Household consumption and monetary policy in China

Household consumption and monetary policy in China

China Economic Review 13 (2002) 27 – 52 Household consumption and monetary policy in China Yin ZHANGa,*, Guang Hua WANb a Discipline of Economics, S...

244KB Sizes 4 Downloads 45 Views

China Economic Review 13 (2002) 27 – 52

Household consumption and monetary policy in China Yin ZHANGa,*, Guang Hua WANb a

Discipline of Economics, School of Economics and Political Science, Faculty of Economics and Business, The University of Sydney, Sydney, NSW 2006, Australia b Department of Agricultural Economics, Faculty of Agriculture, The University of Sydney, Sydney, NSW 2006, Australia Received 1 January 2001; received in revised form 10 April 2001

Abstract In recent years, the People’s Bank of China (PBC) has been relying increasingly on the adjustment of nominal interest rates for stabilization purposes. This paper employs the Euler equation approach to examine the effect of monetary policy, conducted through the interest rate channel, on household consumption. Econometric estimates of six related models suggest a weak substitution effect of the real interest rate. Inflation rates seem to be more relevant to household consumption decisions than do nominal interest rates. The results provide much evidence for the significance of liquidity constraints, but little for that of precautionary savings. D 2002 Published by Elsevier Science Inc. JEL classification: E13; E21; E52; O53 Keywords: Monetary policy; China; Consumption; Interest rate

1. Introduction Since the beginning of the 1990s, the Chinese monetary authority—the People’s Bank of China (PBC)—has faced new challenges with respect to managing household consumption. At the beginning of the last decade, shortages were eliminated in major consumer goods markets, although supply bottlenecks existed in some sectors such as energy, transportation,

* Corresponding author. Tel.: +61-2-9351-5250; fax: +61-2-9351-4341. E-mail address: [email protected] (Y. Zhang). 1043-951X/02/$ – see front matter D 2002 Published by Elsevier Science Inc. PII: S 1 0 4 3 - 9 5 1 X ( 0 1 ) 0 0 0 5 5 - 4

28

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

and raw materials. Slack demand first appeared in the 1989–1991 downturn but was soon relieved by an investment boom triggered by government policy changes.1 The demand constraint gained renewed attention when economic growth remained slow after the muchacclaimed ‘‘soft landing’’ in 1995. For the PBC, stabilization is no longer simply a question of reigning the expansion of demand, it also involves how to cope with lackluster consumption and investment. Meanwhile, structural transformation of the economy invalidated the traditional tool for managing consumption—the cash plan. When segmentation between the circulation’s of ‘‘household money’’ (cash) and ‘‘enterprise money’’ (bank accounts) broke down, the volume of cash flow was no longer a good indicator of the level of households’ liquidity. With mounting financial assets besides cash, households no longer have to confine their consumption to labor income. Theoretically, this expansion of households’ financial base introduces the possibility of managing consumption through interest rate changes. This possibility introduces intriguing research questions, however, because neoclassical consumption theory offers no clear-cut prediction about the effect of interest rate changes on the consumption of households that are net savers in the current period. Moreover, households may encounter more constraints in reality than in most general theoretical models; this complication is especially likely to be true for a transitional and less developed economy such as China. This paper seeks to evaluate the scope for managing household consumption with monetary instruments in China. In particular, it examines the effect of interest rate changes on consumption (from May 1996 to June 1999 there were seven consecutive interest rate cuts). This will be achieved by estimating an array of Euler equation-based models. Twostage least-squares (2SLS) estimation will be employed with instruments chosen from 14 alternative sets. The paper is organized as follows. The next section reviews household consumption theory and introduces econometric models for empirical implementation. Data issues are discussed in Section 3. Section 4 presents regression procedures and results. Section 5 discusses policy implications and limitations of the study.

2. Consumption theory and econometric models 2.1. Consumer behavior, uncertainty, and liquidity constraints The (real) rate of interest bears on household consumption decisions because it is the relative price of present consumption in terms of future consumption. Changes in interest rates will trigger intertemporal resources reallocation through both substitution and wealth effects. Moreover, a rise in the interest rate makes the same amount of savings more

1

Macroeconomic policy was switched from tightening to expanding right after Deng Xiaoping made a famous speech during his spring tour of southern China in early 1992.

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

29

productive, thus increasing lifetime resources available for consumption in the future. Consumers who prefer present consumption to future consumption will attempt to transfer to the present part of the expected expansion in the wealth stock. This can be called ‘‘asset income’’ effect of an interest rate change since it operates essentially through greater income flows from the accumulated wealth stock. These three effects manifest themselves in the standard optimization problem of a representative consumer who maximizes expected lifetime utility subject only to an intertemporal budget constraint. Assuming (i) the underlying utility function is of the constant-relative-risk-aversion (CCRA) type, (ii) the consumer has perfect foresight (uncertainty is absent), and (iii) the interest rate is constant over time, it can be shown that the optimal level of current consumption is given by t1 #  " 1  X bs ð1 þ rÞs 1 ð1Þ Yt , ð1 þ rÞðWt1  Ct1 Þ þ Ct ¼ 1  1þr 1þr t¼t where t and t index time periods, b is the constant subjective discount factor representing time preference, C is the real value of consumption, W is the stock of assets, Y is the nonasset income, s denotes the intertemporal elasticity of substitution, and r is the real rate of interest. The tension between the substitution effect and the asset income effect is illustrated by the terms in the first bracket on the right-hand side of Eq. (1). The values of these terms determine the ratio of lifetime resources designated for current consumption. When r rises, (1 + r)s increases, indicating that consumption will be tilted towards the future. At the same time, 1/(1 + r) decreases, inducing a move toward current consumption. The second pair of brackets comprises a consumer’s lifetime resources measured in current period goods. It is clear that if current net worth, Wt  1  Ct  1, is greater than zero, the asset income effect and wealth value effect would also work in opposite directions. Hence, the asset income effect offsets the substitution and wealth value effects. The theory provides no a priori prediction as to whether interest rate rises will increase or decrease current consumption. The final outcome hinges on the magnitude of s, the elasticity of intertemporal substitution, and on whether the consumer is a net saver at the beginning of the current period. The assumptions of certainty and perfect capital market are frequently violated in the real world. If the third derivative of the utility function u000(C) > 0 or if the marginal utility function u0(C) is convex in C, greater uncertainty of future consumption raises the expected utility of future consumption, forcing a reduction in current consumption. Although this kind of precautionary savings makes consumption under uncertainty (at any level of income and asset stock) lower than it would be without uncertainty, it is more dominant at the lower range of asset levels (Carroll, 1992). Therefore, when a consumer’s asset stock is below a certain level, the elasticity of consumption to the interest rate tends to be small. On the other hand, if the value of assets is sensitive to the interest rate, which will be the case when the majority of assets in the wealth stock are marketable, interest rate changes may deliver an important impact on consumption by altering the value of assets. A binding liquidity constraint is often given as another possible reason that consumer behavior may depart from the certainty case. Liquidity constraints arise from either

30

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

institutional barriers (e.g., lack of consumer credit) or capital market imperfections, which may result in credit rationing (Stiglitz & Weiss, 1981). Deaton (1991) demonstrates that if income flows are (positively) serially correlated, liquidity-constrained consumers tend to keep a larger buffer-stock of savings than do their unconstrained counterparts. The more constrained are consumers, the less fungible is consumption between different periods, and the greater will be precautionary savings. If, however, income generation follows a random walk process, precautionary savings no longer provide the desired insurance. Consumers simply spend their current income, a behavioral pattern that accords with the Keynesian theory of consumption. For liquidity-constrained consumers, interest rate changes do not impinge directly on their consumption decisions. Whatever influence interest rates may have must come through other channels that alter income flows and/or credit conditions. 2.2. Econometric model specification Empirical evaluation of the effect of interest rate changes on household consumption can be pursued using two approaches. One is to specify an aggregate consumption function with the interest rate as one of the independent variables. This method has gradually fallen out of favor since the Lucas critique (Lucas, 1976), which casts doubt on the stability and even the very existence of such a function. Pioneered by Robert Hall (1978), the other approach involves estimating regression models based on Euler equations. One of such models derived by Hall (1988) is Dct ¼ a0 þ srt þ et ,

ð2Þ

where c denotes the logarithm of C, Dct = ln Ct  ln Ct  1, and e is a white noise error term. Eq. (2) implies that the sensitivity of consumption growth to interest rate changes depends on the value of s. With a large s, modest interest rate changes may generate a substitution effect strong enough to offset the income effect. Higher interest rates would thus lower the level of current consumption even if households were net savers. The scope, therefore, exists for the monetary authority to manage household consumption through manipulating interest rates. In Hall’s model, the real rate of interest is the only current period variable that enters the regression equation. Other variables may influence consumption only to the extent that they reflect innovations in the error term et. Results from empirical studies are often at odds with this model. In particular, researchers have found that income growth explains consumption much better than interest rates. One explanation of this contradiction may lie in liquidity constraint (Deaton, 1991). Campbell and Mankiw (1989) incorporate a liquidity constraint into consumption modeling by assuming that aggregate household consumption is composed of the expenditure of two types of consumers. One group allocate their resources intertemporally while the other simply consume their current income. In this case, the regression model becomes (Eq. (3)): Dct ¼ a0 þ a1 yt þ a2 rt þ et ,

ð3Þ

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

31

where a1 can be viewed as the indicator of the ratio of liquidity-constrained consumers, in other words, it measures how pervasive the liquidity constraint is. Another way to explain the empirical puzzle is by exploring the implications of uncertainty. Carroll (1992) argues that ‘‘prudent’’ and ‘‘impatient’’ consumers have a targeted buffer-stock of assets.2 Before the target is achieved, consumers try to save more, leading to low consumption levels but high consumption growth. Once the buffer-stock is attained, consumers return to normal behavior—allocating resources intertemporally. In a growing economy, steady-state consumption is expected to increase at the rate of income growth, holding everything else constant. To allow for uncertainty in consumption, Caroll proposes a model of the form (Eq. (4)) Dct ¼ slnð1 þ rÞb þ 1=sVarðDct Þ þ et ,

ð4Þ

where DVar(Dct) is the expected variance of next period’s consumption growth, conditional upon information available in the current period, b and a are defined as before. Based on the above discussion, the following equations are considered in this study: Dct ¼ a0 þ a1 Dyt þ et , Dct ¼ a0 þ a2 rt þ et ,

ðModel IÞ ðModel IIÞ

Dct ¼ a0 þ a1 Dyt þ a2 rt þ et ,

ðModel IIIÞ

Dct ¼ a0 þ a1 Dyt þ a2 rt þ a3 VarðDct Þ þ et :

ðModel IV Þ

Model I resembles the Keynesian consumption function, Model II is a restatement of Hall’s model, Model III is the Campbell–Mankiw model, and Model IV is a hybrid of Model III and the Caroll model. In addition, the following two auxiliary models are considered as a check on the robustness of Models I–IV. Dct ¼ a0 þ a1 Dyt þ a2 rt þ a3 VarðDct Þ þ a4 Dct1 þ et

ðModel V Þ

Dct ¼ a0 þ a1 Dyt þ a2 rt þ a3 VarðDct Þ þ a5 ðyt1  ct1 Þ þ et

ðModel VIÞ

These two models are proposed by Chyi and Huang (1997). The lagged term of consumption growth in Model V accounts for factors such as adjustment costs of consumption, habit persistence, or the ‘‘keeping up with the Joneses’’ effect (Chyi & Huang, 1997, p. 1278). Model VI is an error-correction model, stipulating the existence of a long-run equilibrium relationship between y and c. A consumer is prudent if s/he has precautionary motives for savings (i.e., u000(C )>0), and is impatient if s ln(1 + r) b < g, where g is the growth rate of her/is permanent income and s ln(1 + r)b equals the growth rate of consumption when there is no uncertainty (Carroll, 1992). 2

32

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

Relying on these models, it is possible to evaluate monetary policy by estimating the following: (a) a2, an indicator of s, which suggests to what extent consumers are willing to substitute consumption between periods; (b) a1, a measure of the proportion of Keynesian consumers; (c) the significance of the term Var(Dct), namely, whether precautionary savings dominate the behavior of Chinese consumers at the current wealth level. While findings from (a) and (b) help assess the responsiveness of consumption to interest rate adjustments, results from (c) can assist in identifying the primary reason behind these findings, be it liquidity constraints or uncertainty.3 This is useful as previous studies4 have revealed low s values and significant a1.

3. Data and preliminary data analysis Two sets of yearly data are used for model estimation. The main set, ranging from 1966 to 1998, covers both the prereform and the postreform periods. The reference set only includes data for the postreform periods. The reference set only includes data for the postreform period (1979–1997). Appendix A contains a description of the variables and the data sources used in this study. Fig. 1 plots real per capita consumption, income, and nominal interest rates. Clearly, consumption and income rose only modestly in the prereform period (1966 – 1978). Moreover, consumption seemed to be proportional to income, suggesting a stable relationship of the form Ct = bYt. In the postreform period, both variables increased more rapidly, but the growth rate of income was obviously greater than that of consumption so that the gap between them widened over time. There are two candidate explanations for this phenomenon. Firstly, a regime shift may have occurred in household consumption–savings behavior. For one reason or another, households became more willing to save or capable of saving more in the postreform period. Given the profound changes in the socioeconomic structure that occurred after 1978, such changes in behavior would not be too surprising. Fig. B1, charting the growth rates of income and consumption, provides further evidence on the causes of this change in consumption behavior. Before 1979, consumption growth generally moved along with income growth, though the latter fluctuated more widely. During 1979–1998, not only did consumption growth become more volatile, but also no clear pattern between consumption growth and income growth can be discerned. In some years, notably 1992–1998, they 3 Distinguishing the influence of liquidity constraint from that of uncertainty cannot be fully accomplished. Liquidity-constrained consumers also keep, and even keep a larger buffer-stock of savings (Deaton, 1991). 4 See Campbell and Mankiw (1989) for estimates for the United States, Chyi and Huang (1997) for Japan, Korea, the Philippines, Thailand, and Taiwan, and Villagome´z (1997) for 16 developing countries in Latin America and Asia.

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

33

Fig. 1. Consumption, income and nominal interest rates.

actually moved in opposite directions. To allow for the possibility of a structural break in household preferences, Models I–VI are estimated for three periods: the whole sample period (1966–1998) and two subperiods. The break point between the two subsamples is chosen in the light of regression results for the entire sample period. Another possible explanation of the growing discrepancy between income and consumption is errors introduced in the data compiling process. As stated in Appendix A, nonasset income Y in data set 1 is proxied by per capita real GNP. This approximation fails to take into account shifts in the proportion of disposable income in total national income. To assess this approximation error, real GNP is plotted with the series of constructed disposable income from data set 2 in Fig. B2. The difference between the two series does increase over time. However, the two series are highly correlated with a correlation coefficient of .9943. Fig. B3 also suggests the possible presence of measurement errors as the distance between income and consumption series from data set 2 increases, though in a smaller scale than that revealed in Fig. 1. In view of this, data set 2 will be used as a reference for corroboration purposes. Additional relevant information revealed in Fig. 1 is the lack of variability in the nominal interest rate.5 This point is demonstrated more clearly in Fig. 2. Consequently, most of the variability in the real interest rate must come from variations in the inflation rate. This may imply that the nominal interest rate, due to its rigidity, lost the function of signalling changes in asset returns. Instead, the inflation rate may be perceived as a financial price variable directly reflecting movements of returns on assets. This hypothesis is tested by replacing the real interest rate with the inflation rate in estimating Models I–VI. 5

Nominal interest rates are adjusted more frequently after 1985, but are still rather inflexible.

34

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

Fig. B1. Consumption growth, income growth and real interest rates (1996-1998).

As an initial assessment of how the models fit the first data set, Fig. B4 and B5 plot the 5-year moving averages of consumption growth versus those of the real interest rate and income growth, respectively. A clear positive relation can be discerned for the growth rates of consumption and income, which conforms to the predictions that a1 >0. However, the message from Fig. B4 is perplexing. Only 5 (1979–1983, 1980–1984, 1981–1985, 1982–1986, and 1983–1987) among the 29 points are consistent with the theory that a2 (and hence s) is

Fig. B2. Comparison of per capita income of the two data sets.

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

35

Fig. B3. Consumption and income in data set 2.

positive. Overall, there seems to be no definite pattern in the relationship between consumption growth and the real interest rate. Table 1 is constructed to demonstrate the possible interactions between monetary policy and household consumption in recent years. No obvious correlation exists between either the magnitude or the timing of nominal interest rate changes (column 1) and the growth rates of retail sales (column 5), which serve as an indicator of household consumption. The real

Fig. 2. Inflation rates, nominal, and real interest rates.

36

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

Fig. B4. Average consumption growth vs. interest rate.

interest rate (column 4) fell in 1996, but rose by 5.07 percentage points over the following seven quarters. Again, the relationship between the real interest rate and household consumption (retail sales) appears ambiguous. If the substitution effect dominated, the

Fig. B5. Average consumption growth versus income growth.

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

37

Table 1 Recent interest rate adjustment and household responses Date of adjustment and nominal interest ratea 1996 30 April: 10.98 1 May: 9.18 23 August: 7.74 1997 23 October: 5.67

1998d 25 March: 5.22 1 July: 4.77

Quarter

Retail price index

Real interest rate

Retail salesb

Percentage change of new deposits

I II III IV I II III IV I II III

7.7 6.8 5.5 4.5 2.6 1.1 0.2  0.8  1.5  2.6  3.4

3.31 3.18 2.89 2.90 4.84 6.57 7.24 6.60 7.18 7.82 7.97

23.6 19.8 17.4 18.7 13.6 8.9 9.8 9.5 6.9 6.7 7.5

8.4c 19.1  18.5  32.8  14.7  45.2  16.1 49.1  17.3 6.3 –

Source: Chen (1998, p. 8). a Interest rate on 1-year savings deposit. b Percentage change over the same period of the last year. c Percentage change for the whole year. d The other two interest cuts are: December 7, 1998 to 3.78 and June 10, 1999 to 2.25.

growth rate of retail sales would have increased in 1996. Conversely, it would have risen after 1996 if the income effect were dominant. Among the three financial variables, only the inflation rate (retail price index in column 3) exhibits a positive correlation with retail sales. It seems that households are not responsive to the nominal interest rate—its substitution and income effects may have cancelled out. More importantly, the rising real interest rate in 1997–1998 implies that the intention of the PBC to induce consumption by cutting nominal interest rates was offset by falling inflation. Interest rate adjustments in 1996–1998 do not seem to have affected household consumption. The above arguments do not preclude the possible impact of interest rates on the composition of wealth and households’ choice of the means of savings. In fact, a decline in the nominal interest rate may prompt households to transfer long-term bank deposits into more liquid assets. Evidence of such an effect is found in the continuous, negative growth rate of new deposits from the third quarter of 1996 to the third quarter of 1997 and in the first quarter of 1998. The same period saw a spectacular boom in China’s two stock exchanges. The market value of shares in the Shanghai and Shenzhen stock exchanges was RMB 93.8 billion at the end of 1995. The figure more than quintupled to RMB 517.8 billion by the end of 1997. The number of investors increased from 1.23 million to 3.3 million. The government bond market also experienced significant growth at about the same time. In 1995, the issue of government bonds was RMB 15.5 billion. Between 1996 and 1998, a total of more than RMB 80 billion new bonds was added to the stock. The average annual issue exceeded RMB 26.6 billion (Chen, 1998, p. 19). There were changes in the term structure of the savings deposits, too. A recent study shows that during this period, January 1997 to October 1999, the proportion of demand deposits in total savings deposits rose by 0.46 of a percentage point 1 month after a 1 percentage point cut in the deposit rate (Wang, Pu, & Xu, 2000, p.137).

38

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

The finding of a shift in household portfolios seemingly contradicts the earlier assertion that the nominal interest rate does not affect the household consumption–savings decision. The values of shares and bonds are more sensitive to interest rates than are illiquid bank deposits, which in turn makes the value of the wealth stock more easily affected by interest rate. Consequently, changes in nominal interest rate are more likely to influence household consumption through the wealth value effect. Our explanation for the paradox is that because only a small part of household assets are held in the form of shares and bonds,6 the wealth value effect may be negligible. Other reasons have also been suggested, among which are liquidity constraints and increasing income uncertainty. Liquidity constraints and uncertainty are of particular relevance to urban households. After rapid popularization of household appliances such as TVs, refrigerators, and air conditioners in the late 1980s and early 1990s, housing and automobiles have become the next ‘‘musthaves’’ in the household demand hierarchy. In contrast, per capita spending on housing in urban China was only RMB 148.66 in 1997, about 3.55% of total consumption expenditure (Yuan & Song, 1999, p. 23). According to a 1999 household survey conducted in Shanghai, 19.2% of households rank the purchase of housing as the most important reason for their savings, and 55% put it among the top three important reasons (Yuan & Song, 1999, p. 23). The PBC recognized the impediment of liquidity constraints to the development of the housing market, and in addition to reducing lending rates on home loans, it also urged banks to increase credit for house purchases. Increasing income uncertainty induces precautionary savings and can thus dampen the wealth value effect. Since 1995, uncertainty has been exacerbated by the government’s determination to tackle the knottiest part of economic reform—restructuring the SOEs. Predictably, the resulting layoffs weakened consumer confidence in their job security and future income stability. Moreover, in the survey cited above, 69.6% of households considered saving for retirement as one of the top three reasons (Yuan & Song, 1999, p. 23). Notwithstanding that precautionary savings may have become a more dominant motive, saving for retirement in itself is not equivalent to precautionary savings. It is a kind of life cycle consumption behavior, but was previously oppressed (or rather, made unnecessary) by the old welfare system.

4. Regression procedure and results 4.1. Regression procedure Based on the preliminary data analysis, three rounds of regressions were carried out. The first round was based on data set 1 with the real interest rate as the r variable. The second round employed the same data set but used the inflation rate for r. In the last round, data set 2

6 The largest component of rural residents’ wealth stock is nonmarketable housing property. The urban residents keep their savings mainly in the form of savings deposits.

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

39

was used and the r variable was the real interest rate. By assumption, the error term et in Models I–VI is orthogonal to information available at period t  1. Yet, it is still possible for et to be correlated with variables in period t. The correlation between et and r1 (and yt) may bring about simultaneity bias under OLS. In such circumstances, the 2SLS technique should be used and is adopted in this study. Appropriate variables lagged one period are the natural choice for instruments if they are not correlated with et. This would be indeed be the case if data were collected at the same periodicity as that of household decision-making. Unfortunately, data used in this study are annual values while households are likely to make their consumption–savings choice on a much more frequent basis. Consequently, the annual figures are averages of observations generated by a process that is of a higher frequency. According to Working (1960), the first differences of the averages of a random walk process have a first-order serial correlation of about .25. Hence, the transformation imposed by the data collection interval implies Corr(C1,Ct  1)ffi.25. Thus, et is correlated with yt  1, ct  1, and rt  1, and the latter cannot serve as valid instruments. Following Campbell and Mankiw (1989), the instruments used in this paper are variables lagged by two or three periods. To minimize possible errors associated with the choice of instruments, endogenous variables Dy and r were regressed upon 14 instrument sets separately.7 A Wald test was then conducted for each instrument set with the null: all coefficients of instruments except the intercept are jointly zero. The instrument sets finally chosen are those with optimal combinations of Wald statistics for the Dy and r equations. Predicted values of Dy and r were then computed and used in the second stage OLS estimations of Models I–VI. Diagnostic tests from the first two rounds of regressions indicated the existence of first-order autocorrelation and a regime shift around 1985.8 Hence, the sample was split into two subperiods: 1966–1984 and 1985–1998. Estimations were then repeated for the two subperiods and estimates of standard deviations were corrected for first-order autocorrelation. Since data set 2 only contains data for the postreform period, further division is not considered. 4.2. Regression results The empirical results from the first round of regression are reported in Tables 2–4. Table 2 presents estimation results for the period of 1966–1998. It is divided into three blocks. OLS estimates for Models I–III form the first block; they are reported here as a reference. The other two blocks present results with two different instrument sets. In each

7

These instrument sets are an intercept plus: (1) Dyt  2, Dyt  3; (2) Dct  2, Dct  3; (3) rt  2, rt  3; (4) Dyt  2, Dyt  3, Dct  2, Dct  3; (5) Dyt  2, Dyt  3, rt  2, rt  3; (6) Dct  2, Dct  3, rt  2, rt  3; (7) Dyt  2, Dyt  3, Dct  2, Dct  3, Dct  3,rt  2,rt  3;(8)Dyt  2,Dyt  3,Dct  2,Dct  3,rt  2,rt  3,yt  2  ct  2;(9)Dyt  2,Dyt  3,Dct  2,Dct  3,pt  2, pt  2, pt  3; (10) Dyt  2, Dyt  3, Dct  2, Dct  3, pt  2, pt  3, yt  2  ct  2; (11) Dyt  2, Dyt  3, Dct  2, Dct  3, it  2,it  3; (12) Dyt  2,Dyt  3,Dct  2,Dct  3,it  2,it  3,yt  2  ct  2; (13) Dyt  2,Dyt  3,Dct  2,Dct  3,rt  2, rt  3, rt  3, pt  2, pt  3; and (14) Dyt  2, Dyt  3, Dct  2, Dct  3, rt  2, rt  3, pt  2, pt  3, yt  2  ct  2. Here,p represents the inflation rate, r represents the real interest rate, and i represents the nominal interest rate. 8 The break point was determined on the basis of sequential Chow test.

40

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

Table 2 Regression results based on data set 1 for the period 1966 – 1998 Model Ia Model IIa Model IIIa OLS

a1

0.4640 (2.811)**

a2

IV set I:b Dyt  2, yt  3, DCt  2, Ct  3, rt  2, rt  3; Wald statistics for: Dy equation: 0.034, r equation: 0.084

adjusted R2 0.303 DW 1.108 a1 0.3632 (1.980)*

0.4579 (2.616)**  0.1028  0.0362 (  0.889) (  0.329)  0.016 0.280 0.9641 1.125 0.3436 (1.794)*

 0.1829  0.1332 (  0.695) (  0.522)

a2 a3

Model IV a Model V a

0.3264 (1.821)*

 0.1174 (  0.521) 0.0894 (0.989)

a4 a5

IV set II:b Dyt  2, Dyt  3, DCt  2, DCt  3, rt  2, rt  3, yt  2  Ct  2; Wald statistics for: Dy equation: 0.065, r equation: 0.036

adjusted R2 DW a1

0.042 1.017 0.3261 (1.697)

a2

 0.016 0.936

0.017

0.055

1.019 0.3049 (1.662)

1.146 0.2992 (1.614)

 0.2597  0.2381 (  1.458) (  1.286)

a3 a4 a5 adjusted R2 0.027 DW 0.999 a

0.014 0.980

0.034 1.042

0.2607 (2.021)*

Model VIa

0.3699 (2.053)*

0.0605  0.1058 (0.246) (  0.516) 0.0123 0.0574 (0.121) (0.578) 0.4499 (3.08)** 0.1110 (2.080)** 0.214 0.127 1.858 0.2378 (1.888)*

1.212 0.3805 (2.096)**

 0.2169  0.0024  0.1119 (  1.290) (  0.011) (  0.644) 0.0367  0.0073 0.0138 (0.426) (  0.085) (0.145) 0.4666 (3.461)** 0.1223 (1.96)* 0.013 0.207 0.099 1.063 1.824 1.169

Figures in parentheses are t ratios. IV set refers to the instrument set selected to obtain the regression results shown in the columns on the right. * Indicates the coefficient is significant at 10% level. ** Indicates the coefficient is significant at 5% level. b

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

41

block, the instruments employed, along with the P values of the Wald test statistics for the Dy and r equations are given in the left-most cell. The six columns on the right side of the table correspond to the six models under consideration. Each column reports the relevant coefficient estimates, their Newey–West method corrected t ratios, R2 , and the Durbin– Watson (DW) statistic. As is evident from columns 3 and 4 of Table 2, neither the simple Keynesian model Model I nor the Hall model Model II fits the data well. The Campbell–Mankiw model Model III and the augmented Caroll model Model IV do not register a significant improvement, either. Judging by R2, the autoregressive model Model V performs the best, which implies a slow response of consumption to new information. Overall, liquidity-constrained Keynesian behavior tends to be significant while the roles of interest rates and uncertainty are negligible. However, given the conclusion of the Chow test that the coefficients are unstable, results in Table 2 cannot be accepted at their face value. Table 3 resembles Table 2 except that data used for estimation only cover the period 1966– 1984. Two points noted in Table 2 reappear here. First, Model V is now by far the best model with the highest R2 and larger coefficients of Dct  1. (The a4 estimates increased by around .24 to .27 compared with those in Table 2.) Second, the coefficients of the interest rate and uncertainty terms continue to be insignificant in most models. Note, however, that when the second instrument set was used, Model IV produced a significant but negative coefficient estimate for the interest rate variable, and Model V produced a significant but negative coefficient estimate for the uncertainty term. Strictly interpreting these estimates, we must infer a nonconcave utility function. Another unexpected result is that the ratio of liquidityconstrained consumers seems to be slightly lower than indicated in Table 2. Since the average income of this period is lower than that for the entire sample period, one would anticipate consumers to be more liquidity-constrained. These results become less puzzling once the special economic environment prevailing in 1966–1984 is considered. For most of the period, the Chinese economy operated under a centrally planned system, which provided households with job security and income stability. Prices, interest rates, and wages were usually fixed for a long time. Health care, education, housing, and pensions were provided by the government. These measures and policies insured households against exceptionally low levels of consumption, diminishing the need for precautionary savings, and inducing smooth consumption paths. Moreover, as consumption was just above the subsistence level and savings were mainly delayed consumption expenditure for consumer durables, both consumption and savings exhibited downward rigidity. Another factor is that government policy had favored heavy industries at the expense of consumption goods industries. Pricing was such that households were more liable to be constrained by supply shortages rather than lack of liquidity. Given the habits that had been formed in the prereform period, increasing uncertainty about the future led households to accumulate consumer goods as protection against possible future shortages, rather than to increase their saving. Accordingly, current spending would increase, leading to a drop in consumption growth and thus the negative sign of the uncertainty term. On the other hand, the negative effect on consumption of the interest rate variable may be construed as reflecting a positive effect of inflation. Since the nominal interest rate barely varied, a change in the real

42

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

Table 3 Regression results based on data set 1 for the period 1966 – 1984 Model Ia Model IIa Model IIIa OLS

a1

0.2769 (1.697)

a2

IV set I:b Dyt  2, Dyt  3, Dct  2, Dct  3, it  2, it  3, yt  2  ct  2; Wald statistics for: Dy equation: 0.147, r equation: 0.005

adjusted R2 DW a1

0.142 0.543 0.1908 (0.803)

a2 a3

 0.468 ( 1.505)  0.009 0.497

0.3241 (2.291)**  0.687 (  2.513)** 0.214 0.875 0.3707

Model IV a Model V a

0.3706 (1.356)

 0.0857  0.7689  0.7681 (  0.160) (  1.082) (  0.863) 0.0004 (0.001)

a4

0.2364 (2.064)*

 0.3749 (  0.669)  0.1643 (  0.710) 0.7052 (3.560)**

a5

IV set II:b Dyt  2, Dyt  3, Dct  2, Dct  3, pt  2, pt  3; Wald statistics for: Dy equation: 0.186, r equation: 0.001

adjusted R2  0.013 DW 0.563 a1 0.1666 (0.723)

a2 a3 a4

 0.070 0.473

 0.017 0.739 0.4187 (1.936)*

 0.102 0.740 0.4805 (2.103)*

0.353 1.525 0.2388 (2.164)*

 0.2122  1.0104  1.4869  0.4542 (  0.425) (  1.692) (  2.583)** (  1.215)  0.2250  0.2664 (  1.442) (  3.226)** 0.7174 (3.827)**

a5 adjusted R2  0.023 DW 0.539 a

 0.061 0.498

0.027 0.986

0.055 0.901

0.480 1.543

Model VIa

0.3962 (3.00)**

0.9407 (1.324) 0.1078 (0.375)

0.3753 (4.296)** 0.275 1.596 0.5755 (3.861)**

 0.4931 (  0.794)  0.1956 (  1.667)

0.3072 (3.111)** 0.337 1.198

Figures in parentheses are t ratios. IV set refers to the instrument set selected to obtain the regression results shown in the columns on the right. * Indicates the coefficient is significant at 10% level. ** Indicates the coefficient is significant at 5% level. b

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

43

Table 4 Regression results based on data set 1 for the period 1985 – 1998 Model Ia OLS

a1

0.9169 (5.248)**

a2

IV set I:b Dyt  2, Dyt  3, Dct  2, Dct  3, it  2, it  3, yt  2  ct  2; Wald statistics for: Dy equation: 0.006, r equation: 0.044

adjusted R2 DW a1

Model IIa

0.596 1.820 0.7358 (3.844)**

 0.0064 (  0.058)  0.083 1.501

Model IIIa

1.026 (8.570)** 0.1602 (2.599)** 0.643 1.919 0.7278 0.8239 (2.224)** (3.068)**

 0.2092  0.0074 (  3.083)** (  0.0421)

a2

Model IVa Model Va Model VIa

a3

0.0039 (0.268) 0.0383 (0.355)

a4

0.8231 0.8289 (2.884)** (3.210)**

0.0387 (0.267) 0.0382 (0.345) 0.0011 (0.07)

a5 adjusted R2 DW

0.252 1.614

0.022 1.536

0.184 1.610

0.124 1.715

0.027 1.716

0.0091 (0.054) 0.0470 (0.427)

0.0664 (0.599) 0.039 1.691

a

Figures in parentheses are t ratios. IV set refers to the instrument set selected to obtain the regression results shown in the columns on the right. ** Indicates the coefficient is significant at 5% level. b

interest rate was just a change of inflation rate in the opposite direction. Because a large part of household savings was in the form of nonmarketable real assets, whose returns rose with inflation, a drop in the real interest rate would raise returns on real assets and encourage savings. Next, we turn to Table 4 to gauge the situation in the postreform period. Since only one instrument set passed the Wald test, one set of estimates is reported here. A striking feature of Table 4 is the large and significant coefficients of Dy. Households seem to have been much more liquidity-constrained during this period than in the past. It is noted that the estimated coefficients of the interest rate variable and the uncertainty term are both positive in Models IV–VI, although they remain small and insignificant. In addition, the autoregressive terms, Dct  1 and yt  1  ct  1, ceased to be significant. Viewed as a whole, Table 4 gives a snapshot that captures reasonably well the consumption behavior of households in a transitional economy: households have started reacting to financial variables and uncertainty, but deficiencies in the financial system and markets confined their ability of allocating resources intertemporally.

44

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

Results in Tables B1–B3 are obtained using the same data as those in Tables 2–4, except that the inflation rate instead of the real interest rate is now used for the variable r. As stated earlier, this is done to test the functioning as price signals of non-market-determined, infrequently adjusted nominal interest rates. This practice is justifiable in a partial equilibrium framework where both the nominal interest rate and the inflation rate can be treated as exogenous. Tables B1 and B2 are very much similar to their parallels (cf. Tables 2 and 3), but with higher R2. In both tables, the coefficients of the inflation rate are positive and significant. The evidence points to the acceptance of the hypothesis that the inflation rate, rather than the interest rate, enters households’ decision-making process. The coefficient estimates of the inflation rate in Table B3 nearly mirror those of the real interest rate in Table 4. The estimates are generally insignificant and tend to be negative. Table B4 presents the regression results using data set 2. As none of the 14 instrument sets demonstrate sufficient power in forecasting Dy and r, the credibility of the estimates must be viewed very cautiously. We note that although a narrower measure of Y is adopted here, the coefficients of Dy are still large, suggesting that perhaps households are more liquidityconstrained in the postreform period.

5. Major findings and policy implications In summary, this study suggests a weak substitution effect of interest rate adjustment on consumption. We find that consumption is more volatile and sensitive to external shocks in the postreform period than in the prereform period. More households are constrained by the lack of liquidity in the postreform period, which in turn hampers the transmission of monetary policy. Unavailability of sufficient data (e.g., more observations for the postreform period, or cross-sectional household data) prevents a conclusive assessment on the importance of uncertainty, but the evidence does not support a significant role of precautionary savings. Our empirical results are similar to those reported in earlier studies (e.g., Li, 1999; Research Bureau of the PBC, 1999; Wang et al., 2000; Yang & Li, 1997; Zhang, 1997). Particularly, they corroborate the insignificance of interest rate effects and the importance of liquidity constraints. However, this study also provides evidence that the direct effect of interest rate adjustment on consumption is not likely to increase much even if some institutional constraints are loosened or removed in the future. Since the substitution effect is shown to be too weak to be operational, and the relative importance of the income and wealth effects will be influenced by such factors as demographic changes and the evolution of the financial system, the outcome from the interaction among the three effects is still hard to predict. The empirical results lead to four implications for the conduct of monetary policy in China: 1. Households demonstrate different consumption behavior in the postreform period than in the prereform period. Most notable is the fact that the autoregressive model Model V and the error-correction model Model VI no longer describe consumption growth in the

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

45

Table B1 Regression results based on data set 1 for the period 1966 – 1998 (inflation rate for variable r) Model Ia OLS

a1

0.4640 (2.811)**

a2

IV set I:b Dyt  2, Dyt  3, Dct  2, Dct  3, rt  2 , rt  3 , yt  2  ct  2: Wald statistics for Dy equation: 0.065, p equation: 0.017

adjusted R2 DW a1

Model IIa

0.303 1.108 0.3261 (1.697)

a2

0.1551 (1.349) 0.034 1.057

0.3019 (2.387)**

Model IIIa 0.4392 (2.476)** 0.0882 (0.804) 0.300 1.185 0.3029 (1.820)*

0.2906 (2.230)**

a3 a4 a5

IV set II:b Dyt  2, Dyt  3, Dct  2, Dct  3, rt  2 , rt  3 , p t  2 , pt  3, yt  2  ct  2; Wald statistics for: Dy equation: 0.011, p equation: 0.035

adjusted R2 DW a1

0.027 0.999 0.3364 (1.484)

a2

0.080 1.139

0.3230 (2.885)**

0.102 1.195 0.2665 (1.300)

0.2842 (2.249)**

a3 a4 a5 adjusted R2 DW a

0.057 0.929

0.108 1.104

0.133 1.075

Model IVa

0.2974 (1.770)*

Model Va

0.2394 (1.920)*

Model VIa

0.3728 (1.860)*

0.2771 0.0674 0.0964 (2.242)** (0.427) (0.398) 0.0313  0.0065 0.0171 (0.357) (  0.076) (0.177) 0.4307 (2.972)** 0.1054 (0.996) 0.079 0.212 0.097 1.192 1.779 1.181 0.2759 0.2329 0.3538 (1.353) (1.625) (1.759)*

0.2738 0.0797 0.0891 (2.298)** (0.555) (0.442) 0.0242  0.0083 0.0264 (0.217) (  0.078) (0.226) 0.4128 (3.063)** 0.1047 (1.194) 0.104 0.231 0.124 1.098 1.675 1.086

Figures in parentheses are t ratios. IV set refers to the instrument set selected to obtain the regression results shown in the columns on the right. * Indicates the coefficient is significant at 10% level. ** Indicates the coefficient is significant at 5% level. b

46

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

Table B2 Regression results based on data set 1 for the period 1966 – 1984 (inflation rate for variable r) Model Ia OLS

a1

0.2769 (1.697)

a2

IV set I:b Dyt  2, Dyt  3, Dct  2, Dct  3, it  2, it  3, yt  2  ct  2; Wald statistics for: Dy equation: 0.147, p equation: 0.052

adjusted R2 DW a1

Model IIa

0.142 0.544 0.1908 (0.803)

a2

1.2420 (3.599)** 0.394 1.177

1.1242 (1.989)*

Model IIIa 0.2686 (2.973)** 1.2253 (4.447)** 0.564 1.920 0.3449 (2.317)**

1.4341 (3.051)**

a3

Model IVa

0.3187 (2.495)**

Model Va

0.2239 (1.922)*

IV set II:b Dyt  2, Dyt  3, Dct  2, Dct  3, rt  2, rt  3, pt  2, pt  3; Wald statistics for: Dy equation: 0.185, p equation: 0.002

a2

0.171 0.811

1.0887 (3.062)**

0.292 1.326 0.3581 (2.457)**

a4 a5

a

0.207 1.084

0.2683 1.152 0.3376 (2.804)**

0.398 1.509 0.2421 (2.127)*

0.1833 (1.088) 0.253 1.189 0.4273 (3.226) **

1.4382 (4.219)**

a3

adjusted R2  0.024 DW 0.539

0.4356 (2.974) **

1.6868 0.7769 0.8002 (3.362)** (1.344) (0.7847)  0.1567  0.1787  0.0911 (  0.735) (  0.877) (  0.3802)

a4 a5 adjusted R2  0.013 DW 0.563 a1 0.1666 (0.723)

Model VIa

0.353 1.905

1.8168 0.9692 1.3110 (6.885)** (1.837)* (3.831)**  0.2799  0.2789  0.2549 (  0.091) (  3.848)** (  2.975)** 0.4667 (1.744) 0.1381 (1.502) 0.4813 0.5453 0.476 1.961 1.855 1.695

Figures in parentheses are t ratios. IV set refers to the instrument set selected to obtain the regression results shown in the columns on the right. * Indicates the coefficient is significant at 10% level. ** Indicates the coefficient is significant at 5% level. b

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

47

Table B3 Regression results based on data set 1 for the period 1985 – 1998 (inflation rate for variable r) Model Ia OLS

a1 a2

IV set I:b Dyt  2, Dyt  3, Dct  2, Dct  3, it  2, it  3, yt  2  Ct  2; Wald statistics for: Dy equation: 0.006, p equation: 0.091

adjusted R2 DW a1

Model IIa

Model IIIa Model IVa

0.9824 (7.585)**  0.0163  0.1269 (0.884) (  1.847)* 0.596  0.082 0.620 1.820 1.494 1.865 0.7358 0.7135 (3.844)** (2.505)**

Model VIa

0.9169 (5.248)**

a2

0.1837 (2.725)**

0.0249 (0.176)

a3

0.7787 (3.070)**

 0.0028 (  0.022) 0.0333 (0.309)

a4 a5 adjusted R2 DW

Model Va

0.252 1.614

0.001 1.601

0.185 1.611

0.122 1.680

0.7802 0.8068 (2.821)** (3.556)**

 0.0025  0.0135 (  0.0198) (0.098) 0.0335 0.0452 (0.301) (0.408)  0.0022 (  0.014) 0.0712 (0.711) 0.024 0.039 1.677 1.678

a

Figures in parentheses are t ratios. IV set refers to the instrument set selected to obtain the regression results shown in the columns on the right. * Indicates the coefficient is significant at 10% level. ** Indicates the coefficient is significant at 5% level. b

postreform period. This behavioral change has implications for macroeconomic management. While it allows for the possibility of manipulating consumption to offset fluctuations in other components of the aggregate demand, more volatile consumption can in itself be a source of instability or a mechanism magnifying instability. 2. The substitution effect of interest rate changes appears rather weak. The magnitude of a2, the coefficient of the interest rate, is positively related to the intertemporal elasticity of substitution, s, and negatively related to the ratio of liquidity-constrained consumers. The estimates of a2 suggest that s cannot be very large. Hence, the cut (on June 10, 1999) in the 1-year savings deposit rate by the PBC from 3.78% to 2.25% would, ceteris paribus, cause little change in the current level of consumption. Because savings are mainly driven by delayed consumption on durables resulting from liquidity constraints, rather than speculative motives, the income effect of interest rate changes is likely to be small as well. Consequently, households do not consider interest accrued to bank deposits as an important component of income. The only mechanism through which adjustment in interest rates can effect consumption is the wealth value effect. Apart from

48

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

Table B4 Regression results based on data set 2 for the period 1979 – 1997 Model Ia Model IIa Model IIIa OLS

IV set I:b Dyt  2, Dyt  3, Dct  2, Dct  3, rt  2, rt  3; Wald statistics for: Dy equation: 0.238, r equation: 0.503

a1

0.5386 (2.743)**  0.2359 a2 (1.468) adjusted R2 0.396 0.080 DW 2.037 1.271 a1 0.8054 (3.204)**

a2

 0.4314 (  1.519)

a3 a4 a5

IV set II:b Dyt  2, Dyt  3, Dct  2, Dct  3, rt  2, rt  3, pt  2, pt  3, yt  2  ct  2; Wald statistics for: Dy equation: 0.669, r equation: 0.215

adjusted R2 0.420 0.117 DW 1.567 1.314 a1 0.6673 (2.772)**

a2

 0.3169 (  1.394)

a3 a4 a5 adjusted R2 0.306 DW 1.466 a

0.111 1.433

0.5335 (3.487)**  0.2289 (  2.096)* 0.503 2.079 0.7447 (3.056)**

Model IVa

0.8948 (4.047)**

Model Va

0.9254 (3.873)**

Model VIa

0.9298 (4.122)**

 0.3282 (  1.822)*

 0.2632  0.2569  0.2663 (  2.153)* (  2.187) * (  2.245)**  0.2042  0.2163  0.1830 (  2.347)** (  2.323)** (  1.891)*  0.0489 (  0.455) 0.0789 (0.976) 0.490 0.582 0.547 0.560 1.621 2.305 2.260 2.205 0.6962 0.7071 0.6679 0.8104 (2.977)** (3.282)** (2.842)** (3.605)**

 0.3443  0.3725  0.3745  0.3581 (  2.274)** (  2.839)** (  2.917)** (  2.609)  0.2146  0.19591  0.1205 (  1.373) (  1.358) (  0.692) 0.0792 (0.501) 0.1424 (0.197) 0.484 0.490 0.450 0.486 1.578 1.874 2.007 1.846

Figures in parentheses are t ratios. IV set refers to the instrument set selected to obtain the regression results shown in the columns on the right. * Indicates the coefficient is significant at 10% level. ** Indicates the coefficient is significant at 5% level. b

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

49

altering the current value of nonasset income, this effect may also work through the inverse relationship between interest rates and asset prices if a large proportion of household assets is marketable such as shares and bonds. In short, deposit rate adjustments in China should not aim to induce direct substitutions between consumption and savings so much as to initiate a shift in the composition of household portfolios. Yet the wealth value effect may be hindered by liquidity constraint and diminished by non-market-determined interest rates. 3. The role of expected inflation in affecting household consumption decision should not be underestimated. Given that interest rates are still administered in China, the expected inflation rate may continue to be perceived as a substitute for market-determined interest rate. However, the sign of the effect of expected inflation could well be reversed in the future. During 1966–1984, expected inflation exhibits a positive effect on consumption growth (and hence a negative effect on current consumption) due to the special structure of household portfolios. As the ratio of financial assets in the total stock of household assets has been constantly on the rise, it is anticipated that expected inflation would assume a different role. The inverse relationship between real returns on financial assets and the inflation rate implies that higher expected inflation will discourage asset accumulation and increase current spending. 4. Finally, the pervasiveness of liquidity constraint constitutes a major obstacle in the transmission of monetary policy. This problem becomes more serious at the trough of the business cycle when macroeconomic policies are directed to reflating the economy. Unless cuts in interest rates can improve conditions in the credit market, much of the central bank’s effort will be in vain. Our results can be challenged on at least two grounds. Evaluating the strength of intertemporal substitution presumes that households plan over multiple periods. While a long planning horizon is inherent in the rational expectations assumption, the institutional setting can be such that individual households find it perfectly rational to consider the current period only. In other words, planning over a lifetime is no different than living for the present. The prereform Chinese economy may be an example of such an environment, especially for urban households. Various subsidies and welfare programs equalized household income, provided insurance, and moderated income fluctuations. Thus, it is possible that the expected income profile of individual consumers was a horizontal line when plotted against time. When tomorrow was almost certain to be a repetition of today, consumers would have little motivation to look beyond the present. If such an environment shortened consumers’ effective planning horizon, the regression models would be misspecified and the results obtained would be invalid. Another specification issue arises from the fact that the whole economic structure in China is undergoing transition from a centrally planned system to a market system. As old social groups disintegrate, people find their socioeconomic status redefined. It is, therefore, hard to contend that preferences of the representative agent are immune to the profound changes in society. If preference shocks not only had occurred but also were correlated, they would have entered the disturbance term in the regression models and invalidated the results.

50

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

Further research would undoubtedly benefit policymaking and evaluation. Among the many directions that might be pursued, separate treatment of rural and urban households is of particular interest. Rural areas were not covered by the old welfare system, and rural reforms started earlier and moved faster than urban reforms. Thus, the consumption behavior of rural households is likely to more closely approximate the households in other developing economies than that in urban China. The persistence of urban–rural duality over the sample period makes separate examination of rural and urban consumption behavior more appropriate than a one-size-fits-all approach. Further, disaggregation by commodity, particularly into durables and nondurables, would also be useful.

Acknowledgments The authors wish to thank Associate Professor Tony Aspromourgos, Associate Professor Jeff Sheen, Dr. Michael Plumb, Atta Adu-Osae, the participants of the research students’ seminar at the School of Economics and Political Science, and the two anonymous referees for their constructive comments on an earlier version of this paper.

Appendix A. The data Two sets of data are used in this paper. The first were taken from the World Development Indicators (WDI, World Bank, 2000) and various issues of the Statistical Yearbook of China (SYC). Specifically, income and consumption series are from WDI, while interest rates and inflation rates are computed from SYC. The second set comes from a 1999 study by the research bureau of the PBC published in Jingji Yanjiu (Economic Research Journal) (Research Bureau of the PBC, 1999). It serves as the reference data set. All variables are annual observations and are expressed on a per capita basis where appropriate. Ideally, per capita real consumption C shall only contain nondurable consumption. However, there is only scattered information about household expenditure on durables. Therefore, the C data in both sets include consumption on durables (but C in data set 2 does not include household investment in fixed assets and inventory). The same problem exists with respect to Y, where per capita nonasset income is replaced by per capita real GNP in data set 1 and by disposable personal income in data set 2. The inflation rate of retail sales is chosen to indicate inflation, as the CPI index was not reported until 1985. The selection of a nominal interest rate has always been a controversial issue in the literature. In this paper, the official 1-year deposit rate is used because there is no publication on the interest rates in the informal market. For most of the sample period, savings deposits were the only form of noncash financial assets available to households. The real interest rate is thus calculated as the nominal deposit rate i subtracted by next period’s inflation rate of retail sales p. Var(Dc1) is proxied by the squared one-step ahead forecast error of consumption growth.

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

51

Some descriptive statistics of the data employed are provided below. Descriptive statistics Number of observations

Mean

Standard deviation

Variance

Minimum

Maximum

Data set 1 Y (yuan) C (yuan) Dc Dy i p r

33 33 33 33 33 33 33

630.85 1251.44 5.35 6.62 5.85 4.01 1.84

369.77 863.96 3.76 5.47 2.62 5.75 4.50

136,727.63 746,420.24 14.12 29.91 6.85 33.07 20.24

276.55 388.34  3.28  8.09 3.24  2.63  9.77

1473.77 3315.68 13.03 16.15 11.34 19.64 9.26

Data set 2 Y (yuan) C (yuan) Dc Dy

20 20 19 19

429.96 561.51 8.03 9.07

194.28 281.04 3.55 4.34

37,743.31 78,982.89 12.63 18.84

183.23 204.38  0.57  0.15

842.60 1145.28 13.89 15.02

Correlation between data of the two sets C .9943 Y .9933 Dc .6768 Dy .3132 For data set 1, Dc and Dy are directly taken from WDI, while for data set 2, they are calculated by the authors from the C and Y series.

References Campbell, J.Y., & Mankiw, N.G. (1989). Consumption, income and interest rates: reinterpreting the time series evidence. In: O. Blanchard, & S. Fischer (Eds.), NBER macroeconomic annual 1989 ( pp. 185 – 216). Cambridge: MIT Press. Carroll, C.D. (1992). The buffer-stock theory of saving: some macroeconomic evidence. Brookings Papers on Economic Activity, 2, 61 – 156. Chen, D.Q. (1998). Reflections on the operation of macroeconomic policy in the last two years. Jingji Yanjiu (Economic Research Journal) 12, 3 – 12. Chyi, Y.L., & Huang, C.H. (1997). An empirical study of the ‘rule of thumb’ consumption model in five East Asian countries. Applied Economics, 29 (10), 1271 – 1282. Deaton, A. (1991). Savings and liquidity constraints. Econometrica, 59 (5), 1221 – 1248. Hall, R. (1978). Stochastic implications of the life cycle-permanent income hypothesis: theory and practice. Journal of Political Economy, 86, 971 – 987.

52

Y. Zhang, G.H. Wan / China Economic Review 13 (2002) 27–52

Hall, R. (1988). Intertemporal substitution in consumption. Journal of Political Economy, 96, 339 – 357. Li, Y. (1999). An inquiry into the relationship between the interest rate and savings of Chinese households. Jingji Yanjiu (Economic Research Journal) 11, 39 – 46. Lucas, R.E. Jr. (1976). Econometric policy evaluation: a critique. Carnegie-Rochester Conference Series on Public Policy, 1, 19 – 46. Research Bureau of the PBC (1999). The influential factors of China’s household savings and national saving. Jingji Yanjiu (Economic Research Journal) 5, 3 – 10. Statistical Yearbook of China. Beijing: Statistical Publishing House of China. (various issues). Stiglitz, J.E., & Weiss, A. (1981). Credit rationing in markets with imperfect information. American Economic Review, 71, 393 – 410. Villagome´z, F.A. (1997). Private saving, interest rates and liquidity constraints in LDCs: recent evidence. Applied Economics, 29 (5), 607 – 615. Wang, X.Y., Pu, Y.X., & Xu, Y. (2000). An empirical analysis of the impact of interest rate cuts on savings, loans and money supply. Jingji Yanjiu (Economic Research Journal) 6, 11 – 18. Working, H. (1960). Note on the correlation of first differences of averages in a random chain. Econometrica, 28, 916 – 918. World Bank. (2000). World development indicators. Washington, DC: Author. Yang, X.D., & Li, Y. (1997). The fundamental consumption theory and the evolution of consumption of urban residents in Shanghai: an empirical study. Jingji Yanjiu (Economic Research Journal) 9, 29 – 36. Yuan, Z.G., & Song, Z. (1999). Consumption behavior of urban households and economic growth in China. Jingji Yanjiu (Economic Research Journal) 11, 20 – 28. Zhang, P. (1997). Statistical test, institutional explanation and macroeffect analysis on consumers’ behavior. Jingji Yanjiu (Economic Research Journal) 2, 43 – 51.