation policies, effectively unleashing medical consumption demand among formerly poor populations, thus partially verifying the substantial success of these policies. (3) Different Regions: Borrowing significantly promotes medical consumption in both formerly impoverished (poverty-alleviated) areas and other regions. A 1% increase in borrowing enhances medical expenditures in formerly impoverished areas by 0.033%, lower than the 0.036% increase in other areas, potentially reflecting the comparatively weaker medical and healthcare service capacity in those regions.

Effect of Household Borrowing on Agricultural Production Expenditures. (1) Full sample: Household borrowing significantly promotes agricultural production expenditures. As shown in

Table 7. Effect of household borrowing on agricultural production expenditures.

Variable	Full Sample	Low-Income Households	Other Households	Formerly Impoverished Areas	Other Areas
Borrowing	0.020 ***	0.033 ***	0.013 **	0.017 ***	0.027 ***
	(0.004)	(0.007)	(0.005)	(0.005)	(0.007)
Average age	0.008 ***	0.009 *	0.007 *	0.008 **	0.010 *
	(0.003)	(0.005)	(0.004)	(0.004)	(0.005)
Avg. years of education	−0.014	0.024	−0.023	−0.026	0.015
	(0.013)	(0.023)	(0.015)	(0.016)	(0.022)
Avg. health level	−0.024	−0.073 **	0.018	−0.029	−0.037
	(0.022)	(0.036)	(0.029)	(0.028)	(0.037)
Household size	0.166 ***	0.129 ***	0.183 ***	0.172 ***	0.150 ***
	(0.022)	(0.032)	(0.030)	(0.027)	(0.040)
Total household assets	0.111 ***	0.162 ***	0.084 ***	0.114 ***	0.100 ***
	(0.012)	(0.021)	(0.015)	(0.016)	(0.020)
Rural household (Yes = 1)	1.043 ***	0.821 ***	0.883 ***	0.964 ***	1.095 ***
	(0.124)	(0.276)	(0.169)	(0.168)	(0.194)
Total household income	0.107 ***	0.134 ***	0.062 ***	0.110 ***	0.104 ***
	(0.017)	(0.024)	(0.023)	(0.021)	(0.028)
Time fixed effects	Controlled	Controlled	Controlled	Controlled	Controlled
Constant	0.803 ***	1.534 ***	0.684 *	1.483 ***	−0.392
	(0.307)	(0.490)	(0.404)	(0.389)	(0.511)
Observations	30064	12024	18040	19487	10577
R-squared	0.034	0.040	0.025	0.033	0.033

Note: Robust standard errors are in parentheses. *, **, *** denote significance at the 10%, 5%, and 1% levels, respectively.

, a 1% increase in borrowing raises agricultural spending by 0.020%, indicating that borrowing facilitates greater investment in agricultural production. Additionally, average age, household size, total assets, total income, and rural household status all positively influence agricultural expenditures. (2) Different Income Groups: Borrowing promotes agricultural production expenditures for both low-income and other households, but the effect is more pronounced for low-income families, indicating that financial assistance, especially microcredit policies under targeted poverty alleviation, effectively boosts agricultural industries among low-income households. (3) Different Regions: Borrowing significantly enhances agricultural production expenditures in both formerly impoverished areas and other regions. Each 1% increase in borrowing raises agricultural expenditures by 0.017% in formerly impoverished areas and by 0.027% in other areas. This finding highlights the relatively more developed rural financial markets and higher degrees of production intensification and industrialization in other areas, as well as differences in credit support for agricultural production across regions.

4.3. Discussion of Endogeneity and Its Treatment

Addressing Reverse Causality: Instrumental Variable (IV) method. This paper focuses on the effect of borrowing on household consumption, which may exhibit reverse causality: as household consumption increases, families might face budget deficits and thus be more likely to borrow [22]. To tackle this endogeneity, we choose the number of household members with public service posts (编制) as the IV for two-stage least squares estimation.

Employment within the formal system typically comes with stable income expectations, low unemployment risks, and higher creditworthiness. The more family members employed within the formal system, the greater the overall occupational stability and creditworthiness of the household. Under the same conditions, such households are more likely to pass the risk assessments of financial institutions and private lenders, thereby objectively increasing the success rate of borrowing. Conversely, informal employment is characterized by flexibility and instability. The more unstable the employment of household laborers, the higher the risk of delayed loan repayment, making it less likely to secure bank loans or private lending. Therefore, the number of family members employed within the formal system is one of the important factors influencing borrowing, satisfying the relevance requirement for instrumental variables. The exogeneity of the instrumental variable is primarily reflected in the following three aspects:

First, as a product of administrative system arrangements, the essential attribute of formal employment lies in job stability rather than differences in income levels. That is, an increase in the number of family members with formal employment mainly reflects a reduction in the household’s occupational risk coefficient, rather than a change in marginal propensity to consume. It does not directly influence household consumption decisions.

Second, household consumption decisions are primarily determined by income levels, lifestyle habits, and family needs, rather than the nature of employment. In other words, formal employment itself neither directly alters the household’s marginal propensity to consume nor directly induces changes in rigid expenditures such as healthcare and education.

Third, the acquisition of formal employment often precedes household consumption behavior and is subject to policy thresholds, making it difficult for ordinary households to influence consumption by adjusting the number of family members with formal employment.

Therefore, the number of family members employed within the formal system is a relatively exogenous variable, satisfying the basic requirements for the exogeneity of instrumental variables.

We further test the validity of this IV. The underidentification test shows that the Kleibergen–Paap rk LM statistic’s p-value is 0.001, indicating no underidentification problem. The weak IV test yields an F-statistic of 15.48, exceeding the 8.96 critical value under 15% bias, ruling out a weak IV. Column (1) of

Table 8. Effects of borrowing on household consumption: IV and treatment-effects models.

Variable	2SLS Regression	Treatment-Effects Model
	(1)	(2)
Borrowing	0.389 ***	2.314 ***
	(0.121)	(0.472)
Public service posts	0.340 ***	0.158 ***
	(0.100)	(0.023)
Control variables	Controlled	Controlled
Time fixed effects	Controlled	Controlled
Constant	4.405 ***	−0.671 ***
	(0.147)	(0.008)
Wald Test (Chi2)		12.07
p-value		0.001
Observations	30,064	30,064

Note: Values in parentheses are robust standard errors. *** indicates significance at the 1% level.

shows the IV estimate is 0.389, significant at the 1% level, implying that after controlling for reverse causality, a 1% increase in borrowing promotes a 0.389% rise in household consumption, consistent with the baseline regression results.

Addressing Self-Selection: Treatment-effects model. Although the IV method mitigates reverse causality, households might still “self-select” into borrowing or not borrowing. Hence, we employ a treatment-effects model. As the endogenous variable is continuous, we assign a value of 1 if a household’s borrowing is >0, otherwise 0, and then estimate the model. Column (2) of Table 8 shows that the coefficient for public service posts is significant at the 1% level in the first-stage regression, indicating a strong positive correlation with the endogenous variable. Using maximum likelihood estimation for endogeneity, the Wald test’s p-value is 0.001, rejecting the null hypothesis of no endogeneity; thus, borrowing is indeed endogenous regarding consumption. Column (2) also shows that the coefficient of borrowing is 2.314, significant at the 1% level, implying that a 1% increase in borrowing promotes a 0.389% rise in household consumption, consistent with baseline results and further validating our core conclusions.

Therefore, after addressing reverse causality and self-selection endogeneity, borrowing continues to significantly promote household consumption growth. This demonstrates that the conclusion—that borrowing enhances household consumption by alleviating liquidity constraints—remains robust.

5. Robustness Tests

To further verify the robustness of the baseline regression results, this section conducts a series of robustness checks by adopting propensity score matching, replacing the dependent variable, and subdividing the key explanatory variables (although the paper mentions robustness tests, it would still have potential data constraints and assumptions within the model).

5.1. Re-Estimation via Propensity Score Matching

Whether or not to borrow is endogenously self-selected by households. As noted, the treatment-effects model partially alleviates the self-selection-driven endogeneity issue, and its results are consistent with the baseline regressions. To strengthen the robustness of our conclusions, we employ propensity score matching (PSM) to mitigate estimation bias caused by self-selection. Column (1) of

Table 9. Effects of borrowing on household consumption: robustness check.

Variable	(1)	(2)	(3)
Borrowing	0.055 ***
	(0.008)
2-Period Lag of Borrowing		0.520 ***
		(0.191)
Formal borrowing			0.018 ***
			(0.002)
Informal borrowing			0.027 ***
			(0.002)
Control variables	Controlled	Controlled	Controlled
Time fixed effects	Controlled	Controlled	Controlled
Constant	5.980 ***	3.463 ***	6.278 ***
	(0.148)	(0.431)	(0.151)
Observations	7727	22,548	30,064

Note: Robust standard errors are in parentheses. *** indicates significance at the 1% level.

shows that after addressing endogeneity via PSM (Table 9 presents the PSM (Propensity Score Matching) estimates. The nearest-neighbor matching method with a one-to-four ratio is used. The results of other estimation methods, such as nearest-neighbor matching with a caliper, radius matching, and kernel matching, are similar and thus not reported), both the sign and significance of the core explanatory variable remain consistent with the baseline results, indicating that borrowing indeed promotes household consumption.

5.2. Replacing the Core Explanatory Variable

While the instrumental variable method helps address potential reverse causality, we further confirm our findings by following Zhang and Zhu [7], using second-order lags of household debt as the key explanatory variable. As earlier-period debts should not be influenced by current household consumption, a two-stage least squares regression reveals that the second-order lag of household debt carries a coefficient of 0.520, significant at the 1% level, reinforcing our conclusion that borrowing facilitates household consumption.

5.3. Re-Estimation by Subdividing the Explanatory Variable

To verify the robustness of our findings, we disaggregate household borrowing into formal and informal categories, then re-estimate their impacts on household consumption. Formal borrowing refers to loans from banks, credit cooperatives, and other financial institutions, whereas informal borrowing refers to loans from relatives, friends, or non-institutional sources. As shown in Column (3) of Table 9, the coefficients for formal and informal borrowing are 0.018 and 0.027, respectively, both significant at the 1% level, implying that regardless of source, borrowing effectively promotes household consumption, thus reinforcing the robustness of our core conclusion.

6. Conclusions

6.1. Research Conclusion

Based on the four-wave balanced panel data from the China Family Panel Studies (CFPS), this paper empirically examines the impact of borrowing on total household consumption and various subcategory expenditures, as well as heterogeneity across households at different relative poverty levels and regions. The findings are summarized as follows:

First, borrowing exerts a significant positive effect on total household consumption, but the magnitude of its impact varies markedly across consumption categories, ranked in descending order as follows: medical expenditures, education and training expenses, agricultural production expenditures, basic subsistence consumption, and leisure-oriented consumption. This indicates that borrowing facilitates the transition from subsistence-oriented consumption to development- and investment-oriented consumption, thereby optimizing and upgrading household consumption structures. Second, while borrowing significantly promotes total consumption and all subcategories for both low-income and non-low-income households, the degree of impact differs substantially. For low-income households, borrowing demonstrates stronger positive effects on total consumption, education and training expenses, leisure-oriented consumption, medical expenditures, and agricultural production expenditures, while exerting a smaller effect on basic subsistence consumption compared to non-low-income households. These results suggest that borrowing more effectively stimulates development- and investment-oriented consumption in low-income households, highlighting its greater efficacy in upgrading consumption structures among this demographic. Finally, borrowing significantly enhances total consumption and subcategory expenditures in both poverty-alleviated regions and other regions, albeit with notable regional disparities. The promotional effects on total consumption and education/training expenses are more pronounced in poverty-alleviated regions, whereas its impacts on basic subsistence, leisure-oriented, medical consumption, and agricultural production expenditures are relatively weaker compared to other regions. This implies that borrowing’s stronger overall consumption-boosting effect in poverty-alleviated regions primarily stems from disproportionately larger increases in education/training expenditures, underscoring greater latent domestic demand and accelerated human capital accumulation in these areas.

6.2. Policy Recommendations

Enhance Financial Market Environments to Foster Development- and Investment-Oriented Consumption: Borrowing positively affects both overall household consumption and its subcategories, with a stronger effect on developmental and investment-oriented consumption than on survival-oriented consumption, contributing to the optimization and upgrading of household consumption structures. Therefore, to alleviate relative poverty, improving financial market environments is crucial to provide robust funding support to rural families while guiding households to shift from survival-oriented to investment- and development-oriented spending, thereby enhancing the sustainable development capacity of low-income and formerly impoverished regions.

Refine Credit Supply to Address Low-Income Households’ Needs: Disparities between low-income households and others extend beyond assets and income to encompass total consumption and consumption patterns. Compared with other households, borrowing exerts a greater positive effect on total, developmental, and investment-oriented consumption among low-income families, thus increasing their living standards, boosting development prospects, and narrowing consumption gaps across varying degrees of relative poverty. In consolidating poverty alleviation outcomes and fully aligning with rural revitalization, policymakers should tailor credit supply systems to the funding characteristics of low-income families, continuously improving the precision of credit disbursements to meet their needs in education/training and industry development.

Unlock Consumption Potential in Formerly Impoverished Regions to Drive Growth: Consumption is a key driver of long-term economic growth, yet China faces significant challenges in expanding domestic demand. Borrowing has a more pronounced effect on total consumption in formerly impoverished areas compared with other regions, implying that, during the transition period, focusing on these areas’ consumption potential can elevate their consumption capacity and, in turn, advance common prosperity for all.

Strengthen Agricultural Finance Support to Bolster Rural Industrial Development: China’s agricultural capital accumulation remains relatively low, and financial services for the agricultural industry vary widely across regions. Borrowing significantly promotes agricultural production expenditures across different levels of relative poverty, thereby validating the effectiveness of China’s agricultural industry credit support policies. Nevertheless, formerly impoverished areas generally exhibit lower levels of agricultural intensification and factor productivity, making them even more reliant on credit support than non-poor areas. In the post-poverty era, policymakers should adapt financial innovation mechanisms, such as credit guarantee insurance, “order + credit + futures” systems, and mutual aid funds, according to local resource endowments, expanding support for agricultural sector development in formerly impoverished regions while managing credit risk effectively.

6.3. Discussion

The above policy recommendations not only play a crucial role in enhancing the consumption capacity of low-income groups and poverty-stricken areas during the relative poverty stage, as well as promoting the optimization and upgrading of consumption structures, but also further improve their sustainable development capabilities, making significant contributions to achieving long-term and stable poverty alleviation.

Nevertheless, it is essential to remain vigilant about the potential risks and long-term impacts of credit policies. For instance, during economic recessions or financial crises, borrowing could theoretically stimulate consumption by alleviating liquidity constraints. However, during such periods, households’ risk aversion tends to increase. If policies are not combined with efforts to restore market confidence and control structural risks, it will be difficult to achieve practical outcomes. Moreover, issues such as repayment difficulties or high interest rates resulting from borrowing could lead to severe consequences, particularly triggering chain reactions among low-income households, which may undermine the stability and sustainability of poverty alleviation. Therefore, policy design must balance short-term benefits with long-term sustainability while addressing potential risks.