I Introduction

Since Aiyagari’s path-breaking work (Aiyagari 1994), saving has been recognized as an important instrument with which to self insure against idiosyncratic income shocks. The labor supply, as another instrument to self insure against idiosyncratic income shocks, has recently gained attention in the literature. Low (2005) shows that allowing flexible working hours in a rather standard Aiyagari-type heterogenous agent model brings the age profiles of working hours and consumption closer to the data. Flexibility of working hours allows individuals to react to income shocks by changing their hours of work, and thus reducing the cost of uncertainty. Pijoan-Mas (2006) also highlights the self-insurance channel of the labor supply and shows that the labor supply is at least as important as savings when it comes to the instruments with which to insure against income risk.

This paper studies, both empirically and quantitatively, the roles savings and the labor supply play as self-insurance channels over the life cycle when one faces not only idiosyncratic income risks, but also changes in longevity risk and pension benefits. These are important risks and uncertainties that an individual faces over the whole life cycle. Theoretically speaking, changes in pension benefits, such as whether or not a pension benefit is linked to an individual’s average (life-cycle) wage, e.g., the average index of monthly earnings (AIME) in the U.S. pension system, would have an impact on the individual’s savings and labor supply behavior. The linkage of one’s working age income to one’s retirement benefits extends the idiosyncratic income shocks faced during working years to retirement period. To hedge against an increase in income shocks during one’s retirement period, one could start to save more and/or increase one’s labor supply during one’s working life. On the other hand, changes in pension benefits via reducing the generosity of the pension system (e.g., the replacement ratio) could also affect a household’s savings and labor supply due to consumption smoothing motive. A lower replacement ratio would reduce the available pension wealth after retirement, and hence would encourage an individual to increase saving while working. The empirical literature, e.g., Attanasio and Bru-giavini (2003) and Attanasio and Rohwedder (2003), find that this is the case and the substitutability between public pension wealth and private saving is relatively high. Erosa, Fuster and Kambourov (2012) link government programs such as pensions to the labor supply, and find that social security rules account for the bulk of cross-country differences in labor supply late in the life cycle.

By the same token, changes in longevity risk could also affect the labor supply and savings. An increase in longevity would allow an individual to live for a longer time during the retirement period. In order to prepare for that, one could increase savings and/or one’s labor supply while working, so as to effectively insure against the change in longevity risk. We call this channel the life expectancy effect (see Bloom et al. 2007).

By putting together idiosyncratic income risks, longevity risk, and uncertainties about the pension system, and allowing both savings and the labor supply to insure against those risks, this paper provides a comprehensive view of individual’s responses to these risks and uncertainties over the life cycle and investigates the aggregate implications of those responses. This is the main contribution of the present paper.

Empirically and quantitatively testing the importance of savings and labor supply as self-insurance channels in response to changes in longevity risk and pension benefits, however, is quite challenging, since not many countries have experienced significant changes in both longevity risk and pension benefits. In this paper, we choose China as a case study because China has undergone a rapid aging and a radical pension reform simultaneously for the period 1995–2009 (more details in Appendix A and B). Therefore, China provides an ideal “natural experiment” to test the impacts of rapid aging and pension reform on savings and the labor supply, both at the household and at the aggregate level. The first research question we ask is: how would changes in longevity risk and pension benefits, such as the ones that China has experienced, quantitatively affect households’ savings and labor supply at the aggregate level, and over the life cycle, in the long run? In addition, as shown in more detail in Section II, both the average household saving rate and weekly working hours per worker have been increasing significantly in urban China since 1997. We therefore ask a further and specific question: to what extent do the rapid demographic changes and the pension reforms mentioned above contribute to the increasing household saving rate and supply of labor in urban China after 1997?

To answer these quantitative questions, we first provide an empirical analysis showing that the Chinese pension reform indeed had an impact on the saving rate and the labor supply at the household level, using data from the Chinese Household Income Project Survey (CHIPS). We use the 2005 Chinese pension reform (for details, see Appendix VII.B) as a quasi-natural experiment to conduct a difference-in-difference (DiD) regression. We find that workers with pension benefits tend to supply less labor and save less, consistent with our theoretical prediction. Households responded to the 2005 pension reform by increasing their savings and labor supply in response to the reduction in the generosity of the pension system. In addition, our empirical analysis shows that the impact of the pension reform was more significant for the saving rate than for the labor supply.

Motivated by our empirical evidence, we then develop a large-scale (70-period) heterogeneous agent overlapping generations general equilibrium model, following the literature such as Auerbach and Kotlikoff (1987), Imrohoroglu, Imrohoroglu, and Joines (1995), Huang, Imrohoroglu and Sargent (1997), and Conesa and Krueger (1999). In the model, an individual faces stochastic income risks up to retirement and a borrowing constraint. Individuals have to make decisions on consumption, asset holding, and their labor supply. We calibrate the model to match the Chinese economy before its pension reform and rapid aging took place, by using data from micro-level Chinese household surveys. We then input the exogenous demographic change and the policy changes in the pension system into the model. We study the individual’s responses to those exogenous changes in longevity risk and pension benefits, and investigate the aggregate implications of these individual responses in the long run. In addition, we also input the annualized demographic change in China, together with a carefully modeled “once-for-all” pension reform, and solve the model along a transition path to investigate the short-run impact of the pension reform and rapid aging on our variables of interest for the period 1995–2009.

To answer the first question, we find that compared to the benchmark model without changes in longevity risk and pension benefits, a rising longevity risk alone increases the household saving rate by 3.6% and its labor supply by 3.2%. On the other hand, the pension reform as a whole package raises the saving rate by 28.7% and its labor supply by 3.0% in the long run. We also further decompose the changes in the pension system, and find that the reduction of the indexation of pension benefits (i.e., increasing the linkage with the AIME) increases the household saving rate by 18.3% and its labor supply by 1.3% in the long run. In contrast, the reduction of the replacement ratio raises the saving rate by 12.2% and the labor supply by 1.8% in the long run. The magnitude we find here from the calibrated model shows that the labor supply as a self-insurance/consumption smoothing mechanism is of the same order of magnitude as savings for insurance against longevity risk. However, our quantitative exercises show that changes in pension benefits seem to affect the saving rate much more significantly than the labor supply in the long run, which is consistent with our empirical analysis.

To highlight the importance of the labor supply as a self-insurance channel in response to uncertainties in longevity risk and pension benefits, we conduct a counter-factual experiment to shut down the flexible labor supply in our benchmark model. We find that the average household saving rate increases by 52% in the long run, compared to 30.2% in the benchmark case. This exercise shows that ignoring the labor supply would lead to a significant over-estimation of the importance of precautionary savings in insuring against uncertainties over the life cycle, not only for idiosyncratic income shocks, as emphasized by Low (2005) and Pijoan-Mas (2006), but also for uncertainties in longevity risk and pension benefits.

As to the life cycle dimension, our simulations show that young people aged 20–35 reduce their savings in responding to rapid aging and pension reform. After age 35, they start to save more. For workers aged 46–60, especially towards the end of their working years, the increase in savings is mostly in reaction to the changes in pension benefits; whereas after age 60, the increase in household savings is mostly driven by rising longevity risk. In other words, the impact of changes in the longevity risk and pension benefits on the household saving rate occurs for relatively older people (aged 45 and older). The impact increases with age. On the other hand, workers of almost all ages (except for young workers aged below 30) increase their working hours in response to changes in longevity risk and pension benefits, but the impact of pension benefit changes exceeds that of demographic changes after age 55. The magnitude of the adjustment of the labor supply is comparable to that of the saving rate until retirement age. However, after retirement, an individual is not able to use their labor supply as an instrument for self-insurance. But the adjustment via savings still exists after retirement, and increases with age. This is the fundamental reason why the impact of longevity risk and pension reform hits savings more significantly than the labor supply.

To what extent do these two self-insurance channels help to explain the rising household saving rate and labor supply in urban China after 1997? The data show that the average household saving rate increased 12.7 percentage points: from 15.5%, which is the average for 1990–1997, to 28.2% in 2009. Weekly working hours per worker as a fraction of weekly discretionary hours increased about 7.7%: from 0.379 in 1996 to 0.408 in 2010. Our transition path results show that the changes in pension benefits due to the pension reform alone contribute about 41% of the increase in the household saving rate and 41% of the increase in labor supply for the time period. While the rapid aging contributes more to the increase of the labor supply than to the saving rate: it, alone, contributes about 36% of the increase in labor supply but only 11% of the increase in household saving rate for the time period. Together, the rapid aging and the pension reforms contribute about 55% of the increase in the household saving rate from 1995 to 2009 and 64% of the increase in the labor supply from 1996 to 2010. The exercise shows that the impact of the demographic changes and pension reform on the aggregate household saving rate and the labor supply in urban China is of first order importance.

This paper is related to several strands of the literature. First of all, the empirical literature finds mixed evidence for the importance of demographic change on the aggregate saving rate, from only a modest contribution (Deaton and Paxson 2000) to one that is significantly large (Bloom, Canning and Graham 2003 and Bloom et al. 2007). We quantify the importance of the rapid aging for the household saving rate in the present paper, and find that the contribution indeed is modest. This might reflect the fact there are two opposite driving forces in action in our general equilibrium framework. On the one hand, rapid aging shifts the age structure in an economy from young (working age) towards old. Compared to retirees, working-age individuals tend to save more. Therefore the composition effect of the demographic change would reduce the household saving rate. On the other hand, the life expectancy effect as mentioned above tends to raise the household saving rate. Probably these two counteracting mechanisms offset each other in the model, to generate the modest quantitative results.

This paper also contributes to the literature on pensions and private savings, such as Attanasio and Brugiavini (2003) and Attanasio and Rohwedder (2003). We confirm their finding in the Chinese context by showing that the significant reduction of generosity of pension system could have had a big impact on the household saving rate.

However, the literature has put less emphasis on the importance of changes in demographic structure and the pension system for the aggregate supply of labor. In that sense, we are closely related to the literature pioneered by Prescott (2004) on fiscal policy and labor supply. More specifically, we are very closely related to Erosa, Fuster and Kambourov (2012). They analyze the link between the labor supply and government programs, and they find that government policies can account for the differences in the labor supply late in the life cycle (age 50+) across European countries. Instead, our focus is to quantify how labor supply can actuate the self-insurance channel against both longevity risk and changes in pension benefits through the whole life cycle. Giavazzi and McMahon (2012) use German microdata and a quasi-natural experiment to provide empirical evidence for how households use both savings and their labor supply to respond to an increase in pension policy uncertainty. We instead use the structural general equilibrium model here for the analysis. Our paper, however, complements their research, by focusing on studying the impact of changes in the pension benefits formula, in terms of both the replacement ratio and the linkage to the AIME, on household savings and labor supply. Similar to their findings, we find that the impact seems to hit savings more.

Although our paper does not aim to solve the “Chinese saving rate puzzle,” it does contribute to understanding why the Chinese household saving rate had been rising.¹ Our quantitative exercise shows that the pension reform, which significantly reduced the generosity of pension benefits for urban workers, could be an important driving force behind the rising saving rate for the period 1995–2009. Our paper also contributes to the strand of the literature on quantifying the effects of China’s demographic transition on its household savings rate, such as Choukhmane, Coeurdacier and Jin (2014) and Curtis, Lagauer and Mark (2015). Both papers emphasize the importance of declining fertility due to the “one child policy” on Chinese households’ savings behavior. We complement their research by focusing on another end of the demographic change spectrum, namely the rising life expectancy due to rapid aging. In addition, both papers abstract from the impact of demographic change on the aggregate labor supply, which is an important focus of our paper.

This paper is also related to the literature on Chinese pension reforms. Feng, He, and Sato (2011) estimate the impact of the 1997 pension reform and find that empirically the pension reform boosted the household saving rate in 1999 by about 6–9 percentage points for the cohort aged 25–29 and by about 2–3 percentage points for the cohort aged 50–59. We emphasize the importance of the labor supply as another instrument to self-insure against pension benefit uncertainties. Our paper is also closely related to Song, Storesletten, Wang and Zilibotti (2015), who highlight the redistribution channel of social security in a fast growing country and focus on the optimal intergenerational redistribution along the transition path of a Chinese economy facing future pension reform and demographic transitions. Our focus here is the current impacts of the demographic change and pension reforms on the household saving rate and labor supply. In addition, Song et al. (2015) abstract from idiosyncratic risks, whereas how individuals use their savings and labor supply to insure against idiosyncratic risks over life cycle is our main focus.

The rest of this paper is organized as follows. Section II provides some stylized facts on the changes in the household saving rate and labor supply in urban China. It also describes the empirical analysis of the impact of the pension reform on the labor supply and the household saving rate, based on Chinese household survey data. Section III describes the model. Section IV illustrates the calibration of the model. Section V presents the simulation results after we input rapid aging and the pension reform into the benchmark model. It also conducts several counterfactual experiments. Finally, Section VI concludes.

II Empirical Evidence

In this section, we first provide data describing the evolution of the household saving rate and labor supply for 1990s and 2000s. We then go deeper, with the micro-level household survey data, to empirically analyze the impact of the pension reforms on households’ saving behavior and labor supply.

A Trend of Household Saving Rate and Labor Supply

What happened to the Chinese economy after the rapid aging and radical pension reform mentioned above? A well known fact is that the household saving rate in China has also been rising very dramatically, making China rank the highest in terms of the saving rate among all countries. Figure 1 shows the aggregate urban household saving rate for 1990–2009, which we constructed from the Chinese Urban Household Survey (UHS).² As shown in this figure, the Chinese urban household saving rate had been fairly stable from 1990 until 1995. It then began to increase very rapidly after 1996. In 2009, the rate exceeded 28%. Compared to its level in 1995, which was about 15.5%, the saving rate had increased by about 12.7 percentage points over these fourteen years.

View Full Size

Chinese Urban Household Savings Rate

Citation: IMF Working Papers 2019, 061; 10.5089/9781498302890.001.A001

• Download Figure
• Download figure as PowerPoint slide

However, a less known fact about China over duirng the 1990s and 2000s is that the Chinese also worked much longer than they did before. In Figure 2, we report the average weekly working hours per worker in urban China for 1996–2010 from a micro-level household survey called Chinese Health and Nutrition Survey (CHNS).³ The panel data structure of the CHNS also makes tracking changes in working hours over time more credible. We therefore use the data from the CHNS as our main data source for the labor supply. The most striking pattern from Figure 2 is that urban workers dramatically increased their labor supply after the late 1990s. The average weekly working hours per worker jumped from 42.4 hours in 1996 to 45.9 hours in 2003. To check the robustness of the pattern found in the CHNS data, we also plot the weekly working hours per worker from the Chinese Household Income Project Survey (CHIPS) in the same graph.⁴ We again see a significant increase in weekly working hours from 1995 to 2007. Finally, the UHS reports the weekly working hours per worker as well, but only for a limited time period from 2002 to 2006. For comparison, we also plot the available data from the UHS in the same figure. The data from the UHS are broadly consistent with the CHNS data on the rising trend even for this limited time period.

View Full Size

Average Weekly Working Hours in Urban China

Citation: IMF Working Papers 2019, 061; 10.5089/9781498302890.001.A001

• Download Figure
• Download figure as PowerPoint slide

B Empirical Analysis Using Micro-level Survey Data

In this section, we provide our empirical analysis of the effect of changes in pension benefits on the labor supply and household saving rate using the CHIPS data.

B.1 The Impact of the Pension Reform on the Labor Supply

We first consider the following regression model for household i to analyze the impact of the changes in pension benefits on the household labor supply:

$$\begin{array}{cc}\ln \left(workinghou{r}_i\right)={\beta }_0+{\beta }_{1}pensio{n}_i+{\beta }_2{X}_i+{\epsilon }_i& \left(1\right)\end{array}$$

where pension_i is a dummy variable that indicates whether household i is enrolled in the pension system or not. X_i is a set of household characteristics including age, age-squared, gender, wage, number of children, family income, spouse’s wage, marital status, education, occupation, health status, disability status, ownership of the firm in which the household head is employed, labor contract tenure, homeownership status, industry, and provincial dummies.

We use CHIPS 2002 and 2007 to construct the variables used in the regression. We measure “working hour” by weekly working hours for worker i.⁵ We drop working hours < 0 and working hours > 112. Pension is a dummy variable for pension status: 1 for having a pension, 0 otherwise.⁶,⁷ We restrict our sample to urban workers aged 25–55 for females and 25–60 for males.⁸

Our theory first implies that across individuals, households with pension = 1 tend to work less than households with pension = 0, other things being equal. In other words, our hypothesis 1 is that for a given year t, α₁ < 0. In addition, as we describe in detail in the Appendix, the pension system in China underwent a significant change in 2005. The generosity of the pension system was significantly decreased in that year.⁹ It thus provides a unique opportunity to conduct a difference-in-differences analysis for Equation (1). Compared to workers who do not have a pension (they thus tend to supply more labor), workers with a pension will work significantly fewer hours in 2002. This should be captured in the coefficient $${\beta }_1^{2002}<0$$. However, in 2007, workers with pension = 1, facing much less generous pension benefits in the future, would increase their labor supply to hedge against the change in pension benefits. This implies that the difference in labor supply between workers with and without pensions in 2007 should be smaller compared to the gap in 2002. Therefore, if we run the regression equation (1) for 2002 and 2007 separately, we expect to see $${\beta }_1^{2002}<{\beta }_1^{2007}$$. And the statistical significance of α₁ might also decrease from 2002 to 2007. This is our hypothesis 2.

We also conduct an alternative empirical specification:

$$\begin{array}{cc}\ln \left(workinghou{r}_{it}\right)={\alpha }_0+{\alpha }_{1}yea{r}_t+{\alpha }_{2}pensio{n}_{it}+{\alpha }_{3}yea{r}_t\times pensio{n}_{it}+{\alpha }_4{X}_{it}+{\epsilon }_i& \left(2\right)\end{array}$$

where year_t is a dummy variable which is equal to one if t = 2007, and zero if t = 2002. Corresponding to hypothesis 1 above, our prediction is α₂ < 0, i.e., workers with a pension tend to supply less labor. And probably more importantly, corresponding to hypothesis 2 above, we expect to see α₃ > 0. In other words, workers with a pension in 2007 are expected to supply more labor since they face less generous pension benefits in the future.

Notice that in the specification above (equation (2)), we assume that other than pension dummy, controls are constant across two years. In the following specification, we relax this assumption and have the more flexible specification to include interaction terms between year dummy and other controls:

$$\begin{array}{cc}\ln \left(workinghou{r}_{it}\right)={\alpha }_0+{\alpha }_{1}yea{r}_t+{\alpha }_{2}pensio{n}_{it}+{\alpha }_{3}yea{r}_t\times pensio{n}_{it}+{\alpha }_4{X}_{it}+{\alpha }_{5}yea{r}_t\times X_{it}+{\epsilon }_i.& \left(3\right)\end{array}$$

We again expect to see α₂ < 0 and α₃ > 0.

We also consider two possible alternative explanations for the dramatically rising working hours after 1997 (shown in Figure 2) in our specification. The first one could be that the remaining SOE workers had to take a larger workload after a large-scale layoff of the SOE workers due to the SOE reform the occurred from 1997 to 2001.¹⁰ In addition, another possible alternative explanation could be that the large-scale layoff provides a very strong incentive for the remaining SOE workers to work harder for signalling purposes. We control for those explanations based on the corresponding CHIPS data in the regression.¹¹

Table 1 shows the estimation results for the coefficients of greatest interest, namely β₁, α₂ and α₃, and other selected coefficients. In this table, the first two columns show the results for Equation (1) using the 2002 and 2007 samples separately. ${\beta }_1^{2002}=-0.024$, meaning that having public pension would reduce an individual’s working time by 2.4 percent, and ${\beta }_1^{2007}=0.007$. We do have ${\beta }_1^{2002}<{\beta }_1^{2007}$. And the statistical significance of β₁ does decrease from the 1% level in 2002 to not statistically significant in 2007. Thus hypotheses 1 and 2 are confirmed. In addition, “laid-off before” turns out to be significantly positive. Individuals who had been laid off tend to supply more labor, possibly due to a stronger precautionary motive. However, the workload channel via firm ownership change seems to be not significant.

Table 1:

The Effect of Pension Reform on Labor Supply

Source: CHIPS 2002, 2007. Note: Robust standard errors in parentheses; * p<0.10, ** p<0.05, *** p<0.01. The unreported controls include education level, health status, disability, firm ownership, occupation, industry, province, and “hukou” dummies.

Table 1:

The Effect of Pension Reform on Labor Supply

	(1)	(2)	(3)	(4)
	Year 2002	Year 2007	NO year×X	Year×X
pension X year			0.045*** (0.015)	0.034** (0.016)
pension	-0.024*** (0.007)	0.007 (0.015)	-0.035*** (0.006)	-0.028*** (0.006)
year			-0.043*** (0.015)	0.069 (0.198)
log(wage)	-0.010*** (0.004)	-0.011 (0.023)	-0.015*** (0.004)	-0.013*** (0.003)
male	0.045*** (0.006)	0.041*** (0.009)	0.044*** (0.005)	0.044*** (0.006)
age	0.000 (0.005)	0.003 (0.005)	0.001 (0.003)	0.000 (0.004)
age2 X1000	-0.020 (0.059)	-0.061 (0.068)	-0.032 (0.039)	-0.013 (0.050)
marriage	0.018 (0.014)	-0.007 (0.015)	0.009 (0.009)	0.017 (0.012)
children	0.025*** (0.007)	0.010 (0.011)		0.018*** (0.006)
log(household income)	-0.000 (0.007)	0.019 (0.015)		-0.005 (0.006)
homeownership	-0.007 (0.008)	-0.030* (0.018)	-0.015** (0.006)	-0.011 (0.007)
change_ownship	0.001 (0.008)
layoff before	0.048*** (0.012)
Constant	3.733*** (0.156)	3.756*** (0.167)	3.841*** (0.085)	3.868*** (0.120)
Observations	6,624	4,624	13,110	13,044
Adjusted R2	0.101	0.038	0.065	0.073

Source: CHIPS 2002, 2007. Note: Robust standard errors in parentheses; * p<0.10, ** p<0.05, *** p<0.01. The unreported controls include education level, health status, disability, firm ownership, occupation, industry, province, and “hukou” dummies.

View Table

Table 1:

The Effect of Pension Reform on Labor Supply

	(1)	(2)	(3)	(4)
	Year 2002	Year 2007	NO year×X	Year×X
pension X year			0.045*** (0.015)	0.034** (0.016)
pension	-0.024*** (0.007)	0.007 (0.015)	-0.035*** (0.006)	-0.028*** (0.006)
year			-0.043*** (0.015)	0.069 (0.198)
log(wage)	-0.010*** (0.004)	-0.011 (0.023)	-0.015*** (0.004)	-0.013*** (0.003)
male	0.045*** (0.006)	0.041*** (0.009)	0.044*** (0.005)	0.044*** (0.006)
age	0.000 (0.005)	0.003 (0.005)	0.001 (0.003)	0.000 (0.004)
age2 X1000	-0.020 (0.059)	-0.061 (0.068)	-0.032 (0.039)	-0.013 (0.050)
marriage	0.018 (0.014)	-0.007 (0.015)	0.009 (0.009)	0.017 (0.012)
children	0.025*** (0.007)	0.010 (0.011)		0.018*** (0.006)
log(household income)	-0.000 (0.007)	0.019 (0.015)		-0.005 (0.006)
homeownership	-0.007 (0.008)	-0.030* (0.018)	-0.015** (0.006)	-0.011 (0.007)
change_ownship	0.001 (0.008)
layoff before	0.048*** (0.012)
Constant	3.733*** (0.156)	3.756*** (0.167)	3.841*** (0.085)	3.868*** (0.120)
Observations	6,624	4,624	13,110	13,044
Adjusted R2	0.101	0.038	0.065	0.073

Source: CHIPS 2002, 2007. Note: Robust standard errors in parentheses; * p<0.10, ** p<0.05, *** p<0.01. The unreported controls include education level, health status, disability, firm ownership, occupation, industry, province, and “hukou” dummies.

The third column of Table 1 shows the results for Equation (2): α₂ = -0.035 and it is significant at the 1% level. More importantly, the coefficient of the interaction term α₃ = 0:045 and it is significant at the 1% level. This shows that individuals did respond to the 2005 pension reform (which further reduced the generosity of the pensions and strengthened the link between the pension benefits and one’s own past labor income) by significantly increasing their working time by about 4.5 percent.

Finally, the fourth column of Table 1 reports the results for the more flexible specification Equation (3). We see that α₂ = 0:028 and it is significant at the 1% level. And α₃ = 0:034 and it is significant at the 5% level. The hypotheses again are confirmed in this more flexible and our most favorable specification.

B.2 Impact of Pension Reform on the Saving Rate

Next, we would like to empirically identify the impact of the pension reform on the household saving rate. Similar to Equation (1), we consider the following regression model for household i:

$$\begin{array}{cc}savingrat{e}_i={\beta }_0+{\beta }_{1}pensio{n}_i+{\beta }_2{X}_i+{\epsilon }_i.& \left(4\right)\end{array}$$

We again use CHIPS 2002 and 2007 to construct the saving rate, which is 1 – household consumption/household disposable income. Following Curtis, Lugauer and Mark (2015), we control for gender, age, age-squared, number of children, household income, and education, in X. Besides these, we also control for marital status, share of elderly (age ≥ 65) in the household, firm ownership, working status, home-ownership, home value, province, and household registration status (“hukou”).

Similar to the argument above for the labor supply, our theory implies that workers with a pension tend to save less, other things being equal, i.e., β₁ < 0. In addition, our DiD approach by comparing 2002 and 2007 also implies that the gap between the saving rate of workers with and without a pension should be smaller (and also might be less statistically significant) in 2007 compared to 2002, i.e., $${\beta }_1^{2002}{\beta }_1^{2007}$$.

We also ran an alternative empirical specification to equation (4):

$$\begin{array}{cc}savingrat{e}_{it}={\alpha }_0+{\alpha }_{1}yea{r}_t+{\alpha }_{2}pensio{n}_{it}+{\alpha }_{3}yea{r}_t\times pensio{n}_{it}+{\alpha }_4{X}_{it}+{\epsilon }_i.& \left(5\right)\end{array}$$

We again expect to see α₂ < 0 and α₃ > 0.

Finally, as in equation (3), we also run regresssion based on a more flexible specification:

$$\begin{array}{cc}savingrat{e}_{it}={\alpha }_0+{\alpha }_{1}yea{r}_t+{\alpha }_{2}pensio{n}_{it}+{\alpha }_{3}yea{r}_t\times pensio{n}_{it}+{\alpha }_4{X}_{it}+{\alpha }_{5}yea{r}_t\times X_{it}+{\epsilon }_i.& \left(6\right)\end{array}$$

Table 2 presents the results for equations (4), (5), and (6). In this table, the first two columns show the results for equation (4) using the 2002 and 2007 samples separately. We find $${\beta }_1^{2002}=-0.027$$ but it is significant at the 5% level. Having pension benefits would reduce the household saving rate by 2.7% in 2002, other things being equal. However, after the 2005 pension reform, $${\beta }_1^{2007}$$ is 0.040 and it is not significant. We thus do have $${\beta }_1^{2002}<{\beta }_1^{2007}$$. The third column shows the results for equation (5): α₂ = 0.028 (significant at 5% level), while α₃ = 0.058 but it is not significant. Finally, results of the more flexible specification equation 6 in the fourth column are similar to those in the third column, with α₂ = 0.027 (significant at the 5% level) and α₃ = 0.067 (now significant at 10% level), showing that the 2005 pension reform significantly increased the household saving rate by 6.7%.

Table 2:

The Effect of the Pension Reform on Saving Rate

Source: CHIPS 2002, 2007. Note: Robust standard errors in parentheses; * p<0.10, ** p<0.05, *** p<0.01. The unreported controls include firm ownership, working status, industry, province, and “hukou” dummies.

Table 2:

The Effect of the Pension Reform on Saving Rate

	(1)	(2)	(3)	(4)
VARIABLES	Year 2002	Year 2007	NO year×X	Year×X
pension × year			0.058 (0.038)	0.067* (0.038)
pension	-0.027** (0.011)	0.040 (0.036)	-0.028** (0.012)	-0.027** (0.012)
male	0.040*** (0.011)	0.045** (0.021)	0.044*** (0.010)	0.039*** (0.011)
age	-0.030*** (0.009)	0.023 (0.017)	-0.011 (0.008)	-0.030*** (0.009)
age2	0.371*** (0.104)	-0.311 (0.200)	0.124 (0.100)	0.373*** (0.105)
marriage	-0.028 (0.023)	0.004 (0.073)	-0.012 (0.033)	-0.029 (0.023)
household size	-0.008 (0.014)	-0.056*** (0.016)	-0.024** (0.011)	-0.008 (0.014)
elder	0.016 (0.022)	-0.034 (0.044)	-0.006 (0.022)	0.016 (0.022)
children	-0.048*** (0.018)	-0.047 (0.030)	-0.043*** (0.015)	-0.048*** (0.018)
log(household income)	0.191*** (0.012)	0.412*** (0.044)	0.290*** (0.021)	0.189*** (0.013)
homeownership	0.332*** (0.085)	0.189 (0.150)	0.349*** (0.080)	0.337*** (0.085)
log(home value)	-0.028*** (0.008)	-0.023* (0.012)	-0.031*** (0.007)	-0.029*** (0.008)
Constant	-1.175*** (0.225)	-4.271*** (0.667)	-2.474*** (0.314)	-1.114*** (0.225)
Observations	4,308	2,166	6,474	6,474
Adjusted R2	0.103	0.233	0.155	0.178

Source: CHIPS 2002, 2007. Note: Robust standard errors in parentheses; * p<0.10, ** p<0.05, *** p<0.01. The unreported controls include firm ownership, working status, industry, province, and “hukou” dummies.

View Table

Table 2:

The Effect of the Pension Reform on Saving Rate

	(1)	(2)	(3)	(4)
VARIABLES	Year 2002	Year 2007	NO year×X	Year×X
pension × year			0.058 (0.038)	0.067* (0.038)
pension	-0.027** (0.011)	0.040 (0.036)	-0.028** (0.012)	-0.027** (0.012)
male	0.040*** (0.011)	0.045** (0.021)	0.044*** (0.010)	0.039*** (0.011)
age	-0.030*** (0.009)	0.023 (0.017)	-0.011 (0.008)	-0.030*** (0.009)
age2	0.371*** (0.104)	-0.311 (0.200)	0.124 (0.100)	0.373*** (0.105)
marriage	-0.028 (0.023)	0.004 (0.073)	-0.012 (0.033)	-0.029 (0.023)
household size	-0.008 (0.014)	-0.056*** (0.016)	-0.024** (0.011)	-0.008 (0.014)
elder	0.016 (0.022)	-0.034 (0.044)	-0.006 (0.022)	0.016 (0.022)
children	-0.048*** (0.018)	-0.047 (0.030)	-0.043*** (0.015)	-0.048*** (0.018)
log(household income)	0.191*** (0.012)	0.412*** (0.044)	0.290*** (0.021)	0.189*** (0.013)
homeownership	0.332*** (0.085)	0.189 (0.150)	0.349*** (0.080)	0.337*** (0.085)
log(home value)	-0.028*** (0.008)	-0.023* (0.012)	-0.031*** (0.007)	-0.029*** (0.008)
Constant	-1.175*** (0.225)	-4.271*** (0.667)	-2.474*** (0.314)	-1.114*** (0.225)
Observations	4,308	2,166	6,474	6,474
Adjusted R2	0.103	0.233	0.155	0.178

Source: CHIPS 2002, 2007. Note: Robust standard errors in parentheses; * p<0.10, ** p<0.05, *** p<0.01. The unreported controls include firm ownership, working status, industry, province, and “hukou” dummies.

In conclusion, our empirical analysis shows that households in urban China did respond to the pension reform by significantly increasing both their labor supply and saving rate. With this strong support from the empirical evidence, we will try to build a theoretical model to analyze an individual’s optimal responses to changes in longevity risk and pension benefits, and the aggregate implications of these responses.

B.3 Heterogeneity Matters on the Impacts

Although we control for a list of variables in our benchmark regressions reported above, by no means we have exhausted the potential important factors affecting households’ labor supply and saving behavior. In what follows, we would like to extend the empirical analysis above to different dimensions including more heterogeneity.

First, the impact of pension reform on labor supply and saving rate could be different between healthy and unhealthy workers. Unhealthy workers might expect to have a shorter life expectancy and a lower income. Expecting a shorter life expectancy, they may decrease labor supply and reduce savings. On the other hand, expecting lower pension income and higher medical expenditures during retirement period, they shall increase working hours and raise savings. According to their self-reporting health status, we divide workers in CHIPS into three groups: healthy (those report their health status to be “excellent” and “good”), fine (those report their health status to be “fine”), and unhealthy (those report their health status to be “bad”). We run regressions based on our most favorable specification in equations (3) and (6), with a triple interaction term including now worker’s health status. To control the potential impact of medical insurance on labor supply and savings, we restrict our data sample to only include those who have medical insurance:

$$\begin{array}{cc}{Y}_{it}={\alpha }_0+{\alpha }_{1}yea{r}_t+{\alpha }_{2}pensio{n}_{it}+{\alpha }_{3}yea{r}_t\times pensio{n}_{it}+{\alpha }_4{X}_{it}+{\alpha }_{5}yea{r}_t\times X_{it}+{\alpha }_{6}yea{r}_t\times pensio{n}_{it}\times healt{h}_{it}+{\epsilon }_i.& \left(7\right)\end{array}$$

Y_it is either log of working hours or saving rate.

Column 1 in Table 3 shows the results of equation (7) for labor supply. Comparing to the group of healthy workers (reference group), the impact of the pension reform on labor supply for workers with fine health would be 8% lower (significant at the 5% level). For unhealthy workers, the impact would be 5.5% lower (but not significant). The life expectancy effect seems dominate the income effect.

Table 3:

Heterogeneity on the Impact of Pension Reform on Labor Supply

Source: CHIPS 2002, 2007. Note: Robust standard errors in parentheses; * p<0.10, ** p<0.05, *** p<0.01. The unreported controls include education level, health status, disability, firm ownership, occupation, industry, province, and “hukou” dummies.

Table 3:

Heterogeneity on the Impact of Pension Reform on Labor Supply

	(1)	(2)	(3)	(4)	(5)	(6)
VARIABLES	Health	Howner	H value	Old	GOV	Migrate
pension*year*health (fine)	-0.080** (0.033)
pension*year*health (bad)	-0.055 (0.081)
pension*year*homeowner		0.064 (0.063)
pension*year*ln(h value)			-0.023* (0.012)
pension*year*older				-0.005 (0.039)
year*gov					-0.034** (0.015)
pension*year*migration						0.066 (0.048)
pension*year	0.052** (0.023)	-0.025 (0.061)	0.309** (0.149)	0.033 (0.028)		0.031** (0.015)
log (wage)	-0.005 (0.004)	-0.013*** (0.003)	-0.013*** (0.004)	-0.016*** (0.003)	-0.013** (0.007)	-0.013*** (0.003)
male	0.031*** (0.006)	0.044*** (0.006)	0.041*** (0.006)	0.041*** (0.007)	0.042*** (0.007)	0.033*** (0.005)
age	0.002 (0.004)	0.000 (0.004)	0.001 (0.005)		-0.002 (0.005)	0.006* (0.004)
age2	-0.038 (0.053)	-0.014 (0.051)	-0.029 (0.056)		0.005 (0.064)	-0.092** (0.044)
marriage	0.008 (0.012)	0.017 (0.012)	0.019 (0.013)	0.009 (0.011)	0.008 (0.013)	0.006 (0.011)
children	0.017*** (0.006)	0.018*** (0.006)	0.013** (0.006)	0.016** (0.006)	0.015** (0.007)	0.003 (0.005)
log(household income)	0.000 (0.007)	-0.008 (0.007)	-0.006 (0.007)	-0.010 (0.008)	0.001 (0.008)	0.004 (0.005)
Constant	3.777*** (0.132)	3.880*** (0.121)	3.908*** (0.140)	3.894*** (0.110)	3.972*** (0.146)	3.686*** (0.101)
Observations	10,142	13,044	11,471	8,388	9,558	18,127
Adjusted R2	0.034	0.073	0.067	0.079	0.040	0.331

Source: CHIPS 2002, 2007. Note: Robust standard errors in parentheses; * p<0.10, ** p<0.05, *** p<0.01. The unreported controls include education level, health status, disability, firm ownership, occupation, industry, province, and “hukou” dummies.

View Table

Table 3:

Heterogeneity on the Impact of Pension Reform on Labor Supply

	(1)	(2)	(3)	(4)	(5)	(6)
VARIABLES	Health	Howner	H value	Old	GOV	Migrate
pension*year*health (fine)	-0.080** (0.033)
pension*year*health (bad)	-0.055 (0.081)
pension*year*homeowner		0.064 (0.063)
pension*year*ln(h value)			-0.023* (0.012)
pension*year*older				-0.005 (0.039)
year*gov					-0.034** (0.015)
pension*year*migration						0.066 (0.048)
pension*year	0.052** (0.023)	-0.025 (0.061)	0.309** (0.149)	0.033 (0.028)		0.031** (0.015)
log (wage)	-0.005 (0.004)	-0.013*** (0.003)	-0.013*** (0.004)	-0.016*** (0.003)	-0.013** (0.007)	-0.013*** (0.003)
male	0.031*** (0.006)	0.044*** (0.006)	0.041*** (0.006)	0.041*** (0.007)	0.042*** (0.007)	0.033*** (0.005)
age	0.002 (0.004)	0.000 (0.004)	0.001 (0.005)		-0.002 (0.005)	0.006* (0.004)
age2	-0.038 (0.053)	-0.014 (0.051)	-0.029 (0.056)		0.005 (0.064)	-0.092** (0.044)
marriage	0.008 (0.012)	0.017 (0.012)	0.019 (0.013)	0.009 (0.011)	0.008 (0.013)	0.006 (0.011)
children	0.017*** (0.006)	0.018*** (0.006)	0.013** (0.006)	0.016** (0.006)	0.015** (0.007)	0.003 (0.005)
log(household income)	0.000 (0.007)	-0.008 (0.007)	-0.006 (0.007)	-0.010 (0.008)	0.001 (0.008)	0.004 (0.005)
Constant	3.777*** (0.132)	3.880*** (0.121)	3.908*** (0.140)	3.894*** (0.110)	3.972*** (0.146)	3.686*** (0.101)
Observations	10,142	13,044	11,471	8,388	9,558	18,127
Adjusted R2	0.034	0.073	0.067	0.079	0.040	0.331

Source: CHIPS 2002, 2007. Note: Robust standard errors in parentheses; * p<0.10, ** p<0.05, *** p<0.01. The unreported controls include education level, health status, disability, firm ownership, occupation, industry, province, and “hukou” dummies.

Column 1 in Table 4 reports the results of equation (7) for saving rate. Comparing to the group of healthy workers, the impact of the pension reform on household saving rate for workers with fine health would be 5% higher (not significant). For unhealthy workers, however, the impact would be 21.8% higher (significant at the 5% level). It appears that the worry of old age consumption dominates the life expectancy effect and pushes unhealthy workers save much more than their healthy counterparts.

Table 4:

Heterogeneity on the Impact of Pension Reform on Saving rate

Source: CHIPS 2002, 2007. Note: Robust standard errors in parentheses; * p<0.10, ** p<0.05, *** p<0.01. The unreported controls include firm ownership, working status, medical insurance, industry, province, and “hukou” dummies.

Table 4:

Heterogeneity on the Impact of Pension Reform on Saving rate

	(1)	(2)	(3)	(4)	(5)	(6)
VARIABLES	Health	Howner	Hvalue	Old	GOV	Migrate
pension*year*health (fine)	0.050 (0.176)
pension*year*health (bad)	0.218** (0.102)
pension*year*homeowner		0.090 (0.079)
pension*year*log(h value)			0.031 (0.035)
pension*year*older				-0.100 (0.141)
year*gov					-0.035* (0.021)
pension*year*migration						-0.107* (0.061)
male	0.046*** (0.013)	0.040*** (0.012)	0.035*** (0.012)	0.025* (0.015)	0.043*** (0.011)	0.042*** (0.010)
age	-0.027*** (0.010)	-0.028*** (0.009)	-0.025** (0.010)		-0.007 (0.009)	-0.034*** (0.007)
age2	0.334*** (0.118)	0.345*** (0.106)	0.328*** (0.120)		0.084 (0.111)	0.398*** (0.090)
marriage	-0.040 (0.027)	-0.031 (0.024)	-0.027 (0.027)	-0.021 (0.032)	-0.039 (0.028)	-0.031 (0.023)
household size	0.001 (0.018)	-0.005 (0.015)	-0.001 (0.016)	0.005 (0.017)	-0.039*** (0.013)	-0.055*** (0.016)
elder	0.008 (0.027)	0.012 (0.023)	0.009 (0.023)	-0.010 (0.029)	-0.014 (0.029)	0.049** (0.022)
children	-0.069*** (0.021)	-0.051*** (0.018)	-0.048** (0.020)	-0.060*** (0.020)	-0.044*** (0.017)	0.001 (0.020)
ln(income)	0.165*** (0.014)	0.193*** (0.013)	0.205*** (0.015)	0.195*** (0.017)	0.314*** (0.028)	0.272*** (0.022)
Observations	4,814	6,176	5,374	3,843	4,396	9,484
Adjusted R2	0.181	0.178	0.186	0.173	0.197	0.190

Source: CHIPS 2002, 2007. Note: Robust standard errors in parentheses; * p<0.10, ** p<0.05, *** p<0.01. The unreported controls include firm ownership, working status, medical insurance, industry, province, and “hukou” dummies.

View Table

Table 4:

Heterogeneity on the Impact of Pension Reform on Saving rate

	(1)	(2)	(3)	(4)	(5)	(6)
VARIABLES	Health	Howner	Hvalue	Old	GOV	Migrate
pension*year*health (fine)	0.050 (0.176)
pension*year*health (bad)	0.218** (0.102)
pension*year*homeowner		0.090 (0.079)
pension*year*log(h value)			0.031 (0.035)
pension*year*older				-0.100 (0.141)
year*gov					-0.035* (0.021)
pension*year*migration						-0.107* (0.061)
male	0.046*** (0.013)	0.040*** (0.012)	0.035*** (0.012)	0.025* (0.015)	0.043*** (0.011)	0.042*** (0.010)
age	-0.027*** (0.010)	-0.028*** (0.009)	-0.025** (0.010)		-0.007 (0.009)	-0.034*** (0.007)
age2	0.334*** (0.118)	0.345*** (0.106)	0.328*** (0.120)		0.084 (0.111)	0.398*** (0.090)
marriage	-0.040 (0.027)	-0.031 (0.024)	-0.027 (0.027)	-0.021 (0.032)	-0.039 (0.028)	-0.031 (0.023)
household size	0.001 (0.018)	-0.005 (0.015)	-0.001 (0.016)	0.005 (0.017)	-0.039*** (0.013)	-0.055*** (0.016)
elder	0.008 (0.027)	0.012 (0.023)	0.009 (0.023)	-0.010 (0.029)	-0.014 (0.029)	0.049** (0.022)
children	-0.069*** (0.021)	-0.051*** (0.018)	-0.048** (0.020)	-0.060*** (0.020)	-0.044*** (0.017)	0.001 (0.020)
ln(income)	0.165*** (0.014)	0.193*** (0.013)	0.205*** (0.015)	0.195*** (0.017)	0.314*** (0.028)	0.272*** (0.022)
Observations	4,814	6,176	5,374	3,843	4,396	9,484
Adjusted R2	0.181	0.178	0.186	0.173	0.197	0.190

Source: CHIPS 2002, 2007. Note: Robust standard errors in parentheses; * p<0.10, ** p<0.05, *** p<0.01. The unreported controls include firm ownership, working status, medical insurance, industry, province, and “hukou” dummies.

Second, housing could play an important role in households’ labor supply and saving behavior. First, households without homeownership need to save (and work harder) to be able to buy a house. Second, for existing homeowners, if home values rise, the wealth effect might lead to reduction in working hours and savings. To test the effects of homeownership (“Howner” in Tables 3 and 4) and home value (“Hvalue” in Tables 3 and 4) on the impacts of pension reform, we run the following regressions:

$$\begin{array}{cc}{Y}_{it}={\alpha }_0+{\alpha }_{1}yea{r}_t+{\alpha }_{2}pensio{n}_{it}+{\alpha }_{3}yea{r}_t\times pensio{n}_{it}+{\alpha }_4{X}_{it}+{\alpha }_{5}yea{r}_t\times X_{it}+{\alpha }_{6}yea{r}_t\times pensio{n}_{it}\times hous{e}_{it}+{\epsilon }_i.& \left(8\right)\end{array}$$

where house is either a homeownership dummy (=1 if owns a home, zero otherwise), or a home value.¹²

Column 2 in Table 3 shows the results of equation (8) for labor supply using home-ownership. The impact of pension reform on labor supply does not exhibit a significant difference between homeowners and renters. However, for results using home value (Column 3 in Table 3), α₆ = 0.023, and it is significant at the 10% level, showing higher home value would reduce the impact of pension reform on labor supply, i.e., a strong sign of wealth effect. For the impact of pension reform on the saving rate, columns 2 and 3 in Table 4 show that there is no significant difference between homeowners and renters.

Third, different age groups might be affected by the pension reform differently. We divide workers into two groups: old (age > 45) and young (age < 35). Dummy “older” is one if workers belong to the first group and zero if workers are in the second group. We then run regressions as in equations (7) and (8), with the triple-interaction term year × pension× older. The results are reported in Column 4 in Tables 3 and 4, respectively. We do not see a significant difference between old and young groups in terms of the impact of pension reform.

Fourth, the pension reform affects mostly state-owned enterprises (SOEs) and leaves the government employees largely untouched. To explore the variation on this dimension, we only include households whose head works either in government or SOEs and has pension in the sample and run the following regression:

$$\begin{array}{cc}{Y}_{it}={\alpha }_0+{\alpha }_{1}yea{r}_t+{\alpha }_{2}pensio{n}_{it}+{\alpha }_{3}yea{r}_t\times pensio{n}_{it}+{\alpha }_4{X}_{it}+{\alpha }_{5}yea{r}_t\times X_{it}+{\alpha }_{6}yea{r}_{it}\times GO{V}_{it}+{\epsilon }_i.& \left(9\right)\end{array}$$

where gov dummy is equal to one if household head works for government, zero if he works for SOEs. Column 5 in Tables 3 and 4 both show that the impact of pension reform to government employees is significantly lower compared to their SOE counterparts, as we expect.

Finally, migrated workers (largely from rural area) often do not have access to a public pension as urban hukou residents do. Therefore, one would expect that their reactions to the pension reform should be much smaller compared to urban residents. We thus run the following regression:

$$\begin{array}{cc}{Y}_{it}={\alpha }_0+{\alpha }_{1}yea{r}_t+{\alpha }_{2}pensio{n}_{it}+{\alpha }_{3}yea{r}_t\times pensio{n}_{it}+{\alpha }_4{X}_{it}+{\alpha }_{5}yea{r}_t\times X_{it}+{\alpha }_{6}yea{r}_t\times pensio{n}_{it}\times Migratio{n}_{it}+{\epsilon }_i.& \left(10\right)\end{array}$$

where migration dummy is equal to one if a worker is a migrated worker, zero otherwise.¹³ Column 6 in Tables 3 and 4 shows that this is indeed a case for the saving rate. The impact of pension reform on saving rate for migrated workers is 10.7% lower than the impact for urban residents for whom pension reform would matter much more. However, there is no significant difference of the impact on labor supply, possibly due to the fact that migrated workers are concentrated in industries that maintain notoriously work longer hours (e.g., construction) and hence it is harder for them to adjust their working schedule.

In summary, our investigation on the impact of pension reform on different dimensions of heterogeneity further confirms that pension reform has a significant impact on labor supply and the savings rate across different groups. And there are several competing effects shaping the impact from different angles.

III The Model

In this section, we describe the model economy. The economy is populated with a continuum of many-period lived overlapping generations of individuals. There is a representative firm that uses capital and labor to produce output. The individuals derive utility from consumption and leisure. They supply capital and labor to the firm. The individuals also own the firm. There is a government that imposes payroll taxes to provide social security to retirees. An individual in the economy faces idiosyncratic income risks and lifetime uncertainty. And we assume that an annuity market is missing.