Chapter

Chapter 5. Interest Rates, Targets, and Household Saving in Urban China

Author(s):
Anoop Singh, Malhar Nabar, and Papa N'Diaye
Published Date:
November 2013
Share
  • ShareShare
Show Summary Details
Author(s)
Malhar Nabar

An important aspect of domestic imbalances in China is the relatively low share of household consumption in the national economy (down to about 36 percent of GDP in 2011 from 46 percent early in the first decade of the 2000s). The flip side is the rising household saving rate. This chapter studies urban household saving rates across China’s provinces for the period 1996–2009, during which the urban household saving rate at the national level increased from 19 percent of disposable income to 30 percent. It finds that urban saving rates rose with the decline in real interest rates for this period. This association suggests that Chinese households save with a target level of saving in mind. A main policy implication is that an increase in real deposit interest rates may help lower household saving and boost domestic consumption.

Introduction

In the mid-1990s, China’s urban household saving rate stood at 19 percent of disposable income. A decade and a half later, the saving rate touched 30 percent. This increase occurred during a period in which the economy grew rapidly and expectations of improved livelihoods, higher earnings, and greater prosperity took root. Such influences might have been expected to dampen rather than trigger savings impulses.

At the same time that household saving rates rose, China’s growth model became more reliant on exports and investment. Rebalancing China’s economy back toward consumption will require, among other policies, measures that induce households to save less and spend more. This chapter examines what role financial reform, in particular, interest rate liberalization and an increase in real deposit interest rates, can play in influencing household saving rates in China.

Although a large body of research has examined the determinants of household saving in China, little attention has been directed to interest rates and their impact on saving decisions. Bank deposits are currently the primary saving vehicle available to Chinese households. Therefore, the return households earn on these bank deposits could potentially influence household saving behavior in a tangible way. This research assesses how interest rates affect household saving decisions using a panel data set that covers China’s 31 provincial-level administrative units for the period 1996–2009.

The main findings are as follows:

  • Panel estimates suggest that household saving responds strongly to a change in the real interest rate. A 1 percentage point increase in the real rate of return on bank deposits lowers the urban household saving rate by 0.6 percentage point.

  • A comparison of the relationship across subperiods shows that the association is stronger in the later period, 2003–09, relative to the earlier period, 1996–2002. The relationship is robust to the inclusion of variables that proxy for other influences on saving such as life-cycle considerations and self-insurance against income volatility.

  • The evidence also indicates that when the return on alternative investments is high (for example, when real property price growth is relatively strong), a decline in the real return on bank deposits does not have as negative an impact on household portfolios.

  • The results suggest that China’s households save to meet a multiplicity of needs—retirement consumption, purchase of durables, self-insurance against income volatility, and health shocks—and act as though they have a target level of saving in mind. An increase in financial rates of return, which would raise the return on saving, makes it easier for them to meet their saving targets. Financial reform that boosts interest rates could therefore have a strong effect on current tendencies to save. As financial development proceeds and alternative investment opportunities become more readily available, portfolio diversification may further enable households to meet their saving targets more easily, potentially contributing to lower rates of saving out of current disposable income.

The chapter is organized as follows: The next section summarizes recent trends in household saving rates in China and provides a brief overview of previous research in this area. Subsequent sections outline the target saving hypothesis, present the empirical analysis, and discuss policy implications and the importance of liberalizing interest rates within the context of an appropriately sequenced set of financial reforms.

Household Saving in China

Context

China’s economic performance has understandably attracted attention from policymakers and academic researchers attempting to explain its various dimensions. The transformation of the economy has taken place at an unprecedented pace. By successfully maintaining close to double-digit rates of GDP growth for nearly two decades, China has lifted millions out of poverty and the prospects of a middle-class lifestyle are increasingly within the reach of an ever-wider cross section of society. These accomplishments have been made without any major disruptions to growth.

In an economy that has experienced such rapid expansion and the accompanying structural transformation in modes of production and distribution, one aspect that has stood out is the decline in the household consumption share of GDP and the associated increase in saving rates. Figure 5.1 (from expenditure-based national accounts data and the flow of funds) shows the relative decline in consumption as a share of GDP over time. China’s consumption share of GDP is now well below the levels seen in countries with similar levels of per capita income (IMF, 2009).

Figure 5.1Relative Decline in Consumption

Sources: CEIC Data Company Ltd.; and IMF staff estimates.

Aziz and Cui (2006) argue that a significant portion of the decline in China’s consumption share of GDP can be accounted for by the decline in the share of GDP of household disposable income. Furthermore, household interest income as a share of GDP has also fallen. Lardy (2008) estimates that in the first quarter of 2008, household interest income as a share of GDP was about 4 percentage points lower than would have been the case had households received the same real rates of return on bank deposits as in 2002 (i.e., had nominal deposit rates been adjusted to keep pace with inflation since 2002).

Although shifts in factor income shares might explain some portion of the decline in the consumption share of GDP, as Figure 5.1 shows, consumption has also been falling as a fraction of disposable income. Guo and N’Diaye (2010) argue that China’s relatively low private consumption can be accounted for by conditioning factors such as the low levels of services sector development and financial development and low real interest rates. They also point out that consumption can be linked to these fundamentals without invoking China-specific cultural attributes as explanatory variables.

The complement of a declining consumption share is an increase in the household saving rate. Figure 5.2 shows the national average urban and rural saving rates computed from household survey data compiled annually by China’s National Bureau of Statistics. The saving rates are calculated using per capita income and consumption. For the urban series, the measure of income used is disposable income per capita; for the rural series the measure used is net income per capita. As seen in Figure 5.2, survey data indicate that urban saving rates have increased since the mid-1990s, with an acceleration in 2003–09. Rural saving rates have not shown as consistent a pattern: they rose in the late 1990s, before reaching a plateau and then declining slightly. Since 2006, they appear to be on the rise again. This chapter focuses on the urban saving rate, leaving the rural saving rate for future analysis.

Figure 5.2Household Saving Rates

Sources: CEIC Data Company Ltd.; and IMF staff estimates.

Previous Explanations

Household saving behavior in China has been the focus of considerable empirical research in recent years. Previous work in this area has mainly analyzed survey data drawn from a sample of Chinese cities or specific rural areas (for example, Banerjee, Meng, and Qian, 2010; Chamon and Prasad, 2010; Giles and Yoo, 2007). A smaller set of literature has looked at developments at the national level (Modigliani and Cao, 2004; Kuijs, 2006) or across provinces (Kraay, 2000; Horioka and Wan, 2007; Wei and Zhang, 2011). The rise in saving rates has been variously attributed to demographic changes and life-cycle considerations, precautionary saving for self-insurance purposes, household formation, and spreading home ownership.

Demographic changes and life-cycle considerations have been linked to the increase in household saving rates in both macro-level analyses (Modigliani and Cao, 2004) and micro-level research (Banerjee, Meng, and Qian, 2010). Modigliani and Cao (2004) argue that the rise in the saving rate is in line with the predictions of a standard life-cycle model in which savings follow a hump-shaped profile (negative saving in early years of life gives way to positive saving and wealth accumulation in the working years, followed by dissaving during retirement). They point out that China’s high growth in recent decades has meant that the saving of the young has more than offset the dissaving of the elderly, leading to a net increase in the saving rate. Banerjee, Meng, and Qian (2010) use a sample of urban households covering 19 cities across nine provinces and find further evidence in favor of the life-cycle model. In the absence of pension systems and developed financial markets, children are viewed as a substitute means of old age support. Using the variation in family size induced by the One Child Policy in the 1970s, Banerjee, Meng, and Qian (2010) find that family planning policies have contributed to an increase in household saving and wealth accumulation for retirement, with a large portion of the increase coming from households that have a daughter.

A second category of explanation ties the increase in household saving to a higher private burden of health and education expenditures since the 1990s. This explanation has been advanced for urban households (Chamon and Prasad, 2010) as well as more broadly for the country as a whole (Blanchard and Giavazzi, 2006). Chamon and Prasad (2010) study a sample of urban households drawn from 10 provinces for 1992–2005, a period in which the iron rice bowl welfare system was dismantled.1 They conclude that the decline in provision of these important services by state-owned enterprises (SOEs) contributed to an increase in saving across the age distribution of the population. Young households increased their saving to finance education expenditures for their children, while older households saved more to provide a buffer against uncertain health expenditures. Blanchard and Giavazzi (2006) document the rising share of out-of-pocket expenses in total health spending for the 1990s and the particularly heavy load borne by rural households in this regard. They argue that the self-insurance motive is a key factor behind the rising saving rate during this period.2

A related hypothesis that connects SOE reform, self-insurance, and the rising saving rate is put forward by Chamon, Liu, and Prasad (2010). The authors use a sample of urban households from nine provinces for the period 1989–2006 to show that higher income uncertainty in the aftermath of SOE reform and restructuring has induced young households to save more over time, while pension reforms and lower retirement benefits have led older households to increase their saving as well. These two developments at opposite ends of the age distribution have contributed to a more pronounced U-shape in the cross-sectional age-saving profile in recent years relative to the 1990s.

Another important development in the 1990s was the privatization of the urban housing stock. Chamon and Prasad (2010) find evidence that the increase in saving rates, particularly among young households, is caused in part by home-ownership objectives. Home ownership in urban areas has grown rapidly since the initial transfer of the housing stock from SOEs to urban residents, often at below-market prices (Naughton, 2007, p. 123). However, the spread of mortgage finance has not kept pace, and home purchases to a large extent are still almost entirely financed out of savings.

The home-ownership and household-formation motives also play a central role in the hypothesis advanced by Wei and Zhang (2011). Their study looks at a province-level panel covering the period 1978–2006 and a sample drawn from the 2002 Chinese Household Income Project. Wei and Zhang’s hypothesis is that home ownership is an effective way of signaling wealth and prosperity in the marriage market, thereby raising the probability of securing a good partner in the competitive matching process. The desire to acquire a home as a signaling device induces higher saving before the purchase of the home. Wei and Zhang find that households with a son tend to save relatively more in provinces with more skewed sex ratios.

The diverse conclusions from the several studies undertaken pose a challenge for distilling the policy implications of measures that might help curb savings and lift household consumption. Across these studies, the samples differ (urban versus rural, the particular subset of cities studied, and the periods covered), making it difficult to compare results effectively and draw conclusions on macro developments or policy measures that can be implemented at a national level. Conclusions themselves vary, sometimes in diametrically opposite ways: Banerjee, Meng, and Qian (2010) find that households with a daughter increase their saving relatively more and interpret this as a sign that children are substitutes for retirement saving; Wei and Zhang (2011) point out that households with sons have seen a relatively larger increase in saving and use this to support their argument on home ownership acting as a signal in the competitive marriage market. A further challenge for drawing clear policy messages arises when increasing household saving is reviewed within the environment of the growing external imbalance since 2003.

Target Saving Hypothesis

To link the increase in saving rates with widening external imbalances and identify possible policy measures that can contribute to the rebalancing effort, factors need to be found that help account for the accelerated increase in saving rates since 2003. One possible factor is the decline in real interest rates that has occurred alongside the rise in saving rates, the shift in the structure of demand, and the increasing reliance on exports and investment during this period (Lardy, 2008). This section provides some motivation and conceptual background for the hypothesis that the decline in real interest rates, particularly in 2003–09, has tangibly affected urban household saving decisions.

Figure 5.3 plots the time series for the national average urban saving rate against the real interest rate on bank deposits. Over the period 2003–09, as the real interest rate on bank deposits declined, the urban saving rate rose steadily.

Figure 5.3Urban Household Saving Rates and Real Interest Rates, 2003–09

Sources: CEIC Data Company Ltd.; and IMF staff estimates.

Looking more closely at the variation across provinces, a similar negative correlation emerges between province-level urban saving rates and real interest rates on bank deposits. Figure 5.4 plots the annual province-level deviation of urban saving rates from the province average for the post-2003 period against the similar deviation of real interest rates from their province-specific period average. The scatter plot suggests that across China’s provinces, as real interest rates fall relative to their mean, urban saving rates move in the opposite direction and tend to increase relative to the average.

Figure 5.4Deviation from Provincial Average, after 2003

Source: IMF staff estimates.

One concern about using the deposit interest rate as a measure of returns to saving is that it does not take into account the returns on other channels of saving, such as wealth management accounts and insurance-linked products. Banerjee, Meng, and Qian (2010) cite estimates from the 2002 China Household Income Project that urban households deposit almost all their savings in banks, with approximately 10 percent held in stocks and bonds. In recent years, as China has been reforming its financial system to allow for nonbank intermediation to emerge, wealth management and insurance-linked products have been growing in importance as alternatives to bank deposits. However, although the heavy reliance on bank deposits may have diminished, they are still the primary channel of household saving (Chamon and Prasad, 2010, p. 96). A change in the return households earn on bank deposits could, therefore, have a large influence on saving behavior.

An immediate reading of Figures 5.3 and 5.4 based on standard intertemporal models of consumption and saving is that the income effect dominates the substitution effect. In other words, as real interest rates fall, the decline in lifetime income is the overriding force that induces households to consume less (and therefore save more) today even though, with lower rates, households have the incentive to do the opposite (i.e., take advantage of the low relative price of current consumption and substitute current consumption for future consumption).

However, a focus on the tension between the income and substitution effects ignores the rapidly changing context within which Chinese households have been making their spending and saving decisions in recent years. These changes include reforms to the social safety net, changes in labor market opportunities and job security, shifting aspirations with regard to the purchase of consumer durables and housing, demographic changes associated with an aging population and skewed gender ratios, and significant changes in investment returns on household saving portfolios.

In such a rapidly changing environment, with limited use of financial services such as consumer loans, insurance, and annuities, households may instead aim to self-insure and build a buffer against income and health shocks, or target a level of wealth accumulation (relative to income) to purchase indivisible consumer durables.3 In this case, the low returns on saving may cause the actual level of wealth to fall below the targeted level of wealth and induce households to save more to make up the shortfall.4

The notion that households behave as though they have some target in mind and base their decisions on wanting to achieve their target has been explored in various contexts. In the macroeconomics literature, Carroll (1997, 2001) develops buffer-stock models of saving in which households choose consumption levels in each period with the intention of meeting a target wealth-to-income ratio that determines how much of their future consumption will rely on returns from capital relative to labor income. Households adjust their current consumption (and hence their saving) depending on where the actual wealth-to-income ratio is relative to its target level. Batini, N’Diaye, and Rebucci (2005); Laxton, N’Diaye, and Pesenti (2006); and Pesenti (2008) develop open economy dynamic stochastic general equilibrium models in which international portfolio diversification is based on a target net foreign asset position, derived from financial intermediaries’ assessment of the optimal exposure to the global economy.

Targets have also been studied in the micro literature. Camerer and others (1997) study the labor supply decisions of New York City cab drivers. Their research concludes that the cab drivers allocate their labor supply with a target income for the day in mind. When the drivers reach their target, the probability of quitting for the day rises sharply. On days when fares are easy to come by, the drivers reach their target relatively quickly and then substitute leisure for labor.

The empirical literature on household saving in China has already pointed to indirect evidence in support of the target saving hypothesis (Chamon and Prasad, 2010). When real interest rates decline, as they have in recent years, wealth is eroded, leading households to save more to replenish depleted wealth in much the same way as, on slow fare days, New York City cab drivers have to ply more routes for longer hours to meet their income targets.

The empirical analysis in the next section studies this hypothesis at length by looking at developments since 1996 and then by splitting the sample into two subperiods, 1996–2002 and 2003–09.

Empirics

Data and Descriptive Statistics

The data for this study are drawn from annual editions of the China Statistical Yearbook and Urban Household Income and Expenditure Surveys (both issued by the National Bureau of Statistics). The variables included in the analysis proxy for the main saving motives studied in previous research (summarized above). The sample covers all 31 provinces, federal municipalities, and autonomous regions for the period 1996–2009.

Table 5.1 reports descriptive statistics for the main variables of interest. The household saving rate is calculated using annual per capita provincial data on urban disposable income and urban consumption expenditure. The real interest rate is calculated using the one-year nominal interest rate on bank deposits deflated by province-level annual inflation. As seen from the table, the averages for the two series move in opposite directions across the two subintervals.

Table 5.1Descriptive Statistics for Household Saving Analysis
VariableDescriptionFull Sample (1996–2009)Subsample (1996–2002)Subsample (2003–09)
Urban household saving ratePercent of disposable income22.820.025.5
(5.2)(3.9)(5.0)
Real interest ratePercent, per year1.42.9−0.1
(2.5)(2.1)(1.8)
Growth rate of disposable incomePercent10.610.610.6
(4.0)(5.2)(2.1)
Income volatilityPercent (coefficient of variation, disposable income growth)13.412.714.1
(5.0)(6.4)(2.9)
Elderly dependency ratioPopulation 65+/working-age population, percent11.210.312.2
(2.5)(2.3)(2.4)
Child dependency ratioPopulation below 15/working-age population, percent30.934.826.5
(9.0)(8.3)(7.6)
SOE shareFraction of workers employed by SOEs68.273.462.2
(13.5)(10.2)(14.4)
Urban property incomeProperty income share of total disposable income1.81.91.7
(1.2)(1.3)(1.0)
Residential property pricesRMB/square meter2,407.01,666.42,860.9
(1,706.0)(892.6)(1,916.0)
Real property price growth (residential)Year-over-year percent change7.94.89.4
(10.9)(9.7)(11.2)
Sex ratioNumber of males per 100 females103.8104.7103.1
(3.5)(3.3)(3.39)
Household sizeAverage number of residents in a household3.53.73.3
(0.5)(0.5)(0.43)
Administrative unitsProvinces, federal municipalities, autonomous regions313131
ObservationsProvince-year units429212217
Sources: China Statistical Yearbook, Urban Household Income and Expenditure Survey (National Bureau of Statistics).Note: Number of observations reported in each sample is for urban household saving rates; number of observations for other variables vary depending on data availability. Standard deviation in parentheses. SOE = State-owned enterprise.
Sources: China Statistical Yearbook, Urban Household Income and Expenditure Survey (National Bureau of Statistics).Note: Number of observations reported in each sample is for urban household saving rates; number of observations for other variables vary depending on data availability. Standard deviation in parentheses. SOE = State-owned enterprise.

At the same time, several notable changes have occurred during this period that are likely to have had some bearing on saving decisions. Income volatility has increased, leaving workers exposed to more variable earnings. The age distribution of the population has shifted to the right, with a slight increase in the elderly dependency ratio and an appreciable decline in the child dependency ratio—both demographic shifts suggest that life-cycle saving influences could also have had an impact on household saving rates during this time. Signs of SOE reform and restructuring are evident in the sharp decline in the fraction of workers employed by SOEs, a change that has left workers with a less comprehensive safety net than they had been accustomed to. Finally, real property price growth accelerated during the second half of the sample, and average household size declined. Both developments point to rising demand for residential property and increases in home ownership, changes that could have accelerated the increase in saving rates to finance purchases of housing and other durable assets. As discussed in more detail below, these changes could have also simultaneously affected the real interest rate even as they had an impact on saving decisions, which would be an argument for their inclusion in the empirical analysis to avoid problems of omitted variable bias.

The remainder of this section provides estimates of the influence of these various determinants and assesses what impact the decline in real interest rates has had on urban household saving rates. Baseline estimates from three different specifications are presented first, before robustness tests are conducted using the variables listed in Table 5.1.

Baseline Estimates

The main specification is

in which the saving rate is regressed on a vector of covariates X and the real interest rate. In the baseline version, the covariates include the growth rate of income and volatility. Additional controls are introduced in the robustness tests.

The main coefficient of interest is β2. A statistically significant finding of β2 < 0 would indicate that the decline in real interest rates since the 1990s could have potentially contributed to the increase in the saving rate.

Ordinary Least Squares (OLS) and Fixed Effects

Table 5.2 reports results from the OLS and fixed effects regressions. The first column confirms a statistically significant negative correlation between real interest rates and urban saving rates across the sample. The magnitude is economically significant as well. A decline in real interest rates from the average for the first half of the sample (2.9 percent) to the average for the second half of the sample (-0.1 percent) is associated with an increase in the saving rate of approximately 1½ percentage points.

Table 5.2Baseline Regression Estimates
OLSFixed Effects
(1)(2)(3)(4)
Real interest rate−0.483***−0.550***
(0.104)(0.0826)
Nominal interest rate−1.775***
(0.235)
CPI inflation0.0388
(0.0701)
Growth rate of income−0.597**−0.779***0.327−1.250***
(0.244)(0.196)(0.236)(0.180)
Volatility of income0.547***0.583***0.382**1.011***
(0.184)(0.160)(0.141)(0.145)
SAMPLE YEARS1996–20091996–20091996–20091996–2009
Number of provinces, regions, municipalities31313131
Observations428428428428
R-squared0.140.240.370.17
Source: IMF staff estimation.Note: Dependent variable: Urban saving rate. Robust standard error in parentheses. CPI = consumer price index; OLS = ordinary least squares.

Significant at the 1 percent level.

Significant at the 5 percent level.

Significant at the 10 percent level.

Source: IMF staff estimation.Note: Dependent variable: Urban saving rate. Robust standard error in parentheses. CPI = consumer price index; OLS = ordinary least squares.

Significant at the 1 percent level.

Significant at the 5 percent level.

Significant at the 10 percent level.

The two other covariates included in the regression are standard proxies for life-cycle saving influences and precautionary saving motives (Kraay, 2000; Horioka and Wan, 2007).5 If the realized growth rate of income is regarded as a predictor of future income growth, an increase in the growth rate of income should be associated with a decrease in current saving rates because households will defer retirement saving to a future stage in the life cycle (Chamon, Liu, and Prasad, 2010). This prediction is borne out by the negative coefficient on the growth rate of mean disposable income in the first column. The OLS regression also indicates that higher income volatility is associated with an increase in the saving rate, suggesting that households hold back on consumption and prefer to build up their buffers against future shocks when income uncertainty increases.

The regression in the second column controls for time-invariant, province-level fixed effects. The fixed effects primarily capture locational differences (coastal versus interior provinces) that affect the economic environment in which the saving decisions are made and real interest rates are determined. The statistical significance and signs of all three coefficients are robust to this change in specification. In particular, the real interest rate continues to be negatively associated with the urban saving rate, suggesting that a decline in the rate of return on household saving leads to a compensating increase in the saving rate.

An immediate concern is that the real interest rate is simply reflecting variations in the inflation rate rather than the financial (nominal) rate of return on bank deposits. The third and fourth columns report results from fixed effects regressions that separately introduce the nominal rate of return and the inflation rate. Although the nominal interest rate is administered centrally and is identical across provinces, its level varies over time. The effects can be distinguished from general year effects because there are multiple instances of the nominal interest rate remaining unchanged across years. In other words, the number of distinct observations for the nominal interest rate is fewer than the number of years in the sample. Of course, if an omitted variable changes at exactly the same time as the nominal interest rate, inference on the effects of the nominal interest rate will be prevented, but this seems unlikely. The results in the third and fourth columns indicate that the coefficient on the nominal interest rate is negative and significant, but that a statistically significant relationship between inflation and the saving rate is absent. This suggests that it is indeed the financial rate of return that seems to matter for the effect of the real interest rate reported in the first and second columns.

Panel Generalized Method of Moments (GMM) Estimation

To allow for the possibility that saving behavior is persistent, the regressions in Table 5.3 introduce lagged saving rates to the set of regressors. Because the introduction of the lagged dependent variable biases fixed effects estimates, the specification used is Arellano-Bond panel GMM estimation, which allows for the inclusion of the lagged saving rate along with other explanatory variables.

Table 5.3Baseline Estimates, Split Samples
(1)(2)(3)(4)(5)(6)
Lagged saving rate0.736***0.1920.720***0.638***0.1610.593***
(0.0740)(0.128)(0.108)(0.0364)(0.115)(0.0778)
Real interest rate−0.167*0.148−0.177***−0.1100.114−0.155**
(0.0860)(0.144)(0.0680)(0.0747)(0.137)(0.0702)
Growth rate of income−0.0549−0.487**0.303−0.120−0.536***0.0227
(0.110)(0.201)(0.191)(0.124)(0.205)(0.271)
Volatility of income0.03400.317**−0.004290.08640.337**0.172
(0.0945)(0.160)(0.128)(0.104)(0.162)(0.156)
SAMPLE YEARS1996–20091996–20022003–091996–20091996–20022003–09
Autonomous regions includedYESYESYESNONONO
Number of provinces, regions, municipalities313131262626
Observations394177217335153182
Arellano-Bond test of no second-order autocorrelation in first-differenced errors (p-value)0.570.880.510.510.700.80
Source: IMF staff estimations.Note: Dependent variable: Urban saving rate. Panel generalized method of moments estimation. Robust standard errors in parentheses.

Significant at the 1 percent level.

Significant at the 5 percent level.

Significant at the 10 percent level.

Source: IMF staff estimations.Note: Dependent variable: Urban saving rate. Panel generalized method of moments estimation. Robust standard errors in parentheses.

Significant at the 1 percent level.

Significant at the 5 percent level.

Significant at the 10 percent level.

The first column shows that the relationship is robust to this change in specification. However, when the sample period is split into the two subintervals (1996–2002 and 2003–09), as seen from Columns 2 and 3, the effect is significant only in the second subperiod. This suggests that the association has emerged in recent years, as the decline in real interest rates has become more entrenched.

Columns 4–6 repeat the specifications but for a smaller sample that excludes the five autonomous regions (Guangxi, Inner Mongolia, Ningxia, Xinjiang, Xizang/Tibet).6 The pattern is robust to focusing only on the smaller subset of provinces: Even after the autonomous regions are excluded, the association between real interest rates and urban saving rates is negative and significant in the second half of the sample period. The absolute value of the coefficient on the real interest rate in Column 6 is slightly smaller than the corresponding coefficient in Column 3, suggesting that saving rates are slightly less responsive to changes in the interest rate in the group of 26 provinces and federal municipalities than they are for the larger sample covering all 31 province-level administrative units.

Robustness: Additional Controls

Although the panel GMM estimation controls for the two main variables used in province-level macro analyses of the determinants of saving rates in China (growth rate of disposable income and volatility), concerns remain that the coefficient on the real interest rate may be soaking up the influence of other omitted variables that affect the saving rate. This subsection considers the main hypothesis put forward from the survey-based micro literature. All regressions are estimated for the period 2003–09, during which the negative association between real interest rates and urban saving rates appears to have emerged.

Age Distribution

As noted above, the age structure of the population has been changing rapidly in China, in part as a result of the One Child Policy instituted in the 1970s, in part due to improvements in health care, and also because of changes in desired fertility. Changes in the age structure could potentially trigger changes in spending patterns that influence consumer prices and thus the real interest rate, even as they directly influence the saving rate. This would point to potential omitted variables bias in the baseline regressions reported in Tables 5.2 and 5.3.

Table 5.4 reports results from regressions that include the elderly dependency ratio and the child dependency ratio measured by province. Columns 1 and 2 report the effects of the elderly dependency ratio for the full set of provinces and for the subset excluding the autonomous regions, respectively. The relationship between the real interest rate and the saving rate is robust to the inclusion of the elderly dependency ratio, and the magnitude itself is larger than in the baseline (Table 5.3, Columns 3 and 6). The coefficient on the elderly dependency ratio itself is robust and suggests that as the population ages, saving rates increase, possibly to finance larger health expenditures.

Table 5.4Analysis of Saving Rate, Controlling for Age Structure
(1)(2)(3)(4)
Lagged saving rate0.653***0.446***0.485***0.326***
(0.123)(0.101)(0.0732)(0.110)
Real interest rate−0.315**−0.204***0.00871−0.0459
(0.153)(0.0733)(0.115)(0.104)
Growth rate of income0.1990.07820.1290.169
(0.255)(0.299)(0.239)(0.255)
Volatility of income0.02570.178−0.01960.112
(0.138)(0.149)(0.131)(0.147)
Elderly dependency ratio0.351*0.190
(0.182)(0.181)
Child dependency ratio−0.597**−0.306**
(0.262)(0.148)
SAMPLE YEARS2003–092003–092003–092003–09
Autonomous regions includedYESNOYESNO
Number of provinces, regions, municipalities31263126
Observations186156186156
Arellano-Bond test of no second-order autocorrelation in first-differenced errors (p-value)0.490.990.580.96
Source: IMF staff estimations.Note: Dependent variable: Urban saving rate. Panel generalized method of moments estimation. Robust standard errors in parentheses.

Significant at the 1 percent level.

Significant at the 5 percent level.

Significant at the 10 percent level.

Source: IMF staff estimations.Note: Dependent variable: Urban saving rate. Panel generalized method of moments estimation. Robust standard errors in parentheses.

Significant at the 1 percent level.

Significant at the 5 percent level.

Significant at the 10 percent level.

The final two columns report estimates from regressions that include the child dependency ratio. The coefficient on the real interest rate is no longer statistically significant. Although the coefficient is not statistically significant, the negative relationship between the real interest rate and the urban saving rate emerges in the sample that excludes autonomous regions. The child dependency ratio itself is associated negatively with the saving rate, suggesting that children and financial saving are possible substitutes in the sense that as the number of children declines, the potential for intergenerational transfers later in life diminishes, leading parents to save more today. As Columns 3 and 4 demonstrate, the effect is present regardless of whether the entire set of provinces is included or only the subset excluding the autonomous regions is examined. Thus, variations in the implementation of the One Child Policy across the autonomous regions compared with the remaining provinces do not seem to affect the relationship between the child dependency ratio and the urban saving rate. It is therefore difficult to establish whether changes in desired fertility or the effects of the policy lie behind the negative relationship between the child dependency ratio and the urban saving rate.

Employment Structure and Sources of Household Income

The SOE reform and restructuring of the 1990s, involving mass layoffs and downsizing,7 introduced significant and far-reaching changes in labor market prospects, retirement income security, health insurance, and ownership of the urban housing stock. Such a shift in the implicit social safety net would be expected to trigger large changes in household spending and saving decisions. Comparing across provinces and over time, as SOE influence declines (captured by SOE share of employees), a rise in saving rates should appear to protect against the higher uncertainty to which the workers formerly employed by SOEs are now exposed.

If SOE reform and restructuring also affected production and distribution networks and introduced significant province-level variation in pricing of consumer products, the reforms would have been associated with fluctuations in the underlying structural inflation rate across provinces, and thus the real interest rate. Columns 1 and 2 of Table 5.5 include the SOE share of employment as an additional robustness check. The relationship between the real interest rate and the household saving rate is robust to including this proxy for the erosion of the implicit social safety net previously provided by SOEs. Furthermore, the coefficient on the SOE share goes in the direction predicted by theory. As the SOE share declines, household saving rates rise to strengthen household defenses against higher income volatility and a less comprehensive social insurance mechanism.

Table 5.5Analysis of Saving Rate, Controlling for SOE Share of Employment, Property Income Share
(1)(2)(3)(4)
Lagged saving rate0.630***0.307***0.579***0.432***
(0.143)(0.113)(0.0751)(0.0982)
Real interest rate−0.259**−0.177**−0.286**−0.192***
(0.128)(0.0722)(0.127)(0.0731)
Growth rate of income−0.07580.0321−0.01680.0549
(0.264)(0.250)(0.294)(0.271)
Volatility of income0.06990.09660.04020.160
(0.113)(0.130)(0.137)(0.147)
SOE share of employment−0.231***−0.208***
(0.0668)(0.0589)
Property income share1.939*0.669**
(1.054)(0.314)
SAMPLE YEARS2003–092003–092003–092003–09
Autonomous regions includedYESNOYESNO
Number of provinces, regions, municipalities31263126
Observations186156186156
Arellano-Bond test of no second-order autocorrelation in first-differenced errors (p-value)0.450.940.480.90
Source: IMF staff estimations.Note: Dependent variable: Urban saving rate. Panel generalized method of moments estimation. Robust standard errors in parentheses. SOE = State-owned enterprise.

Significant at the 1 percent level.

Significant at the 5 percent level.

Significant at the 10 percent level.

Source: IMF staff estimations.Note: Dependent variable: Urban saving rate. Panel generalized method of moments estimation. Robust standard errors in parentheses. SOE = State-owned enterprise.

Significant at the 1 percent level.

Significant at the 5 percent level.

Significant at the 10 percent level.

Along with SOE reform, in the 1990s households also received a large transfer from the state in the form of residential property stock and the potential for another form of income through rents. To the extent that provinces where households have come to rely on property income as a more important source of disposable income also experienced more rapid increases in inflation (which triggered larger declines in the real interest rate), a regression that fails to include property income share as an additional control will be subject to omitted variable bias. To address this concern, Columns 3 and 4 report results from regressions in which the property income share of household disposable income is included as an additional control. The relationship between the real interest rate and the urban household saving rate is robust to including this additional control. The sign on the coefficient itself suggests that in provinces where households depend more on property income, households tend to have higher saving rates out of disposable income, possibly because more productive saving opportunities are available (such as through the purchase of residential property that can then be rented out).

Saving to Buy Property

The privatization of the urban housing stock that started in the late 1990s introduced a noticeable change in household tenure choices with an increasing fraction of urban households choosing to buy homes (by 2005, the proportion of owner-occupied urban housing units had exceeded 80 percent, up from 48 percent in 1999; Naughton, 2007, p. 123). Housing finance options have remained limited, however, because bank lending continues to be primarily focused on enterprises (only 13 percent of the outstanding loan portfolio as of end-2010 was allocated to residential mortgages). As a result, one possible explanation for the rapid increase in household saving is that it is intended for housing purchase. In an environment in which home ownership is increasingly widespread, saving decisions would be expected to conform over time as households plan for an eventual home purchase.

To the extent that rapid house price increases are associated with other relatively more rapid general consumer price growth, residential property prices could have a negative association with real interest rates even as they are positively associated with the urban saving rate. This raises yet another concern of omitted variable bias in the regressions reported in Table 5.3.

The regressions reported in Table 5.6 use two proxies to control for the influence of the property motive: the level of the residential property price and the growth rate of real property prices. Columns 1 and 2 indicate that the association between the real interest rate and urban saving rates is robust to including the level of the residential property price. The pattern is repeated in Columns 3 and 4, in which the growth rate of real property prices is introduced. The property motive itself is significant in the first two columns, indicating that urban saving rates tend to be higher in provinces with high levels of property prices.

Table 5.6Analysis of Saving Rate, Controlling for Property Prices
(1)(2)(3)(4)
Lagged saving rate0.617***0.385***0.712***0.733***
(0.0598)(0.0789)(0.0400)(0.0789)
Real interest rate−0.254***−0.191***−0.177**−0.144*
(0.0788)(0.0628)(0.0799)(0.0733)
Growth rate of income−0.233−0.1350.2940.154
(0.301)(0.252)(0.201)(0.278)
Volatility of income0.09520.1110.0006600.147
(0.125)(0.136)(0.129)(0.168)
Property price level3.283***3.335***
(0.807)(0.669)
Real property price growth0.000693−0.00841
(0.0113)(0.0127)
SAMPLE YEARS2003–092003–092003–092003–09
Autonomous regions includedYESNOYESNO
Number of provinces, regions, municipalities31263126
Observations217182217182
Arellano-Bond test of no second-order autocorrelation in first-differenced errors (p-value)0.500.890.140.11
Source: IMF staff estimation.Note: Dependent variable: Urban saving rate. Panel generalized method of moment estimation. Robust standard error in parentheses.

Significant at the 1 percent level.

Significant at the 5 percent level.

Significant at the 10 percent level.

Source: IMF staff estimation.Note: Dependent variable: Urban saving rate. Panel generalized method of moment estimation. Robust standard error in parentheses.

Significant at the 1 percent level.

Significant at the 5 percent level.

Significant at the 10 percent level.

Additional Household Attributes

A final set of robustness checks introduces the gender ratio and average household size as additional controls. The gender composition of the population may influence marriage prospects, household formation, and the impulse to save to purchase a home (Wei and Zhang, 2011). To the extent that marriages are largely formed within the province (i.e., few matches take place across provinces), provincial variation in the gender composition (number of males per 100 females) can be used as a proxy for the household-formation motive for saving.

Average household size reflects the influence that the number of coresidents may have on the saving rate of the average household. As household size increases, at least two possible effects on saving emerge. First, a larger household could imply more coresident adults (elderly, for example) and so greater possibilities to tap into resources accumulated through their past saving.

Second, variation in household size could also arise because of children in the home. Those households with a child may be relatively less concerned about saving for retirement purposes than those households without a child because the possibilities for intergenerational transfers are greater in the first set.

As the first two columns in Table 5.7 indicate, introducing the variation of gender composition as an additional control does not affect the significance or the sign of the coefficient on the real interest rate. Gender composition itself does not have a significant effect on the urban saving rate. This may reflect, in part, two competing effects that cancel each other. A larger number of males relative to females may lead to lower saving for retirement because there may be an expectation of a greater degree of intergenerational transfer from the male child to the elderly parents relative to the transfers that potentially take place from a female child. But an increase in the sex ratio may increase the competition for marriage matches and therefore strengthen motives to save to purchase a home.

Table 5.7Analysis of Saving Rate, Including Gender Ratios and Household Size
(1)(2)(3)(4)
Lagged saving rate0.718***0.593***0.752***0.743***
(0.0445)(0.0774)(0.0453)(0.0651)
Real interest rate−0.191**−0.159**−0.178**−0.193***
(0.0836)(0.0704)(0.0870)(0.0733)
Growth rate of income0.352**0.03160.1220.399
(0.175)(0.269)(0.244)(0.272)
Volatility of income−0.03660.166−0.115−0.111
(0.128)(0.152)(0.160)(0.182)
Sex ratio (number of males per 100 females)0.1040.0225
(0.133)(0.0721)
Household size−4.594**−3.635***
(2.060)(1.401)
SAMPLE YEARS2003–092003–092003–092003–09
Autonomous regions includedYESNOYESNO
Number of provinces, regions, municipalities31263126
Observations217182155130
Arellano-Bond test of no second-order autocorrelation in first-differenced errors (p-value)0.140.800.780.39
Source: IMF staff estimation.Note: Dependent variable: Urban saving rate. Panel generalized method of moment estimation. Robust standard error in parentheses.

Significant at the 1 percent level.

Significant at the 5 percent level.

Significant at the 10 percent level.

Source: IMF staff estimation.Note: Dependent variable: Urban saving rate. Panel generalized method of moment estimation. Robust standard error in parentheses.

Significant at the 1 percent level.

Significant at the 5 percent level.

Significant at the 10 percent level.

Columns 3 and 4 introduce average household size as an additional control. The main coefficient of interest on the real interest variable continues to be negative and significant with the introduction of this new control. Average household size itself is negatively associated with the urban saving rate: As household size declines across provinces and over time, the urban saving rate rises. The negative coefficient is consistent with the two possible explanations outlined above.8

As this section has demonstrated, the baseline estimates are robust to the inclusion of several additional controls, each of which could affect household saving and the real interest rate, as argued above. Nevertheless, these results should be interpreted with some caveats. First, other factors could also influence the province-level saving rate, such as changes in the household income and wealth distribution during the 1990s and 2000s. A more disaggregated, household-level data analysis would be required to capture the variation in saving rates induced by shifts in the income and wealth distribution. Second, differences in local government spending on social insurance versus physical infrastructure might introduce variations in provincial social safety nets. Since the global financial crisis, for example, numerous measures have been taken to enhance social safety nets. Access to primary health care has been improved through the construction of new health facilities. A new government health insurance program has been launched. The existing government pension scheme is being expanded to cover urban unemployed workers across the country, and the newly introduced rural pension scheme now covers 60 percent of all counties (see IMF, 2011, for more details). The SOE share variable used in the analysis here might not be adequately capturing these differences across provinces that could influence saving behavior. A more detailed breakdown of government expenditures would need to be included in the analysis to reflect the influence of social safety nets on household saving behavior.

The next section discusses possible policy options for increasing the return on saving within the broader agenda of financial reform in China.

Interest Rate Liberalization and Financial Reform

China has made significant progress toward strengthening its financial system. Bank balance sheets have been made more robust by shifting nonperforming loans to asset management companies, and banks have subsequently been recapitalized. The major banks have been listed on international stock exchanges, and the system has become more commercially oriented. Risk management standards have been enhanced and supervisory and regulatory capacity has been significantly widened to cover banking, insurance, and capital markets. Financial market development has proceeded at a steady pace on multiple fronts with the establishment of exchange-traded and interbank bond markets (IMF, 2011).

Nevertheless, as Table 5.1 and Figure 5.3 indicate, real interest rates have declined steadily since 2003. Banks have so far not had to compete aggressively for household deposits because the ceiling on deposit rates limits how far they can go to lure depositors away from competitors; furthermore, the combination of limited alternative saving options through nonbank channels and capital controls reduces contestability from outside the banking system. But, as attested to by the growing popularity of wealth management products, households would move deposits out of the banks to earn higher returns if given more choice regarding saving options. Greater flexibility and choice for household portfolio allocation would exert more competitive pressure on banks, raising interest rates and the return on household saving in the process.9

As outlined in previous IMF work in this area, interest rate liberalization would need to take place within the context of a financial reform agenda that is appropriately sequenced while also maintaining sufficient flexibility to adapt to evolving circumstances (IMF, 2011, and Chapter 13 of this volume). Before interest rates can be deregulated to enable them to guide the allocation of saving and credit in the system, the significant existing liquidity overhang in the financial system would need to be drained. Greater exchange rate flexibility, which would minimize the injection of liquidity via the balance of payments, and more use of central bank open market operations to influence interbank rates are essential prerequisites.

To minimize the risks of a disorderly liberalization, loosening of lending standards, and a surge in credit, regulatory and supervisory capacity would need to be strengthened to close gaps in coverage. With surplus liquidity drained from the system and an enhanced regulatory framework in place, the authorities could begin liberalizing interest rates by lifting the deposit rate ceiling.

As interest rate liberalization proceeds and a strong regulatory framework is developed, a wider range of saving options could be made more easily accessible to households by encouraging greater participation of mutual funds, money market funds, and insurance-linked investment products in the capital market. This broader financial market development will, in time, cover a range of securities from equity to corporate fixed-income instruments to government bonds across the maturity spectrum.

Finally, the range of instruments and investment options can be expanded to include cross-border claims by removing capital controls and allowing for unrestricted two-way flows of capital.10

Eventually, with this greater flexibility and choice in saving vehicles, households will be able to achieve more efficient allocations of saving and extract higher returns from their portfolios. As demonstrated in the regression analysis, an increase in financial rates of return could potentially contribute to a decline in the share of household disposable income that is saved. Combined with the impact of higher interest rates on corporate saving and investment behavior (see Chapter 4 of this volume), the cumulative effect of financial reform on the saving-investment imbalance would be large. An appropriately sequenced set of financial reforms, therefore, could have a substantial impact on rebalancing China’s economy.

Conclusion

This chapter argues that the increase in urban saving rates in China is linked, in part, to the decline in real interest rates, particularly after 2003. Although the result can be interpreted as a case of the income effect outweighing the substitution effect, this would be a narrow reading of the context in which Chinese households have been making their saving decisions in recent years. In an environment in which the underlying fundamentals are changing as rapidly as has been the case in China since the late 1990s, a more relevant framework for analyzing saving responses is a target saving model. Chinese households appear to save to meet multiple needs, including self-insurance, retirement consumption, and purchase of durable assets, and act as though they have a saving target in mind. Changes in real interest rates affect the rate of return on household saving and the degree of difficulty in meeting that saving target. Since 2003, the decline in real interest rates seems to have lowered actual saving accumulation below the target. Households appear to have responded by boosting their rates of saving out of current disposable income to raise actual financial wealth closer to the target.

The relationship between real interest rates and the urban saving rate was found to be robust to the inclusion of other variables that proxy for commonly invoked savings impulses such as life-cycle considerations or the precautionary motive. The association was also found to be robust to changes in the sample of provinces, moving from the full set of China’s 31 provincial-level administrative units to a smaller subset that excludes the five autonomous regions.

The results of this study suggest that financial reform, a sustained increase in interest rates on bank deposits, and wider access to alternative investment opportunities would make it easier for households to meet their target saving accumulation. To the extent this lowers saving out of current disposable income and increases private consumption, financial reform could potentially contribute significantly toward rebalancing China’s economy.

References

    AzizJ. and L.Cui2007Explaining China’s Low Consumption: The Neglected Role of Household Income,IMF Working Paper 07/181 (Washington: International Monetary Fund).

    BanerjeeA.XinMeng and N.Qian2010The Life Cycle Model and Household Savings: Micro Evidence from Urban China” (unpublished; New Haven, Connecticut: Yale University).

    BatiniN. P.N’Diaye and A.Rebucci2005The Domestic and Global Impact of Japan’s Policies for Growth,IMF Working Paper 05/209 (Washington: International Monetary Fund).

    BlanchardO. and F.Giavazzi2006Rebalancing Growth in China: A Three-Handed Approach,CEPR Discussion Paper No. 5403 (London: Center for Economic Policy Research).

    CaiFangJ.Giles and XinMeng2006How Well Do Children Insure Parents against Low Retirement Income? An Analysis Using Survey Data from Urban China,Journal of Public Economics Vol. 90 No. 12 pp. 222955.

    CamererC. L.BabcockG.Loewenstein and R.Thaler1997Labor Supply of New York City Cabdrivers: One Day at a Time,Quarterly Journal of Economics Vol. 112 No. 2 pp. 40741.

    CarrollC.1997Buffer-Stock Saving and the Life Cycle/Permanent Income Hypothesis,Quarterly Journal of Economics Vol. 112 No. 1 pp. 155.

    CarrollC.2001A Theory of the Consumption Function, With and Without Liquidity Constraints,Journal of Economic Perspectives Vol. 15 No. 3 pp. 2345.

    ChamonM.KaiLiu and E.Prasad2010Income Uncertainty and Household Savings in China,IMF Working Paper 10/289 (Washington: International Monetary Fund).

    ChamonM. and E.Prasad2010Why Are Saving Rates of Urban Households in China Rising?American Economic Journal: Macroeconomics Vol. 2 No. 1 pp. 93130.

    FeyzioğluT.N.Porter and E.Takáts2009Interest Rate Liberalization in China,IMF Working Paper 09/171 (Washington: International Monetary Fund).

    GilesJ. and RenMu2007Elderly Parent Health and the Migration Decision of Adult Children: Evidence from Rural China,Demography Vol. 44 No. 2 pp. 26588.

    GilesJ. and K.Yoo2007Precautionary Behavior, Migrant Networks, and Household Consumption Decisions: An Empirical Analysis Using Household Panel Data from Rural China,Review of Economics and Statistics Vol. 89 No. 3 pp. 53451.

    GuBaochangWangFengGuoZhigang and ZhangErli2007China’s Local and National Fertility Policies at the End of the Twentieth Century,Population and Development Review Vol. 33 No. 1 pp. 12947.

    GuoK. and P.N’Diaye2010Determinants of China’s Private Consumption: An International Perspective,IMF Working Paper 10/93 (Washington: International Monetary Fund).

    HoriokaC. and JunminWan2007The Determinants of Household Saving in China: A Dynamic Panel Analysis of Provincial Data,Journal of Money Credit and Banking Vol. 39 No. 8 pp. 207796.

    International Monetary Fund (IMF)2009Regional Economic Outlook—Global Crisis: The Asian Context (Washington). http://www.imf.org/external/pubs/ft/reo/2009/APD/eng/areo0509.pdf.

    International Monetary Fund (IMF)2011People’s Republic of China: Staff Report for the 2011 Article IV Consultation” (Washington). www.imf.org/external/pubs/ft/scr/2011/cr11192.pdf.

    KraayA.2000Household Saving in China,World Bank Economic Review Vol. 14 No. 3 pp. 54570.

    KuijsL.2006How Will China’s Saving-Investment Balance Evolve?World Bank Policy Research Working Paper No. 3958 (Washington: World Bank).

    LardyN.2008Financial Repression in China,Policy Brief 08-8 (Washington: Peterson Institute for International Economics).

    LardyN. and P.Douglass2011Capital Account Liberalization and the Role of the Renminbi,PIIE Working Paper 11-6 (Washington: Peterson Institute for International Economics).

    LaxtonD. P.N’Diaye and P.Pesenti2006Deflationary Shocks and Monetary Rules: An Open-Economy Scenario Analysis,NBER Working Paper No. 12703 (Cambridge, Massachusetts: National Bureau of Economic Research).

    ModiglianiF. and ShiLarry Cao2004The Chinese Saving Puzzle and the Life-Cycle Hypothesis,Journal of Economic Literature Vol. 42 No. 1 pp. 14570.

    NabarM.2011Targets, Interest Rates and Household Saving in Urban China,IMF Working Paper 11/223 (Washington: International Monetary Fund).

    NaughtonB.2007The Chinese Economy: Transitions and Growth (Cambridge, Massachusetts: MIT Press).

    PesentiP.2008The Global Economy Model (GEM): Theoretical Framework,IMF Staff Papers Vol. 55 No. 2 pp. 24384.

    WeiShang-Jin and XiaboZhang2011The Competitive Saving Motive: Evidence from Rising Sex Ratios and Savings Rates in China,Journal of Political Economy Vol. 119 No. 3 pp. 51164.

See Naughton (2007), Chapters 8 and 13, for an overview of this reform including the restructuring of SOEs. Starting in 1993, more than 30 million SOE workers (about 40 percent of the SOE workforce) were laid off (p. 301). The total number of industrial SOEs declined from 120,000 in the mid-1990s to 31,750 in 2004 (Naughton, 2007, p. 313).

Self-insurance and intrafamily in-kind transfers have also had a bearing on other outcomes, such as participation in the migrant labor market (Giles and Mu, 2007). In turn, micro-level evidence from rural surveys indicates the role of migrant labor networks in determining the strength of the precautionary saving motive: as the size of the migrant labor network expands, both poor and non-poor rural households reduce their precautionary saving when a wider pooling of risks becomes possible (Giles and Yoo, 2007). The importance of intrafamily transfers in compensating for the shortfalls of the social safety net has been documented by Cai, Giles, and Meng (2006).

This is not to suggest that households target a particular fixed level of wealth or saving. As the scale of the economy grows and income rises, the actual targeted wealth or saving will also evolve in line with other indicators.

The intuition is seen easily in a simple two-period model as outlined in Nabar (2011).

The growth rate of mean disposable income is calculated as the annual percent change of four-year moving averages and the volatility of disposable income is calculated as the coefficient of variation over four-year windows.

Variations in population policy, which affect the ethnic populations in the autonomous regions differently from the rest of the country, may introduce differences in saving behavior between the autonomous regions as a group and the rest of the country. See Gu and others (2007) on the exemptions to the One Child Policy and details of variations in implementation across the provinces.

The number of SOEs went from 120,000 in the mid-1990s to 31,750 in 2004 and nearly 40 percent of the SOE workforce was laid off during this time; see Naughton (2007) for details.

See Nabar (2011) for additional tests that suggest if financial reform creates a greater menu of higher-paying saving options for households, households could substitute and switch their portfolios toward higher-return assets. In turn, households would find it easier to meet their target accumulation, meaning they could save less and consume more out of disposable income.

Feyzioğlu, Porter, and Takáts (2009) use model-based simulations to show that interest rate liberalization in China would lift interest rates and the return on saving.

Lardy and Douglass (2011) argue that interest rate liberalization must precede capital account liberalization to lower the risk of households transferring their deposits out of the banks and thereby undermining the stability of the domestic banking system.

    Other Resources Citing This Publication