How Fragmented is the Capital Market in China? Genevieve Boyreau-Debray * Shang-Jin Wei **

Transcription

1 Preliminary version June 2002 Comments welcome How Fragmented is the Capital Market in China? Genevieve Boyreau-Debray * Shang-Jin Wei ** Abstract This paper uses two capital mobility tests to analyze financial integration across 28 Chinese provinces over the period. The first test, the so-called Feldstein and Horioka (1980), examines the correlation between investments and savings. The second test, which is drawn from the risk-sharing literature, uses consumption data to evaluate financial integration. We find that capital is much less mobile within China than it is within financially integrated countries like Japan or the United States. Overall the degree of inter-provincial capital mobility in China appears closer to the one of international capital mobility between OECD countries than an unified country. We also find that financial integration significantly decreased in the 1990s. This decrease is explained by the surge of foreign direct investment over the 1990s, which may have intensified disparities in investment financing between provinces and so further fragmented regional capital markets. * Development Research Group, The World Bank, gboyreaudebray@worldbank.org. ** Research Department, International Monetary Fund, swei@imf.org.

2 Introduction How fragmented is the capital market in China? How common is it for savings generated in one part of China to be channeled to another area where investment may be more productive? How has the efficiency of the regional allocation of capital evolved over time? These questions are of great importance but, as far as we know, have not received a formal investigation. China spatial economic organization is traditionally described as a de facto federalism, involving a decentralized economic system in which each region can be considered as autonomous economic entity, in terms of both goods and production factors 1. By contrast to the well documented patterns of intra-national trade, no formal study is available on the degree of intra-national capital mobility in China, albeit available information supports the view of a high degree of capital market fragmentation 2. This paper aims to fill this void by applying two basic tools commonly used in the literature on measuring international capital mobility and financial integration. The first is the so-called Feldstein and Horioka (1980) (FH) test, which examines the correlation between investment and saving. The idea behind the FH test is that under the null hypothesis of a perfectly integrated capital market, investment in one region should not be constrained by the available savings in that region. Hence, the correlation between local savings and local investment should not be high. The second approach uses consumption-based tests from the risk sharing literature to evaluate the degree of financial integration. The idea is that under the hypothesis of perfect capital mobility and asset market completeness, households should be able to smooth consumption over time and across regions by transacting in capital markets. As a result, consumption volatility should be smaller than income volatility and consumption growth rates should be more correlated across countries or regions than income growth rates. The paper is divided into two main sections which respectively analyze the regional correlations of investment and saving on the one hand, and, on the other hand, consumption risk-sharing using a dataset of 28 Chinese provinces between 1978 and 1 See Qian and Xu (1993), Poncet (2002). 2 The only comprehensive report on the topic is provided by the World Bank (1994), which finds no evidence of price or returns to capital convergence across provinces as it would be the case if provinces were financially integrated. 2

3 2000. The organization of the two sections is symmetric. First we present descriptive statistics and we compare them to other corresponding international and intra-national figures. Second, we test more formally for capital market integration within China by applying two methodologies borrowed from the literature on measuring capital mobility. The first is proposed by Iwamoto and van Wincoop (2000) for to investment and saving co-movements (section 2.2). The second is developed by Obstfled (1994) for risk-sharing (section 3.2). Section 4 concludes. 1. Investment-Saving Correlation within China Feldstein and Horioka (1980) propose a simple test for capital mobility. With perfect capital mobility, there should be no relation between domestic saving and domestic investment, as saving responds to the worldwide opportunities for investment while investment is financed by the worldwide pool of capital. Conversely, as in a closed economy incremental saving is invested in the country, a correlation of one to one between investment and saving should be found. Therefore, the closer to one (zero) the correlation between investment and saving, the lower (higher) the degree of capital mobility. The authors test for the proposition using a sample of OECD countries and find a positive and close to one correlation between investment and saving, which has proved to remain robust since then to different specifications or econometric methods. This result sharply contrasts with the common view of an increasing capital integration within those countries and thereby has questioned the validity of the test taken as a test for capital mobility. The reasons why the FH test is not ideal in a cross-country context have been extensively debated in the literature. It does not give conclusive evidence on the degree of international capital market integration as a low saving-investment correlation can be consistent with various alternatives other than a low degree of capital market integration across countries. These alternatives include the presence of currency devaluation risk premium, and governments effort to target the level of current account balance by manipulating the exchange rate. However, within a country, FH test turns out to be a reasonable indicator on the degree of capital market integration across different regions. 3

4 The main reason is that the alternative interpretations mentioned above are not operative within a country. Table 1 reports the results of the intra-national evidence of investmentsaving correlations provided in the literature. Of five separate papers that have looked at Canada, Germany, Japan, United States and United Kingdom, countries that are known to have an integrated capital market internally, all have found a very low or even negative saving-investment correlation across the regions. This provides a justification for using the FH test to examine capital market integration within China. In this section, we first present and compare simple correlations between saving and investment in the Chinese provinces with corresponding international and intra-national figures. We transpose thereafter to the Chinese provinces the test of investment-saving correlation applied to the Japanese prefectures by Iwamoto and van Wincoop (2000) Descriptive Statistics We use a dataset of 28 Chinese provinces (all the Chinese provinces but Tibet) for the reform period ( ). The data come from the database published by A11 China Marketing Research (2001) and consists of provincial GDP desegregated into household and government consumption, gross capital formation, change in inventory and net exports. Provincial savings are defined as regional GDP minus private consumption and government consumption. Provincial investment is defined as the change in gross capital formation both of the private and public sectors. Both saving and investment are expressed as a ratio of provincial GDP 3. Table 2 reports the correlations of investment and saving rates in the Chinese provinces averaged over the reform period ( ) and two sub-periods ( and ). For comparison purposes, Table 2 also reports the corresponding figures for the OECD national economies 4. The investment and saving correlation in the Chinese provinces averages 0.49 between 1978 and By comparison, correlation between investment and saving rates 3 We have tested for the order of integration of the series using panel unit-root tests (Im, Pesaran and Shin, 1997). Both investment and saving rates are stationary. 4 The data of investment and saving for OECD countries come from the United Nations National Accounts Statistics. 4

5 within OECD countries such as Japan or United States, where it can be assumed that there is no regional barrier to intra-national capital flows are either insignificant or even negative (see Table 1). At the opposite extreme, the correlation between investment and saving rates between OECD countries is of 0.45 over the same period, as reported at the bottom of Table 2. Hence, the intra-national correlation levels that are found between investment and saving in China are much closer to other corresponding international figures than to the patterns of investment and saving intra-national movements. We also examine the variation of the correlation over time. The correlation increases sharply between the 1980s and the 1990s from 0.30 to 0.52 (column 4, Table 2). This result is counterintuitive as we would have expected the correlation to decrease - not to increase- over time along with the progress in economic and financial transition. As already mentioned, simple correlations based on raw data only provide a rough measure on financial integration, as a positive correlation between investment and saving rates can co-exist with a high degree of capital mobility. In the following section, we apply to the Chinese provinces the framework proposed by Iwamoto and van Wincoop (2000) that controls for the factors that can explain investment and saving co-movements even when economies financially integrated Testing for Investment-Saving Correlation in China Following FH (1980) finding of a close relationship between national saving and investment rates, a substantial literature has argued that such a positive relationship does not necessarily imply barriers to capital flows as factors exist that can explain saving and investment co-movements even when financial markets are perfectly integrated 5. Iwamoto and van Wincoop (2000) propose an original approach consisting of first controlling for those factors and, second, recalculating accordingly the correlation between saving and investment rates. Their idea is that if financial markets are integrated, investment depends on factors g that determine the global and national supply of funds and factors that determine the productivity of capital which are again global and national factors g and a set of region-specific factors. Saving in the same country also depends on 5 See for instance Coakley at al. (1998) for a review of the literature. 5

6 global and region-specific factors. Those region-specific factors that affect investment and saving can be organized as a set of uncorrelated vectors v, w, z, whereby v and w affect investment, and v and z affect saving. In other words, v is a vector of regionspecific factors that affect both investment and saving. w (z) is a vector of factors that affect investment (saving) independently. Therefore, investment and saving can be expressed respectively as the following functions: I = I(g,v,w) and S = S(g,v,z). In an environment of high capital mobility, investment and saving can still be correlated as a result of global and national factors g or region-specific factors v. After controlling for those factors, saving and investment should be uncorrelated. If the correlation is still significant, it provides a strong indication that financial integration is low. In order to evaluate the level of financial integration between the Chinese provinces, we follow step by step the methodology applied by Iwamoto and van Wincoop to the Japanese prefectures. First, we control for national shocks by subtracting national saving and investment rates from the province s saving and investment rates: S it * = S it S -it I it * = I it I -it (1a) (1b) Where S it and I it are respectively the saving and investment rates in province i, S -it is the national saving rate outside province i, and I -it is the national investment rate outside province i 6. Second, region-specific shocks are controlled by regressing saving and investment rates for each province separately on region-specific determinants that can affect saving and investment co-movements: S it * = D i +D y y it + D F F it + e S it (2a) I it * = E i +E y y it + E F F it + e I it (2b) where S* and I* are respectively region-specific investment and saving rates, after subtracting Chinese national saving and investment rates. 6 Some of the provinces constitute a non-trivial fraction of national saving and investment. We therefore use rest of the country instead of national aggregates in order to distorting the sample correlations upwards. 6

7 We include in the vector v of region-specific factors a measure of the business cycle y and a fiscal variable F. The provincial series of business cycles are generated by applying the Hodrick and Prescott (HP) filter to the GDP series 7. The fiscal variable is the ratio of government consumption to GDP. The residuals e S it and e I it from equations (2a) and (2b) respectively are used to compute the new correlation between saving and investment. Panel A of Table 3 reports the results. The average correlation between investment and saving drops from 0.49 to 0.30 after controlling for national shocks over the whole period. When region-specific factors are also controlled, the correlation decreases further to 0.15 but remains significantly positive. The increase in the correlation between the 1990s and the 1980s signaled in section 2.1 remains significant once the respective national components of local saving and investment rates are subtracted. While the correlation looses its significance over the 1980s when national and regional shocks are held constant, it remains significant over the 1990s with a value of We also control for the endogeneity of the regressors in the saving and investment regressions, by introducing the one-year lagged values of the region-specific variables in equations 2a and 2b. As reported in line 4 of Panel A, using lagged instead of contemporary values leads to very similar results. Overall, the finding of a positive and significant correlation after controlling for the factors that could explain saving and investment co-movements within China suggests that there are still barriers to intra-national capital mobility in this country. For comparison purposes, columns 5 and 6 report the results found by Iwamoto and van Wincoop for Japanese prefectures and OECD countries. For the Japanese prefectures, the correlation drops to 0.04 and 0.01 when respectively national and national and regional factors are held constant (column 5). By contrast, the respective figures for OECD countries average 0.44 and 0.28 and remain significantly positive (column 6). Hence, even when national and region-specific components are controlled, patterns of capital flows within China are closer to the patterns of international capital movements than that of other intra-national studies. Finally, we control for the amount of investment financed by foreign saving. By doing so we intend to measure the degree of capital mobility within the provinces as 7 We use HP(100) filtered log(province GDP) minus HP(100) filtered log(national GDP). 7

8 opposed to capital mobility between the provinces and the rest of the world. In a given province (i), domestic investment I i can either be financed by domestic saving S i, intranational saving from the other provinces S intra,it or foreign saving S F,i : I it = S it + S intra,it + S F,it (3a) In China, foreign direct investment has been the main component of foreign capital inflows, as foreign short term capital movements as well as long term borrowing have been restricted. For instance, in 2000, foreign direct investment accounted for 68% of total foreign capital inflows 8. If foreign saving is approximated by foreign direct investment (FDI it ), equation (3a) can be re-written: I it = S it + S intra,it + FDI it with FDI it S F,it (3b) By subtracting foreign direct investment on both sides, we obtain the following expression: I it - FDI it = S it + S intra,it (3c) The left-hand side expression of equation (3c) corresponds to the amount of investment financed by domestic sources, which can be either local saving generated in the province itself (S it ) or saving coming from the other provinces (S intra,it ). Intuitively, if capital mobility between provinces is low, the correlation between (I it - FDI it ) and S it should be low. Alternatively, if capital flows easily from one province to another, then the amount of investment domestically financed and local saving should not be correlated. To test for inter-provincial capital mobility, we thus subtract from total investment the amount of foreign direct investment 9. The new investment figures are used in turn to re-compute the investment-saving correlations by using the same methodology as above, e.g. by controlling successively for national and regional factors. 8 Source: China Statistical Yearbook (2000). 8

9 Panel B of Table 3 reports the results. Over the whole period, the correlation averages 0.43 which is only slightly lower than the correlation of 0.49 using total investment. By contrast, the correlation drops from 0.30 and 0.15 to 0.21 and when national shocks and national and region-specific shocks are respectively controlled (lines 2 and 3 respectively, Panel B). Interestingly, the change in the correlation between the 1980s and the 1990s is not significant any more (line 3, column 6, Panel Be). This suggests that the increase in the correlation previously found was due to the increase in foreign direct investment inflows in the 1990s. One interpretation is the following: foreign saving has become an important source of finance for investment in China over the 1990s, on average of 8.1% of total investment financing and up to 12% in the mid-1990s. However, the geographical allocation of foreign direct investment has been concentrated on the coastal part of China, leaving the western and central regions aside of the process 10. Hence the very localized nature of foreign direct investment flows may have contributed to the increase in the internal fragmentation of capital markets in the country over the 1990s as some provinces have benefited from an easy access to foreign capital while the other have been severely financially constrained. In the next section, we examine the extent to which capital markets allow risk-sharing in consumption. in China. This will provide another evidence of the extent of financial integration of the Chinese provinces. 9 Provincial FDI series are available from 1984 onwards. As in 1984 foreign investment inflows were negligible (0.9% of GDP in 1984) they are assumed to be zero. 10 Over the 1990s, the average share of investment financed by foreign is as high as 17.7 and 16.6% in Fujian and Guangdong provinces respectively. By contrast, the corresponding shares in Ningxia, Yunnan and Qinghai provinces equal 2.4%, 2.8% and 2.9% respectively. 9

10 2. Risk-sharing within China According to the risk sharing theory, in a fully integrated world, individuals can, by pooling their risk, insure against uncertainty in their resources. Under the assumption of asset market completeness and perfect capital mobility, household changes in consumption should move one for one with aggregate changes in consumption absent of idiosyncratic fluctuations of preferences or measurement error. Furthermore, differences in consumption changes across households should not be correlated with changes in household resources. However and by contrast with the theoretical predictions of the risksharing model, the empirical literature on international consumption smoothing finds the opposite result: consumption growth rates are not highly correlated with each other and are less correlated than output growth rates 11. Despite the lack of conclusive evidence of the theoretical predictions of the model, risk-sharing empirical tests still provide useful information about the degree of financial market integration for the following reasons: First, given that assumptions of perfect capital mobility and market completeness are odds with reality, one would not expect to find a perfect coherence in the evolution of per capita consumption across countries but rather a weaker correlation. A more reasonable result would be to observe, in the presence of increasing international capital mobility and financial sophistication, an increasing coherence of consumption fluctuations and a decreasing volatility of consumption 12. Second, consumption is more correlated within OECD economies, where we already know that the level of financial integration is high, than between them 13. The general conclusion of the intra-national literature is that intranational evidence for risk-sharing, although not perfect, is greater than that for international evidence. Hence those intra-national studies provide a reasonable benchmark for evaluating the results found on other intra-national samples. Finally, consumption-based tests that do not assume asset market completeness provide a more 11 For a survey of the literature on risk-sharing see Crucini and Hess (2000). 12 For instance, Obstfeld (1994) notes that among the G-7 countries over the period 1951 to 1988 there has been a tendency for domestic consumption to become more closely correlated with the world consumption 13 The intra-national risk-sharing literature has used regional data within Canada (Crucini (1998) and Bayoumi and McDonald (1995)), Japan (van Wincoop (1995)) and the United States (Atkeson and Bayoumi (1993), Crucini (1998), Hess and Shin (1997), Asdrubali, Sorensen and Yosha (1996) and Sorensen and Yosha (1996)). 10

11 realistic framework for assessing the degree of financial integration on the basis of risksharing theoretical predictions. This section is organized as follows: we first present descriptive statistics in order to characterize consumption smoothing in the Chinese provinces. We apply thereafter the framework proposed by Obstfeld (1994) that allows testing for risk-sharing by assuming either a complete or an incomplete asset market Descriptive statistics Table 4 shows the correlations of provincial per capita consumption growth with the rest of the country per capita consumption growth rate and the correlation of per capita GDP growth with the rest of the country per capita GDP growth rate 14. Consumption and GDP series come from the 2001 A11 China Marketing Research database for 28 Chinese provinces (All Chinese provinces but Tibet because of data unavailability). All the data are expressed at 1978 prices and in log-difference. We proceed as in the previous section: correlation in the Chinese provinces are averaged over the reform period ( ) and two sub-periods ( and ). For comparison purpose the corresponding figures for OECD countries are also reported. Consumption correlations in the Chinese provinces average 0.53 over the whole period. In comparison, the corresponding figure for OECD countries is Intranational correlations of consumption found in the empirical literature on risk-sharing are shown in Table 5. Local per capita consumption growth rates are more correlated within China than within OECD economies such as USA or Japan where financial markets are integrated. Thus and quite surprisingly, per capita consumption growth rates are more correlated within China than in any other international or intra-national sample of OECD countries where we know that financial integration is high. As a consequence, this comparison suggest a high degree of risk-sharing within China. A high level of consumption correlation can however also reflects a close association of local output growth rates coupled with the typical high correlation between 14 Some of the provinces constitute a non-trivial fraction of national output. We therefore use rest of the country instead of national aggregates in order to avoid distorting the sample correlations upwards. 11

12 consumption and output growth rates. We thus need to control for corresponding GDP co-movements by dividing the correlation of local consumption growth rates by the correlation of local output growth rates (line 3, Table 4). Over the whole period, the ratio stands significantly below unity, as local consumption growth rates are less correlated than GDP. In comparison, the corresponding ratio for OECD economies is not significantly different from one. Hence the high correlation of per capita consumption between the Chinese provinces is better explained by a high level of correlation between regional outputs than by a high level of regional risk-sharing. Column 4 reports the difference-in-mean test of any significant variation of the correlations between the 1980s and the 1990s. The consumption correlation declines markedly from 0.69 to 0.19 while output correlation increases. Those two opposite movements lead to a sharp drop in the ratio of consumption to output correlation, from 1.29 to 0.3. In comparison, the corresponding OECD figure decreases but to a much lesser extent, from 1.16 to The weakening of the consumption correlation over the 1990s can be paralleled with the tightening of the correlation between saving and investment found in the previous section. Both findings suggest a decrease in the degree of financial integration in the recent period. Table 6 reports the volatilities of per capita consumption and output growth rates. If the degree of financial integration is high, individuals should be able to smooth their consumption over time. In other words, as a result of inter-temporal risk-sharing or permanent-income hypothesis, one can expect consumption volatility to be lower than output volatility. In the Chinese provinces, consumption is more volatile than output. Over the period, output and consumption time-series standard deviations average 5.24 and 3.99 respectively. Table 6 also reports consumption and GDP volatilities of OECD countries. Consumption and output in the Chinese provinces appear more volatile than the corresponding OECD national figures. Local consumption is also more volatile within China than within Japan and USA. Hence the level of inter-temporal risk-sharing appears low within China. Overall, there is limited statistical evidence of risk-sharing in the Chinese provinces, considering consumption correlations or volatilities and relatively to international or intra-national standards. 12

13 Measuring capital mobility by consumption correlations or related measures does not allow to give conclusive evidence on the degree of capital mobility, as a low degree of risksharing can also come from asset market incompleteness. In the next section, we apply the framework proposed by Obstfeld (1994) for testing risk-sharing which explicitly recognizes that co-variation in per capita consumption depends not only of the degree of financial integration of capital markets but also on the completeness of those markets Testing for Risk-Sharing in the Chinese provinces Obstfeld (1994) develops a model of international consumption risk-sharing from which are derived several alternative tests. The model allows testing the two extreme cases of perfect and no capital mobility under the complete market hypothesis as well as the case for the broad middle ground where asset markets are incomplete. In this section, we apply Obstfeld s empirical methodology to evaluate the extent of risk-sharing in China 15. The first test supposes a complete asset-market environment and consists of running the following regression: dln(c it )= a + b.dln(c -it ) + u i. + e it (4a) where C it is per capita private consumption in real terms, C -it is aggregate per capita private consumption outside province i, u i. is the provincial fixed effect and e it is the error term 16. Under the null hypothesis of perfect risk-sharing, the coefficient estimate of aggregate consumption b should not be significantly different from unity. Tables 7a reports the results of the panel estimation. The hypothesis of perfect risksharing over the reform period is rejected, as shown by the consumption coefficient of 0.88 which is significantly different from one 17. Splitting the reform period reveals that the coefficient of aggregate consumption decreases sharply in the 1990s. In the first sub- 15 As we do not aim to discuss the model but simply to apply Obstfeld s framework to our sample of Chinese provinces, we refer to the original article for further theoretical readings 16 Because they are collinear with the aggregate consumption variable, fixed time effects are not introduced in equation 4a. The series of per capita consumption growth rate is integrated of order one. 13

14 period, the coefficient is not significantly different from one. By contrast, the estimate drops to 0.56 over the 1990s suggesting a decrease in risk-sharing in the 1990s 18. The validity of those results depend however critically on the validity of the assumption of asset-market completeness, which is questionable in the case of China s emerging capital markets. We now turn to the second test proposed by Obstfeld, that relaxes the assumption of market completeness. It consists of estimating the following equation: dln(c it )= a + b.dln(c -it ) + c.ln(drl it ) + u i + e it (4b) where DRL it is the Domestic Resource Limit defined as DRL it = GDP it - I it G it which corresponds to the maximum consumption level when international markets are closed. If domestic investment is constrained by domestic saving, then consumption is constrained by the domestic resource limit and the hypothesis b=0 and c= 1 should not be rejected. Alternatively, if financial integration is high, we would expect b=1 and c=0 to be true. Intuitively, equation (4b) gives an indication of the extent to which consumption is more closely related to global or local factors respectively represented by aggregate consumption C -it or Domestic Resource Limit DRL it. As emphasized by Obstfeld, this framework is closely related to the investment-saving test for capital mobility. Table 7b shows reports the estimate of equation 4b and the test statistics of the alternative hypotheses. The null hypothesis of a low degree of financial integration is rejected for all periods (H3: b=1 & c=0). At the opposite extreme, the hypothesis of a high financial integration is also systematically rejected (H4: b=0 & c=1). Similarly, the coefficients of dln(drl it ) and dln(c -it ) are individually significantly lower than one and higher than zero. Overall, the results do not allow to discriminate between the two extreme bounds of perfect and no risk-sharing. They rather suggest a middle range financial integration between the Chinese provinces when incomplete markets are 17 When significantly different from one, the coefficients are reported in bold. 18 Equation 4 can be further refined by adding per capita world output growth among the regressors (see Obstfeld 1994). This is in response to the concern that the estimate of equation 4 may only reflect the fact that domestic consumption growth and domestic output growth on the one hand and domestic output growth and world output growth one the other hand are typically highly correlated. Adding aggregate output growth among the regressors produce however similar results (available upon request). 14

15 assumed. Over the 1990s, the coefficient estimate of consumption decreases which, coupled with the finding of a higher coefficient estimate of domestic resource limit suggests again a weakening of risk-sharing in the recent period. Finally, we examine the role of foreign direct investment and more generally by openness in risk-sharing. Intuitively controlling for international openness allows to distinguish the channels of risk-sharing through international capital markets from that through inter-provincial capital markets. As we are interested in the impact of international integration on risk-sharing in the provinces, the coefficient of aggregate consumption should vary according to the level of foreign direct investment. Fdi is therefore introduced as an additive and multiplicative variable as follows: dln(c it )= a + b 0.dln(C -it ) + b 1.ln(C -it ).Fdi it + e.fdi it + u i + e it (5a) dln(c it )= a + b 0.dln(C -it ) + b 1.ln(C -it ).Fdi it + e.fdi it +c 0.dln(DRL -it ) + c 1.ln(DRL -it ).Fdi it + u i + e it (5b) Table 8 reports the estimate of equations 5a and 5b over the whole period and over the two sub-periods. The first three columns report the results when complete asset markets are assumed (equation 5a). Foreign direct investment has a positive impact on the per capita consumption growth rate. This result is not surprising, as foreign investment has been identified in China as an important engine of real growth. More importantly for our purpose, the coefficient of the interaction term between foreign direct investment and aggregate consumption is negative and significant, suggesting that an increased availability of foreign capital lowers the degree of risk-sharing in the provinces. Splitting the sample into two sub-periods reveals that the significance of FDI is true only over the recent period, when foreign direct investment has become an important source of finance for investment. Columns 4 to 6 show the results when incomplete markets are assumed (equation 5b). As previously found, FDI acts positively on consumption growth and negatively on risksharing as shown by the interaction term. The interaction term involving the Domestic Resource Limit variable is insignificant. Again, the impact of foreign investment is 15

16 insignificant over the 1980s and positive over the 1990s. More interestingly, once we control for the impact of foreign capital inflows, the hypothesis of a high level of financial integration cannot be rejected as reported at the bottom of Table 8 (H3: b=1 & c=0). Hence, once the impact of FDI is neutralized, the degree of financial integration increases. This suggests that foreign investment inflows in the 1990s have caused an increase in the fragmentation of regional capital markets. Furthermore we test whether controlling for FDI stabilizes the estimates over the reform period by applying Chow tests that compare the estimate using the whole sample to the estimates over the 1990s. The test-statistics for equations 5a and 5b are respectively F(323, 275) = 0.80 and F(323, 277) = 0.79 and are not significant. We thus do not reject the hypothesis of the global stability of the estimate. Finally, Table 8 reports the results when the ratio of exports plus imports to GDP (Open) is used as an alternative indicator of international integration. By contrast with foreign direct investment, trade openness never proves significant. To summarize, the surge of foreign capital inflows in the 1990s explains relatively well the drop in the coefficient of aggregate consumption previously found. In other words, the increasing international financial integration of the some of the Chinese provinces could be responsible for the decrease in risk-sharing in the Chinese provinces over the 1990s. 3. Concluding Remarks In this paper, we have examined the extent of financial integration of the Chinese provinces on the basis of tests of investment-saving correlation and consumption smoothing. Simple descriptive statistics first suggest that capital is much less mobile within China than within tightly financially integrated countries like Japan or the United States. Overall, patterns of capital flows within China appear closer to the patterns of international capital movements than that of other intra-national studies. We also find that capital mobility in China significantly decreased in the 1990s. When we control for the degree of international integration of the provinces proxied by foreign investment, this result is not significant any more. Hence the decrease in capital mobility can be 16

17 interpreted as a consequence of the surge of foreign direct investment over the 1990s, which may have intensified disparities in investment financing between provinces and so further fragmented regional capital markets. References - Asdrubali, P., Sorensen, B. E., and O. Yosha, 1996, Channels of Interstate Risk Sharing: United States Quarterly Journal of Economics 111, 4, Bayoumi T. A. and A. K. Rose, 1993, Domestic savings and intra-national capital flows, European Economic Review, 37, , - Bayoumi T., and R. McDonald, 1995, Consumption, Income and International Capital Market Integration, IMF Staff Papers, 42, 3, , - Bayoumi T., and M. W. Klein, 1997, A Provincial View of Economic Integration, IMF Staff Papers, 44, 4, , - Coakley J., Kulai F. and R. Smith, 1998, The Feldstein-Horioka Puzzle and Capital Mobility: A Review, International Journal of Finance and Economics, 3, Crucini, M., 1999, On International and National Dimensions of Risk Sharing, Review of Economics and Statistics, 81, 1, 73-84, - Crucini, M., and G. D. Hess, 2000, International and Intranational Risk Sharing, in Gregory D. Hess and Eric van Wincoop (eds.), Intranational Macroeconomics, Cambridge University Press, NY. - Dekle R., 1996, Saving-Investment associations and capital mobility: on the evidence from Japanese data, Journal of International Economics, 41, Feldstein M. and C. Horioka, 1980, Domestic saving and international capital flows, The Economic Journal, 90, , - Frankel J. A., 1992, Measuring international capital mobility: A review, American Economic Review, Papers and Proceedings, 82, , - Hess G., and K. Shin, 1997, International and Intranational Business Cycles, Oxford Review of Economic Policy, 13, 3, , - HoTsung-wu, 2002, A panel cointegration approach to the investment saving correlation, Empirical Journal of Economics, forthcoming, - Iwamoto Y. and E. van Wincoop, 2000, Do border matter? Evidence from Japanese regional net capital inflows, International Economic Review, 41, 1, , - Obstfeld, M., 1994, Are Industrial-Country Consumption Risks Globally Diversified? In Leonardo Leiderman and Assaf Razin (eds.), Capital Mobility: the Impact on Consumption, Investment, and Growth, Cambridge University Press, Cambridge. - Obstfeld, M., 1995, International Capital Mobility in the 1990s, in Peter B.Kenen (ed.), Understanding Interdependence: The Macroeconomics of the Open Economy, Princeton University Press. - Pakko, M.R., 1996, Characterizing Cross-Country Consumption Correlations, Review of Economics and Statistics,

18 - Poncet S., 2002, Is China Disintegrating? The Magnitude of Chinese Provinces' Domestic and International Integration, 2001, CERDI Etudes et Documents, 5, - Qian Y. and C. Xu, 1993, «Why China s Economic Reform Differ : the M-form Hierarchy and Entry/Expansion of the Non-State Sector, The Economics of Transition, 1, 2, , - Sinn S., 1992, Saving-investment correlations and capital mobility: on the evidence from annual data, The Economic Journal, 102, , - Thomas A. H., 1993, Saving, investment and the regional current-account: an analysis of Canadian, British, and German regions, IMF Working Paper, 93/62, - van Wincoop E., 1995, Regional Risk Sharing, European Economic Review, 37, , - World Bank, 1994, China: Internal Market Development and Regulation, World Bank Country Study, The World Bank, 248 p., - Yamori N., 1995, The relationship between domestic savings and investment: the Feldstein-Horioka test using Japanese regional data, Economic Letters, 48, , Statistical Sources: - State Statistical Bureau, various years, Almanac of China Foreign Relations and Trade, China Economics Publishing House, - State Statistical Bureau, various years, China Statistical Yearbook, China Statistical Bureau, Beijing, - State Statistical Bureau, 1996, China Regional Economy, a Profile of 17 years of Reform and Opening-up, China Statistical Bureau, Beijing - A11 China Marketing Research, 2001, China Statistical Data Compilation. 18

19 Table 1: Investment-Saving Correlation Intra-national Evidence Country [time period] Author Saving coefficient a Japan [ ] Deckle (1996) [-0.24 to -0.44] Japan [ ] Yamori (1995) [-0.26 to -0.36] Japan [ ] Bayoumi and Rose (1993) [-0.48,0.24, 0.01] USA [ ] Sinn (1992) [-0.11] b UK [ ] Thomas (1993) [-0.56] Canada [ ] [-0.11] Germany [ ] [-0.06] c a coefficient of the saving rate in the investment regression b correlation c private saving and investment 19

20 Table 2: Investment-Saving Correlation - Chinese Provinces and OECD Countries (1) (2) (3) (4)=(3)-(2) China diff. (1) Investment-saving correlation a (0.103) (0.102) (0.091) (0.069) *** OECD diff. (2) Investment-saving correlation a (0.069) (0.099) (0.094) (0.138) a average time correlation between investment and saving rates (% GDP) *(**, ***) significant at the 10% (5%, 1%) level. Standard errors in parentheses Source (OECD) : UN National Accounts, World Development Indicators countries: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy, Japan, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, United States. Source (China): A11 China Marketing Research, 2001, China Statistical Data Compilation, China Statistical Bureau, Beijing. China Statistical Yearbook, various years, China Statistical Bureau, Beijing. Provinces: Beijing, Tianjin, Hebei, Shanxi, Nei Monggol, Liaoning, Jilin, Heilongjiang, Shanghai, Jiangsu, Zhejiang, Anhui, Fujian, Jiangxi, Shandong, Henan, Hubei, Hunan, Guangxi, Guizhou, Yunnan, Shaanxi, Gansu, Qinghai, Ningxia, Xinjiang, Guangdong + Hainan, Sichuan + Chongqing,

21 Table 3: Investment-Saving Correlations (Iwamoto and van Wincoop, 2000) Panel A (1) (2) (3) (4)=(3)-(2) (5) (6) diff Japan OECD (1) Raw Data (0.103) ** (0.102) ** (0.091) ** (0.069) *** (0.04) *** (0.04) *** Controlling for (2) national factors (0.105) ** (0.1056) (0.093) ** (0.066) *** (0.04) (0.07) *** (3) national & regional factors & business cycle & fiscal policy (0.098) ** (0.102) (0.086) ** (0.067) *** (0.04) (0.05) *** Using one-year lagged regional variables (4) & business cycle & fiscal policy (one year lagged) (0.094) ** (0.098) (0.083) ** (0.069) *** Panel B (1) Substracting FDI from investment (0.096) ** (0.097) (0.100) ** (0.118) ** (2) Controlling for national factors (0.110) ** (0.102) (0.097) ** (0.093) * (3) national & regional factors & business cycle & fiscal policy (0.094) (0.097) (0.086) (0.097) *(**, ***) significant at the 10% (5%, 1%) level. Standard errors in parentheses Source: own calculations (columns 7 and 8, Iwamoto and van Wincoop, 2000) 21

22 Table 4: Consumption Correlations Chinese Provinces and OECD Countries (1) (2) (3) (4)=(3)-(2) China diff. (1) Correlation of consumption *** (0.027) (0.030) (0.051) (0.057) (2) Correlation of output *** (0.029) (0.038) (0.039) (0..048) (3) Ratio of consumption *** =(1)/(2) to output correlations (0.080) (0.111) (0.149) (0.200) OECD diff. (1) Correlation of consumption (0.047) (0.064) (0.057) (0.089) (2) Correlation of output (0.050) (0.069) (0.066) (0..092) (3) Ratio of consumption * =(1)/(2) to output correlations (0.065) (0.085) (1.168) (0.089) correlation of per capita provincial (national) consumption and the rest of the country (of the world) per capita consumption. Per capita consumption and GDP are expressed in real terms and by taking the first difference of the logarithm *(**, ***) significant at the 10% (5%, 1%) level. Standard errors in parentheses in bold: significantly different from 1 at the 10 % (or lower) level Source (OECD) : UN National Accounts, World Development Indicators Source (China): A11 China Marketing Research, 2001, China Statistical Data Compilation, China Statistical Bureau, Beijing. China Statistical Yearbook, various years, China Statistical Bureau, Beijing 22

23 Table 5: Consumption Smoothing-Intranational Evidence Country [# regions] Author Cross-correlation Volatility Cross-correlation Volatility Time period Consumption Consumption Output Output Japan [47] van Wincoop (1995) (0.36) (nr) (0.31) (nr) Canada [10] Crucini (1995) 0.56 nr 0.38 nr (0.17) (nr) (0.31) (nr) USA [51] Crucini (1995) 0.40 nr 0.50 nr (0.22) (nr) (0.21) (nr) USA [19] Hess and Shin (1995) (0.41) (1.21) (0.43) (0.93) nr: not reported Standard errors in parentheses source: Hess and Shin (1997) 23

24 Table 6: Consumption Volatilities Chinese Provinces and OECD Countries (1) (2) (3) (4)=(3)-(2) China diff. (1) Volatility of consumption *** (0.191) (0.22) (0.32) (0.39) (2) Volatility of output *** (0.15) (0.23) (0.23) (0.34) (3) Ratio of consumption =(1)/(2) to output volatilities OECD diff. (1) Volatility of consumption ** (0.183) (0.233) (0.171) (0.212) (2) Volatility of output (0.117) (0.129) (0.157) (0.185) (3) Ratio of consumption =(1)/(2) to output volatilities (0.138) (0.142) (2.12) (2.20) Volatility: time-series standard deviation of per capita consumption (GDP) for each province (country) Per capita consumption and GDP are expressed in real terms and by taking the first difference of the logarithm *(**, ***) significant at the 10% (5%, 1%) level. Standard errors in parentheses in bold: significantly different from 1 at the 10 % (or lower) level Source (OECD) : UN National Accounts, World Development Indicators Source (China): A11 China Marketing Research, 2001, China Statistical Data Compilation, China Statistical Bureau, Beijing. China Statistical Yearbook, various years, China Statistical Bureau, Beijing 24

25 Table 7a : Risk-Sharing - Complete Asset market (Obstfeld (1994)) Dependent variable: logarithm of per capita household real consumption (variation) Period dln(c-it) *** *** ** (0.053) (0.055) (0.234) adj. R # obs # provinces H1: b=1 F(1,601)=5.02 ** F(1,294)=2.54 F(1,279)=3.51 * C-it: national per capita household real consumption outside the province i ln: logarythm, d: first difference in bold: significantly different from 1 (5% level or below) *(**, ***) significant at the 10% (5%, 1%) level. all the regressions include fixed provincial effects. Robust standard errors in parentheses 25

26 Table 7b : Risk-Sharing - Incomplete asset market (Obstfeld (1994)) Dependent variable: logarithm of per capita household real consumption (variation) Period dln(c-it) *** 0.88 *** ** (0.049) (0.051) (0.224) dln(drlt) *** *** *** (0.016) (0.012) (0.035) adj. r # obs # provinces H1: b=1 F(1,601)=10.78 *** F(1,293)=5.52 ** F(1,278)=5.58 ** H2: c=1 F(1,601)=3231 *** F(1,293)=5465 *** F(1,278)=585.7 *** H3: b=1 & c=0 F(2,601)=14.44 *** F(2,293)=12.79 *** F(2,278)=9.09 *** H4: b=0 & c=1 F(2,601)=1649 *** F(2,293)=2820 *** F(2,278)=294.1 *** C-it: national per capita household real consumption outside the province i Yit: per capita household real GDP DRLit= Yit - Iit - Git (Domestic Ressource Limit) Iit: provincial per capita real investment Git: provincial per capita government real consumption ln: logarithm, d: first difference in bold: significantly different from 1 (5% level or below) *(**, ***) significant at the 10% (5%, 1%) level. all the regressions include fixed provincial effects. Robust standard errors in parentheses 26

27 Table 8: Risk-Sharing and Foreign Direct Investment Dependent variable: logarithm of per capita household real consumption (variation) Period dln(c-it) *** *** *** *** *** *** (0.057) (0.061) (0.284) (0.052) (0.056) (0.273) dln(drlt) *** *** 0.13 *** (0.015) (0.013) (0.04) dln(c-it)*fdi ** ** *** ** (3.93) (5.83) (5.17) (3.52) (6.09) (4.71) dln(drlt)*fdiit (0.650) (0.650) (0.81) FDIit ** *** *** *** (0.255) (0.518) (0.362) (0.520) (0.34) adj. r # obs # provinces H1: b=1 F(1,600)=1.20 F(1,292)=2.32 F(1,277)=0.39 F(1,598)=3.86 ** F(1,290)=4.59 ** F(1,275)=1.22 H2: c=1 F(1,598)=3776 *** F(1,290)=4864 *** F(1,275)=387 *** H3: b=1 & c=0 F(2,598)=12 *** F(2,290)=11.75 *** F(2,275)=5.24 H4: b=0 & c=1 F(2,598)=1950 *** F(2,290)=4864 *** F(2,275)=194 *** C-it: national per capita household real consumption outside the province i ; Yit: per capita household real GDP ; DRLit= Yit - Iit - Git (Domestic Ressource Limit) with Iit: provincial per capita real investment and Git: provincial per capita government real consumption ; FDIit: Foreign direct invesment as a proportion of GDPi (assumption: FDI= 0 before 1984 fin all the provinces) in bold: significantly different from 1 (5% level or below) *(**, ***) significant at the 10% (5%, 1%) level. all the regressions include fixed provincial effects. Robust standard errors in parentheses. 27