- Email: [email protected]
Saving, investment and international capital mobility in East Asia
Japan and the World Economy 19 (2007) 279–291 www.elsevier.com/locate/econbase
Saving, investment and international capital mobility in East Asia§ Soyoung Kim a, Sunghyun H. Kim b,*, Yunjong Wang c a
Department of Economics, Korea University, 5-1 Anam-Dong, Sungbuk-Ku, Seoul 136-701, Republic of Korea b Department of Economics, Tufts University, Medford, MA 02155, United States c SK Research Institute, Seoul Finance Center, 84 Taepyungro 1-ga, Chung-gu, Seoul 100-101, Republic of Korea Received 17 November 2005; received in revised form 9 May 2006; accepted 12 May 2006
Abstract This paper estimates the degree of international capital mobility in East Asia using the saving– investment correlation originated in Feldstein and Horioka [Feldstein, M., Horioka, C., 1980. Domestic saving and international capital flows. Economic Journal 90, 314–329]. We apply the empirical method used in Kim [Kim, S.H., 2001. The saving–investment correlation puzzle is still a puzzle. Journal of International Money and Finance 20, 1017–1034] to control for cyclical effects in estimating a time-series saving– investment correlation of 10 Asian countries from 1980 to 2002. Our conclusion is that the saving– investment correlation in East Asia steadily decreases over time but is still higher than that of the OECD countries over all studied periods. These results are consistent with the fact that capital mobility in East Asia is lower than that in the OECD countries. In addition, regional saving and investment data demonstrate that investment in East Asia is largely financed by regional savings. # 2006 Elsevier B.V. All rights reserved. JEL classification: F32; F41 Keywords: Saving–investment correlation; Feldstein–Horioka puzzle; Capital mobility; East Asia; Business cycles
1. Introduction During the 1990s, a number of East and Southeast Asian countries liberalized their financial markets to foreign capital by reducing restrictions on inward and outward capital flows. As a result, net private capital inflows to this region in the mid-1990s were conspicuous in the postwar
§
An earlier version of this paper was published as KIEP Working Paper No. 04-01. * Corresponding author. Tel.: +1 617 627 3662; fax: +1 617 627 3917. E-mail address: [email protected] (S.H. Kim).
0922-1425/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.japwor.2006.05.001
280
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
period in terms of the size of the flow to emerging markets. This paper attempts to measure the degree of international capital mobility in East Asia using a saving–investment correlation (S–I correlation hereafter). Since the seminal paper by Feldstein and Horioka (1980, denoted by FH hereafter), numerous economists have attempted to quantify the degree of international capital mobility using both a cross-sectional and a time-series S–I correlation. Under perfect capital mobility, national saving and national investment should not be correlated as domestic agents search for worldwide investment and saving opportunities regardless of the nationality of capital. Therefore, a high correlation between national saving and investment would imply a low degree of capital mobility. However, since many factors other than capital mobility can simultaneously affect saving and investment, it is difficult to interpret the S–I correlation as a correct measure for the degree of capital mobility.1 This paper adopts the empirical method in Kim (2001) to estimate the time-series S–I correlation after controlling for business cycle shocks, to provide an accurate measure of capital mobility. We remove the cyclical effects from saving and investment data by using the residuals from the regressions of saving and investment on cyclical shocks including productivity shock, fiscal shock and the terms of trade (TOT) shock. In this paper, we test whether this newly estimated S–I correlation correctly reflects the liberalization of capital markets in East Asia. In particular, we compare the S–I correlation in East Asia with that of the OECD countries to examine the differences in the degree of capital mobility.2 We also investigate whether country differences such as the size of the GDP and the non-traded sector have any systematic relationship with S–I correlation. Finally, we use regional saving and investment data to investigate how much of the investment in East Asia is financed by regional savings. The data used originates from panel data of saving and investment from 10 East Asian countries between 1980 and 2002. For panel data regression, we use the GLS estimation of the SUR system to control for cross-country heteroscedasticity and cross-country correlation in the error term. The estimation results illustrate that the cross-sectional S–I correlation steadily decreases over time (0.76 in the early 1980s to 0.53 in the late 1990s) but remains high in absolute terms. This result follows previous papers, including the original FH study. Panel data regression shows that the time-series S–I correlation in East Asia is approximately 0.88 and slightly decreases after controlling for cyclical shocks. These estimates are higher than those of OECD countries (0.70 before controlling cyclical shocks), which is consistent with the fact that capital mobility in East Asia is lower than that among the OECD countries. Country differences do not significantly impact the S–I correlation. Finally, aggregate regional data of saving and investment show that investment in East Asia is largely financed by regional savings. The following is a layout of what follows in this paper. In Section 2, we provide an overview of prior empirical studies on the S–I correlation. Section 3 explains the empirical methodology that we adopt in this paper. Section 4 reports the main statistical properties of the saving and investment data and estimation results including cross-sectional and time-series panel regressions. Finally, Section 5 offers our conclusion.
1
This endogeneity problem has been a main econometric concern since the original paper by Feldstein and Horioka (1980). They assumed that population structure is the main source of endogeneity problem in cross-sectional regressions, while business cycle shocks are the main source of endogeneity problem in time-series regressions. 2 Estimation results for the OECD countries are taken from Kim (2001).
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
281
Box 1. Summary of empirical studies on S–I correlation Paper
Data
Estimation results
Cross-sectional analysis Feldstein and Horioka (1980) Feldstein (1983) Murphy (1984) Dooley et al. (1987) Bayoumi (1989) Golub (1990) Tesar (1991)
16 OECD countries, 1960–1974, S,I as shares of GDP 17 OECD countries, 1960–1979 17 OECD countries, 1960–1980 14 OECD countries, 1960–1980 10 OECD countries, 1965–1986 16 OECD countries, 1960–1988 16–23 OECD countries, 1960–1986
0.89 (gross measures), 0.94 (net measures) 0.80 0.85 1960–1973: 0.75, 1974–1980: 0.74 OLS: 0.97, two-stage LS: 0.85 0.76 1960–1986: 0.84, 1960–1974: 0.89 1975–1986: 0.81 1960–1986: 0.83, 1960–1969: 0.85, 1970–1979: 0.67, 1980–1986: 0.87 1960–1969: 0.87, 1970–1979: 0.82, 1980–1989: 0.68 1974–1990: 0.72, 1974–1980: 0.87, 1981–1990: 0.64
Feldstein and Bacchetta (1991) Sinn (1992)
23 OECD countries, 1960–88
Obstfeld (1995)
20 OECD countries, 1974–1990
Time-series analysis Bayoumi (1989)
23 OECD countries, 1960–1986
1961–1986
Tesar (1993)
1960–1988 OLS, S,I as ratios of GDP
Obstfeld (1995)
1974–1990 OLS, S,I as ratios of GDP
Kim (2001)
21 OECD countries, 1960–1992, panel (GLS) with first differenced data
US: 1.00, Japan: 0.84, Germany: 0.87, UK: 0.33, France: 0.80, Canada: 0.83, Finland: 0.98, Norway: 0.21 US: 0.752, Canada: 0.848, France: 0.929, Germany: 0.886, Italy: 0.063, UK: 0.592 US: 0.848, UK: 0.113, Japan: 1.161, Italy: 0.214, France: 0.909, Germany: 0.327 0.70
2. Overview of literature A number of papers have empirically estimated the cross-sectional and time-series correlation of saving and investment. Box 1 summarizes the main results of these papers.3 Most crosssectional regressions reveal that the S–I correlation was high even in the 1990s, while time-series estimation results depend too much on individual countries to generate any general conclusion. Since the late 1980s, there has been an increasing number of simulation studies based on real business cycle models that explain a high S–I correlation using different types of shocks, even under perfect capital mobility.4 Although numerous papers have suggested multiple shocks and economic structures to explain the high S–I correlation, many have been limited to simulation or vector autoregression methods and have not offered strong empirical support. Moreover, their results depend on the specification of the model, particularly the structure of investment and the specification of shocks.
3
Refer to Coakley et al. (1998) for a detailed survey on this literature. See for example, Mendoza (1991), Backus et al. (1995), Baxter and Crucini (1993), Taylor (1994) and Glick and Rogoff (1995). 4
282
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
3. Empirical methodology Most earlier empirical papers applied a cross-sectional analysis with time-averaged data and related the S–I correlation to long-term capital movement. Such an application averaged out high-frequency fluctuations in the data. However, a cross-sectional analysis is not appropriate in examining Asian countries since the capital market liberalization process was quite rapid and concentrated in a short period of time, and since the timing of liberalization was different among countries. Time-averaging removes these characteristics from the data. Therefore, a time-series analysis is more appropriate, as it reflects short-term capital movements over time. A time-series analysis does pose several econometric problems. First, S–I correlations from country-by-country time-series regressions are insignificant in many cases and too different across countries to identify any systematic relationship with capital mobility. Second, both saving and investment data are subject to cyclical shocks, which can create an endogeneity problem in which regressors are correlated with error terms. Third, time-series data of saving and investment are usually non-stationary, which can result in a spurious regression. We overcome these problems by adopting a panel data analysis and by controlling business cycle effects caused by exogenous shocks. The use of a panel data set provides a more efficient estimator compared to the case of a single time-series estimation. It is difficult to reflect a crosscountry correlation in the cross-sectional regression and without considering cross-country correlation, the inference on coefficient estimates may not be consistent. The panel data regression allows us to reflect a cross-country correlation and therefore avoid this inconsistency problem. We are also able to specifically incorporate country-specific effects in the panel regression. In order to remove cyclical properties from the data, we run the regressions of saving and investment on business cycle shocks and use the residuals from these regressions to derive the S–I correlation. We avoid the spurious regression problem by using the first differenced data.5 For estimation of the panel data system, we apply the GLS estimation calculated by iterating the SUR system using newly computed covariances and system equation estimates. Though we set a maximum of 20 iterations, all of the results converge before reaching this limit. We test if there is cross-country heteroscedasticity and a cross-country correlation in the error term using the Lagrange Multiplier tests. Test statistics indicate that strong evidence exists for both. In addition, we use the t-test for the within-country autocorrelation in the error term assuming AR(1) error structure. The estimated AR(1) coefficient is 0.1 for the whole sample and we can reject the null hypothesis of AR(1) error structure. Therefore, we adopt the GLS estimation assuming an i.i.d. error structure. 4. Empirical results First, we describe the general statistical properties of savings and investment data in East Asia. Then, we report the estimation results of the S–I correlation before and after controlling the relevant factors, cyclical shocks and country differences. Three shocks are considered in the regression: productivity, fiscal and terms of trade shocks. A number of previous studies
5 We use the Augmented Dickey–Fuller unit root test to check the order of integration of the data. The test results show that both saving and investment are I(1) processes for most countries, which guide us to use the first differenced data for regression analysis.
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
283
Table 1 Properties of saving and investment (percentage, period average) 1980–2002
1980–1989
1990–2002
S/Y
I/Y
S/Y
I/Y
S/Y
I/Y
China Hong Kong Indonesia Japan Korea Malaysia Philippines Singapore Taiwan Thailand
37.9 32.3 30.2 30.6 33.5 37.8 18.9 44.9 30.3 30.8
37.2 28.6 27.2 29.0 32.4 32.4 22.0 37.1 23.1 31.6
34.8 33.5 31.0 31.7 31.0 33.2 23.0 41.6 34.1 26.0
35.4 28.1 28.8 29.8 30.3 30.7 22.2 42.4 23.7 29.4
40.3 31.5 29.6 29.8 35.5 41.4 15.7 47.4 27.4 34.5
38.5 28.9 26.0 28.4 34.1 33.7 21.9 32.9 22.6 33.3
Total Big 3 ASEAN Greater China
32.7 34.0 32.5 33.5
30.1 32.9 30.1 29.6
32.0 32.5 31.0 34.1
30.1 31.8 30.7 29.1
33.3 35.2 33.7 33.1
30.0 33.6 29.6 30.0
Note: Big 3 (Korea, Japan, China), ASEAN (Indonesia, Thailand, Malaysia, Singapore and the Philippines), Greater China (Hong Kong, Taiwan, China).
confirmed that these shocks have significant effects on both saving and investment.6 Finally, we report S–I correlation using regional data. A detailed explanation of the data set is in Appendix A. 4.1. Statistical properties of saving and investment Table 1 reports the major statistical properties of saving and investment (as a ratio of GDP) in East Asia. The first column reports the statistics of the whole period (1980–2002). Average savings and investment rates of the 10 countries are 33 and 30 percent, respectively. Singapore exhibits the highest savings rate (about 45 percent), while China shows the highest investment rate of 37 percent. The Philippines have the lowest savings and investment rates, 19 and 22 percent, respectively. The average savings rate is higher than the average investment rate in all countries except for the Philippines and Thailand. Next, we divide the whole sample period into two: 1980–1989 and 1990–2002. Four countries exhibit higher savings and investment rates in the second period: China, Korea, Malaysia and Thailand. In particular, the savings rate in China, Malaysia and Thailand increase by 6–7 percent over this period, while investment rates go up by 3–4 percent. The table also reports average statistics of sub-group countries: the big three (China, Japan and Korea), ASEAN (Indonesia, Thailand, the Philippines, Malaysia, and Singapore), and greater China (Hong Kong, Taiwan and China). Both the savings and investment rates of the big three are the highest among the three sub-groups. In particular, the average investment rate of the big three is about 33 percent, while it is around 30 percent in the other two groups.
6
See Kim (2001) for details.
284
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
Table 2 Saving–investment correlation (I/Y)i = a + b(S/Y)i
OLS
Cross-sectional regression Whole period 1980–1989 1990–2002 1980–1994 1981–1995 1982–1996 1983–1997 1984–1998 1985–1999 1986–2000 1987–2001 1988–2002
0.65 0.85 0.51 0.76 0.75 0.72 0.70 0.67 0.63 0.58 0.55 0.53
(0.14) (0.26) (0.12) (0.21) (0.20) (0.19) (0.18) (0.15) (0.14) (0.13) (0.12) (0.13)
DIit = ai + bDSit
OLS
GLS
Panel regression Asian countries OECD (1960–1992)
1.06 (0.054) 0.91 (0.041)
0.88 (0.040) 0.70 (0.019)
Note: Numbers in the parentheses are S.E.
4.2. Cross-sectional and panel data estimation of S–I correlation First, we report the cross-sectional regression results of saving and investment rates using the OLS estimation method. Cross-sectional data are constructed by taking the averages of saving and investment rates (as of GDP) over different time periods. In the first panel of Table 2, we report the coefficients b from the following regression7: ðI=YÞi ¼ a þ bðS=YÞi :
(1)
Regression of the whole period data produces an S–I correlation of 0.65. The subperiod analysis shows that the S–I correlation decreases from 0.85 in the 1980–1989 period to 0.51 in the 1990– 2002 period. We also calculate the rolling S–I correlation in a 15-year window. The results show that the S–I correlation consistently decreases over time from 0.76 in the 1980–1994 period to 0.53 in the 1988–2002 period. These results confirm that the S–I correlation decreased significantly over time in East Asia. To measure the S–I correlation using panel data, we run the following regressions: DI it ¼ ai þ bi DSit ;
(2)
where ai denotes the country-specific intercept. The second panel of Table 2 reports the panel data regression results assuming that bi is common across countries. Estimation results with GLS show that the S–I correlation is approximately 0.88 in the entire sample period. Compared to the 7 We name b saving–investment correlation for convenience, even though it is not correlation coefficient by definition. Other terms have been used to name this coefficient such as saving ratios coefficient, saving–investment relation, saving retention coefficient, etc.
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
285
Table 3 Saving–investment correlation (country-specific coefficient) (DIit = ai + biDSit) Country
OLS
China Hong Kong Indonesia Japan Korea Malaysia Philippines Singapore Taiwan Thailand
0.88 0.66 1.23 1.07 1.39 0.82 1.01 1.15 0.63 2.04
Average
1.09
GLS (0.16) (0.75) (0.48) (0.06) (0.41) (1.15) (1.47) (1.28) (0.83) (0.86)
1.66 0.73 0.73 1.00 0.04 0.45 1.02 0.45 0.22 1.63
(0.12) (0.07) (0.05) (0.07) (0.20) (0.16) (0.13) (0.11) (0.10) (0.15)
0.79
results of the 19 OECD countries (0.70 taken from Kim (2001)), the S–I correlation in East Asia is higher. The OLS estimation produces slightly higher coefficients (1.06 and 0.91 for East Asian and OECD countries, respectively) and larger standard errors than GLS. However, the main result remains the same—the correlation coefficient of the East Asian sample is higher than that of the OECD countries. Table 3 presents the S–I correlation estimated without equality restrictions on the coefficient. Although the S–I correlation coefficients differ across countries, they are statistically significant in most cases. The average coefficient is 0.79. Results show that China and Thailand exhibit a high correlation, approximately 1.6, while other countries show a correlation between 0.4 and 1, with the exception of Korea and Taiwan. For Korea, S–I correlation is near zero, while it is around 0.2 for Taiwan. With the OLS estimation, the S–I correlation generally increases compared to GLS estimation. 4.3. Exogenous shocks and S–I correlation In order to control for cyclical shocks, we use the residuals from the following panel regressions for saving and investment: DSit ðDI it Þ ¼ ai þ b0 shockit þ b1 shockit-1 þ b2 shockit-2 þ residualsit ;
(3)
where coefficients bs are assumed to be constant across countries. We set the lag length up to 2 for shocks because the coefficients from the lag length 3 and above are insignificant in most cases. Note that we do not specify any prior structure of the estimation model. In order to avoid multicollinearity problem resulting from the use of three shocks in the same regression, we first check if these shocks are systematically correlated. If the correlations among shocks are low, then we do not need to concern ourselves with the multicollinearity problem in regressions with multiple shocks. The results demonstrate that the average pair wise correlations of the three shocks are low: 0.13 (prod/fiscal), 0.07 (prod/TOT) and 0.04 (Fiscal/TOT) with standard errors of 0.19, 0.25, and 0.36, respectively. The low correlation coefficients confirm that we can use multiple shocks in the same regression.
286
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
Table 4 Effects of shocks on aggregate variables Lag
Prod shock
Fiscal shock
TOT shock
Output
0 1 2
0.164 (0.015) 0.142 (0.015) 0.019 (0.014)
0.024 (0.016) 0.184 (0.016) 0.194 (0.016)
0.024 (0.011) 0.03 (0.011) 0.04 (0.011)
Consumption
0 1 2
0.015 (0.007) 0.017 (0.007) 0.001 (0.007)
0.01 (0.015) 0.046 (0.015) 0.031 (0.015)
0.01 (0.007) 0.014 (0.007) 0.019 (0.007)
Gov spending
0 1 2
0.008 (0.002) 0.012 (0.002) 0.011 (0.002)
0.075 (0.003) 0.009 (0.003) 0.001 (0.002)
0.001 (0.001) 0.003 (0.001) 0.009 (0.001)
Saving
0 1 2
0.026 (0.010) 0.013 (0.010) 0.001 (0.010)
0.096 (0.018) 0.038 (0.019) 0.073 (0.018)
0.016 (0.009) 0.005 (0.009) 0.005 (0.009)
Investment
0 1 2
0.066 (0.016) 0.022 (0.016) 0.004 (0.015)
0.099 (0.028) 0.05 (0.028) 0.006 (0.027)
0.004 (0.010) 0.027 (0.010) 0.066 (0.010)
In Table 4, we report the coefficients from the panel regressions of DY, DC, DG, DS, and DI on each shock with a lag structure.8 The numbers in parentheses are standard errors. The signs of the coefficients are as expected in most cases. First, an increase in productivity initially has positive and significant effects on output, consumption, and investment, but these effects diminish over time. Saving also increases because productivity shocks increase output more than consumption (consumption smoothing). Second, a positive fiscal shock (an unexpected increase in government spending) has a positive impact on output but ambiguous effects on consumption. Saving initially decreases due to an initial increase in government spending (S = Y C G). However, effects on saving become positive over time as output rises. Investment initially increases but rapidly decreases over time. In theoretical models, the effects of fiscal shocks on saving and investment are ambiguous since the effects depend on the specification of the shocks. The empirical results in the table also indicate ambiguous effects of fiscal shocks. Third, an increase in TOT has a positive effect on output with a time lag, an initial decrease followed by a significant increase over time. This observation is consistent with the well-known J-curve effects of the terms of trade: it takes time for the production sector to reflect the benefit from an improvement in TOT. Consumption increases over time following an initial downturn. Government spending is not affected by a change in TOT. Saving initially increases but then decreases over time. Investment decreases for the first 2 years followed by a significant increase in the third year showing a strong J-curve effect. The results in Table 4 provide additional findings. First, productivity shocks have the largest impact on all aggregate variables among the three shocks, which is consistent with the fact that productivity shocks are the most important source of business cycles in the economy. Second, the effects of lagged shocks are quite weak in
8 The shocks are all multiplied by the mean of each country’s real GNP over the estimation period. We can infer, therefore, that the coefficients are the change of left-hand side variables as a percent of average GNP in response to a one percent increase in each shock.
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
287
Table 5 S–I correlation after controlling for shocks Lag
Prod shock
Fiscal shock
TOT shock
All shocks
Benchmark value: 0.88 0 0.82 (0.04) 1 0.82 (0.04) 2 0.82 (0.04)
0.86 (0.05) 0.87 (0.05) 0.83 (0.05)
0.92 (0.04) 0.9 (0.04) 0.85 (0.04)
0.83 (0.04) 0.61 (0.05) 0.4 (0.03)
Benchmark value: 0.70 (OECD) 0 0.75 1 0.7 2 0.64
0.7 0.7 0.65
0.68 0.7 0.6
0.64 0.58 0.42
Note: All the correlation coefficients are significant with the 1 percent level in the OECD case.
the case of productivity shocks, while fiscal and TOT shocks have prolonged effects, especially on investment. Third, in most cases, the effects on consumption are less than the effects on output, which supports the consumption-smoothing property. Fourth, the effects of productivity and TOT shocks on government spending are significantly weak compared to the effects on other variables, Y, C, and I. This suggests that the government spending may not respond to the current shocks to the economy but does respond to other targets such as income, inflation, current account, or government debt of current and previous periods. In Table 5, we report the S–I correlations after controlling the cyclical properties caused by exogenous shocks in the economy. The correlation coefficients are derived from the regression of the residuals of investment on the residuals of saving where the residuals are derived from the regressions on exogenous shocks, as in (3), with different number of lags. In the first row, we report the S–I correlations after controlling each shock setting the number of lags at 0–2. Compared to the benchmark value of 0.88, the correlations decrease only by a small number. However, when taking all three shocks at the same time, the S–I correlation decreases to 0.83, 0.61 and 0.4 as the number of lags increases. Table 5 also reports the S–I correlation of OECD countries after controlling for cyclical shocks (Kim, 2001). The OECD countries’ results share some common properties with the results from the Asian sample: the S–I correlation decreases only by a small amount when controlling individual shocks (from benchmark of 0.7 to approximately 0.6) but when including all three shocks with the lag length 2, the correlation decreases to 0.42. In most instances, the S–I correlation for the OECD countries is smaller than that of the Asian countries, which is consistent with the case before controlling shocks. Table 6 presents the S–I correlation estimated without equality restrictions after controlling for all shocks. As before, the S–I correlation significantly differs across countries. Among the 10 countries, the S–I correlation decreases after controlling for all shocks except for two countries: Korea and Taiwan. Note that these two countries show the lowest S–I correlation among the 10 countries before controlling for shocks. These differences in coefficients are mainly due to the lack of data points. In individual country analysis, the number of observation is only 20. 4.4. Country differences and S–I correlation Panel data analysis enables us to examine the effects of country differences on the S–I correlation. Unlike cyclical shocks, we analyze the effects of country differences by running separate regressions with sub-country data sorted by each category, because these variables affect
288
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
Table 6 S–I correlation after controlling for shocks (country-specific coefficient) Country
OLS
GLS
No control
Control
No control
Control
China Hong Kong Indonesia Japan Korea Malaysia Philippines Singapore Taiwan Thailand
0.88 0.66 1.23 1.07 1.39 0.82 1.01 1.15 0.63 2.04
0.82 0.46 1.26 1.07 1.01 0.61 0.80 1.04 0.44 2.10
1.66 0.73 0.73 1.00 0.04 0.45 1.02 0.45 0.22 1.63
0.02 0.27 0.26 0.71 0.47 0.11 1.07 0.08 0.39 0.68
Average
1.09
(0.16) (0.75) (0.48) (0.06) (0.41) (1.15) (1.47) (1.28) (0.83) (0.86)
0.96
(0.16) (0.89) (0.70) (0.09) (0.49) (1.40) (1.18) (1.55) (1.13) (1.20)
0.78
(0.12) (0.07) (0.05) (0.07) (0.20) (0.16) (0.13) (0.11) (0.10) (0.15)
(0.11) (0.05) (0.03) (0.07) (0.19) (0.06) (0.12) (0.09) (0.08) (0.19)
0.41
Note: ‘Control’ denotes the case when all shocks up to lag length 2 are controlled.
the S–I correlation cross-sectionally or over relatively long horizon. We focus on two variables: country size and the size of the non-traded sector. First, a country with a large share of the world output is likely to have a relatively large share of the world’s total saving and investment. Small countries take the world interest rate as given, while changes in the investment and saving behavior of large countries will have an impact on the world interest rate. For example, an increase in the national saving of a large country lowers the world interest rate, which increases the investment of all countries. Therefore, the saving– investment correlation of a large country tends to be high.9 In Table 7, we sort 10 Asian countries into three groups according to the size of the real GDP denominated in constant US dollar, and run separate regressions with each country group.10 Theory predicts that the S–I correlation increases with country size. However, in both models, with and without controlling for cyclical effects, the S–I correlation does not have a positive relationship with country size. Both OLS and GLS estimations produce similar results. Second, as long as domestic residents consume traded as well as non-traded goods, an increase in saving leads to an increase in wealth and in the future consumption path. However, consumption in non-traded goods rises only if production of non-traded goods increases, which necessitates a rise in investment in the non-traded goods sector. Therefore, as long as non-traded goods exist, we should expect to find a correlation between saving and investment, even if capital is perfectly mobile across countries (Wong, 1990; Tesar, 1993). The second panel of Table 7 shows the effects of the non-traded sector on the S–I correlation. We sort 10 Asian countries into three groups using the ratio of (import/GNP) in the year 2002 as the proxy of the size of the nontraded sector.11 We assume that the smaller the relative size of the import sector, the larger the
9
See, for example, Murphy (1984), Dooley et al. (1987), and Bayoumi and Rose (1993). Large countries are China, Japan and Korea. Medium countries are Thailand, Taiwan, Singapore and Hong Kong, while small countries include Indonesia, Malaysia and the Philippines. 11 We used the same proxy that Wong (1990) used for estimating the size of the non-traded sector. Countries with large non-traded sectors (i.e., have small share of import) are China, Japan and Indonesia. Countries with medium non-traded sectors are Thailand, Taiwan, Korea and the Philippines, while countries with small non-traded sectors include Singapore, Hong Kong and Malaysia. 10
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
289
Table 7 S–I correlation and country differences OLS
GLS
No control
Control
No control
Control
1.06 (0.10) 1.07 (0.16) 1.16 (0.13)
0.93 (0.11) 0.9 (0.19) 1.32 (0.17)
0.98 (0.09) 1.01 (0.13) 0.83 (0.08)
0.78 (0.09) 0.82 (0.13) 0.72 (0.09)
Size of non-traded sector Large 1.05 (0.08) Medium 1.35 (0.28) Small 0.8 (0.20)
0.80 (0.10) 1.49 (0.30) 0.72 (0.19)
1.08 (0.08) 1.04 (0.13) 0.58 (0.16)
0.83 (0.08) 0.98 (0.13) 0.45 (0.15)
Size of GDP Large Medium Small
Table 8 Aggregate saving–investment correlation All Big 3 ASEAN Greater China
1.25 1.02 1.73 1.11
(0.20) (0.20) (0.26) (0.21)
non-traded sector. The regression results before controlling for shocks are consistent with the theoretical predictions: the larger the non-traded sector is, the higher the S–I correlation. However, after controlling for cyclical effects, this positive relationship disappears. 4.5. Aggregate S–I correlation Table 8 reports the S–I correlation using aggregate saving and investment data of the sample Asian countries. Aggregate data are constructed by taking the sum of each country’s variables (adjusted for currency unit and inflation). Regressions based on the aggregate data may indicate how much investment of a certain region is financed by regional savings. That is, we treat the whole Asian region as one country and examine the S–I correlation. Note that this aggregate level correlation may not reflect the true degree of international capital mobility because we do not consider intra-regional borrowing and lending. Regressions with regional savings and investment data show high S–I correlation, suggesting that the Asian investment is largely financed by savings within the region. The table also shows that the S–I correlation is larger in the ASEAN countries compared to those in the big three or in greater China. This suggests that investment in the ASEAN countries is more likely to be financed by regional savings (less capital mobility), while investment in big three countries are relatively more financed by savings outside of the region (more capital mobility). 5. Conclusion In this paper, we have explored the S–I correlation of East Asian countries in relation to international capital mobility. Generally, the direction of changes in the S–I correlation over time is consistent with the changes in the degree of capital mobility; the S–I correlation decreases as
290
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
capital mobility increases over time. Additionally, the S–I correlation in East Asia is always less than that in the OECD countries in all periods. This is consistent with the fact that capital mobility in East Asia is lower than that among the OECD countries. On the other hand, even after controlling cyclical shocks, the S–I correlation remains positive, which may imply that the absolute degree of capital mobility is still low. Despite these findings, it is difficult to directly infer the degree of capital mobility of individual countries because the S–I correlation varies greatly across countries in a countryspecific analysis. This cross-country distinction may reflect either the presence of other countryspecific factors that are not considered in this paper or the small sample size. For example, if a government maintains a current account targeting policy, then the current account would not fluctuate to the optimal level and in turn, saving and investment would be highly correlated even in the presence of perfect capital mobility. Therefore, the S–I correlation may provide indirect evidence of a government’s stance on the current account policy, whether governments target the current account balance as their policy goal or simply allow for the current account as a residual of economic activity. Acknowledgement We are grateful for financial support from the Korea Institute of International Economic Policy. Appendix A Ten Asian countries are analyzed in this paper: China, Hong Kong, Indonesia, Japan, Korea, Malaysia, the Philippines, Singapore, Thailand, and Taiwan. We use annual data from 1980 to 2002. Saving is defined as the GDP minus government consumption and private consumption. Investment is gross fixed capital formation plus changes in stocks. Most of the national income accounting data is taken from the International Financial Statistics, IMF. The Asian Development Bank, Bank of Korea, Ministry of Finance in Japan and Taiwan are additional sources of data. We convert local currency data into the US dollar using period average exchange rates and nominal values into real values using GDP deflator. Productivity shocks are defined as annual percentage changes in productivity. For productivity measure, we use Solow residuals. The share of labor in manufacturing output is assumed to be 0.6. Industrial production data has been used for output and the employment data for labor input. Fiscal shocks are defined as percentage changes in unexpected government spending. The use of government spending data for fiscal shocks may cause an endogeneity problem with other national income accounting variables. We construct the data for the unexpected government spending by assuming that policymakers determine the growth rate of government spending at the start of a fiscal year considering the predicted annual GDP growth rate and the net government debt at the start of the year. Government debt data are taken from the IFS (line 80, net government surplus/deficit in national currencies). We assume that policymakers have perfect foresight and run the country-by-country OLS regression of the growth rate of real government spending at time t on the real GDP growth rates and the real net government debt at the start of the year. We use the residuals of these regressions for the unexpected government spending data. Terms of trade (TOT) shocks are defined as the percentage changes in the TOT—export price/import price. Export and import prices are unit values of exports and imports taken from IFS (lines 74, 75), respectively. For Indonesia and Malaysia, we use only export prices as import price index data are
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
291
not available. For Malaysia, export price index is constructed by taking the weighted average of three main export goods’ prices, rubber, tin and palm. References Backus, D., Kehoe, P., Kydland, F., 1995. International business cycles: theory and evidence. In: Cooley, T. (Ed.), Frontiers of Business Cycle Research. Princeton University Press, Princeton, NJ. Baxter, M., Crucini, M., 1993. Explaining saving–investment correlations. American Economic Review 83, 416–436. Bayoumi, T., 1989. Saving-investment correlations: immobile capital, government policy or endogenous behavior? IMF Working Paper 66. Bayoumi, T., Rose, A., 1993. Domestic savings and intra-national capital flows. European Economic Review 37, 1197– 1202. Coakley, J., Kulasi, F., Smith, R., 1998. The Feldstein–Horioka puzzle and capital mobility: a review. International Journal of Finance and Economics 3, 169–188. Dooley, M., Frankel, J., Mathieson, D., 1987. International capital mobility: what do saving–investment correlations tell us? IMF Staff Papers 34, 503–529. Feldstein, M., 1983. Domestic saving and international capital movements in the long run and the short run. European Economic Review 21, 129–151. Feldstein, M., Horioka, C., 1980. Domestic saving and international capital flows. Economic Journal 90, 314–329. Feldstein, M., Bacchetta, P., 1991. National saving and international investment. In: Bernheim, B., Shoven, J. (Eds.), National Saving and Economic Performance. University of Chicago Press, Chicago, IL. Glick, R., Rogoff, K., 1995. Global versus country-specific productivity shocks and the current account. Journal of Monetary Economics 35, 159–192. Golub, S., 1990. International capital mobility: net versus gross stocks and flows. Journal of International Money and Finance 9, 424–439. Kim, S.H., 2001. The saving–investment correlation puzzle is still a puzzle. Journal of International Money and Finance 20, 1017–1034. Mendoza, E., 1991. Real business cycles in a small open economy. American Economic Review 81, 797–818. Murphy, R., 1984. Capital mobility and the relationship between saving and investment in OECD countries. Journal of International Money and Finance 3, 327–342. Obstfeld, M., 1995. International capital mobility in the 1990s. In: Kenen, P. (Ed.), Understanding Interdependence: The Macroeconomics of the Open Economy. Princeton University Press, Princeton, NJ. Sinn, S., 1992. Saving–investment correlations and capital mobility: on the evidence from annual data. Economic Journal 102, 1162–1170. Taylor, A.M., 1994. Domestic saving and international capital flows reconsidered. NBER Working Paper No. 4892. Tesar, L., 1991. Savings, investment and international capital flows. Journal of International Economics 31, 55–78. Tesar, L., 1993. International risk sharing and non-traded goods. Journal of International Economics 31, 55–78. Wong, D., 1990. What do saving–investment relationships tell us about capital mobility? Journal of International Money and Finance 9, 60–74.
Saving, investment and international capital mobility in East Asia§ Soyoung Kim a, Sunghyun H. Kim b,*, Yunjong Wang c a
Department of Economics, Korea University, 5-1 Anam-Dong, Sungbuk-Ku, Seoul 136-701, Republic of Korea b Department of Economics, Tufts University, Medford, MA 02155, United States c SK Research Institute, Seoul Finance Center, 84 Taepyungro 1-ga, Chung-gu, Seoul 100-101, Republic of Korea Received 17 November 2005; received in revised form 9 May 2006; accepted 12 May 2006
Abstract This paper estimates the degree of international capital mobility in East Asia using the saving– investment correlation originated in Feldstein and Horioka [Feldstein, M., Horioka, C., 1980. Domestic saving and international capital flows. Economic Journal 90, 314–329]. We apply the empirical method used in Kim [Kim, S.H., 2001. The saving–investment correlation puzzle is still a puzzle. Journal of International Money and Finance 20, 1017–1034] to control for cyclical effects in estimating a time-series saving– investment correlation of 10 Asian countries from 1980 to 2002. Our conclusion is that the saving– investment correlation in East Asia steadily decreases over time but is still higher than that of the OECD countries over all studied periods. These results are consistent with the fact that capital mobility in East Asia is lower than that in the OECD countries. In addition, regional saving and investment data demonstrate that investment in East Asia is largely financed by regional savings. # 2006 Elsevier B.V. All rights reserved. JEL classification: F32; F41 Keywords: Saving–investment correlation; Feldstein–Horioka puzzle; Capital mobility; East Asia; Business cycles
1. Introduction During the 1990s, a number of East and Southeast Asian countries liberalized their financial markets to foreign capital by reducing restrictions on inward and outward capital flows. As a result, net private capital inflows to this region in the mid-1990s were conspicuous in the postwar
§
An earlier version of this paper was published as KIEP Working Paper No. 04-01. * Corresponding author. Tel.: +1 617 627 3662; fax: +1 617 627 3917. E-mail address: [email protected] (S.H. Kim).
0922-1425/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.japwor.2006.05.001
280
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
period in terms of the size of the flow to emerging markets. This paper attempts to measure the degree of international capital mobility in East Asia using a saving–investment correlation (S–I correlation hereafter). Since the seminal paper by Feldstein and Horioka (1980, denoted by FH hereafter), numerous economists have attempted to quantify the degree of international capital mobility using both a cross-sectional and a time-series S–I correlation. Under perfect capital mobility, national saving and national investment should not be correlated as domestic agents search for worldwide investment and saving opportunities regardless of the nationality of capital. Therefore, a high correlation between national saving and investment would imply a low degree of capital mobility. However, since many factors other than capital mobility can simultaneously affect saving and investment, it is difficult to interpret the S–I correlation as a correct measure for the degree of capital mobility.1 This paper adopts the empirical method in Kim (2001) to estimate the time-series S–I correlation after controlling for business cycle shocks, to provide an accurate measure of capital mobility. We remove the cyclical effects from saving and investment data by using the residuals from the regressions of saving and investment on cyclical shocks including productivity shock, fiscal shock and the terms of trade (TOT) shock. In this paper, we test whether this newly estimated S–I correlation correctly reflects the liberalization of capital markets in East Asia. In particular, we compare the S–I correlation in East Asia with that of the OECD countries to examine the differences in the degree of capital mobility.2 We also investigate whether country differences such as the size of the GDP and the non-traded sector have any systematic relationship with S–I correlation. Finally, we use regional saving and investment data to investigate how much of the investment in East Asia is financed by regional savings. The data used originates from panel data of saving and investment from 10 East Asian countries between 1980 and 2002. For panel data regression, we use the GLS estimation of the SUR system to control for cross-country heteroscedasticity and cross-country correlation in the error term. The estimation results illustrate that the cross-sectional S–I correlation steadily decreases over time (0.76 in the early 1980s to 0.53 in the late 1990s) but remains high in absolute terms. This result follows previous papers, including the original FH study. Panel data regression shows that the time-series S–I correlation in East Asia is approximately 0.88 and slightly decreases after controlling for cyclical shocks. These estimates are higher than those of OECD countries (0.70 before controlling cyclical shocks), which is consistent with the fact that capital mobility in East Asia is lower than that among the OECD countries. Country differences do not significantly impact the S–I correlation. Finally, aggregate regional data of saving and investment show that investment in East Asia is largely financed by regional savings. The following is a layout of what follows in this paper. In Section 2, we provide an overview of prior empirical studies on the S–I correlation. Section 3 explains the empirical methodology that we adopt in this paper. Section 4 reports the main statistical properties of the saving and investment data and estimation results including cross-sectional and time-series panel regressions. Finally, Section 5 offers our conclusion.
1
This endogeneity problem has been a main econometric concern since the original paper by Feldstein and Horioka (1980). They assumed that population structure is the main source of endogeneity problem in cross-sectional regressions, while business cycle shocks are the main source of endogeneity problem in time-series regressions. 2 Estimation results for the OECD countries are taken from Kim (2001).
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
281
Box 1. Summary of empirical studies on S–I correlation Paper
Data
Estimation results
Cross-sectional analysis Feldstein and Horioka (1980) Feldstein (1983) Murphy (1984) Dooley et al. (1987) Bayoumi (1989) Golub (1990) Tesar (1991)
16 OECD countries, 1960–1974, S,I as shares of GDP 17 OECD countries, 1960–1979 17 OECD countries, 1960–1980 14 OECD countries, 1960–1980 10 OECD countries, 1965–1986 16 OECD countries, 1960–1988 16–23 OECD countries, 1960–1986
0.89 (gross measures), 0.94 (net measures) 0.80 0.85 1960–1973: 0.75, 1974–1980: 0.74 OLS: 0.97, two-stage LS: 0.85 0.76 1960–1986: 0.84, 1960–1974: 0.89 1975–1986: 0.81 1960–1986: 0.83, 1960–1969: 0.85, 1970–1979: 0.67, 1980–1986: 0.87 1960–1969: 0.87, 1970–1979: 0.82, 1980–1989: 0.68 1974–1990: 0.72, 1974–1980: 0.87, 1981–1990: 0.64
Feldstein and Bacchetta (1991) Sinn (1992)
23 OECD countries, 1960–88
Obstfeld (1995)
20 OECD countries, 1974–1990
Time-series analysis Bayoumi (1989)
23 OECD countries, 1960–1986
1961–1986
Tesar (1993)
1960–1988 OLS, S,I as ratios of GDP
Obstfeld (1995)
1974–1990 OLS, S,I as ratios of GDP
Kim (2001)
21 OECD countries, 1960–1992, panel (GLS) with first differenced data
US: 1.00, Japan: 0.84, Germany: 0.87, UK: 0.33, France: 0.80, Canada: 0.83, Finland: 0.98, Norway: 0.21 US: 0.752, Canada: 0.848, France: 0.929, Germany: 0.886, Italy: 0.063, UK: 0.592 US: 0.848, UK: 0.113, Japan: 1.161, Italy: 0.214, France: 0.909, Germany: 0.327 0.70
2. Overview of literature A number of papers have empirically estimated the cross-sectional and time-series correlation of saving and investment. Box 1 summarizes the main results of these papers.3 Most crosssectional regressions reveal that the S–I correlation was high even in the 1990s, while time-series estimation results depend too much on individual countries to generate any general conclusion. Since the late 1980s, there has been an increasing number of simulation studies based on real business cycle models that explain a high S–I correlation using different types of shocks, even under perfect capital mobility.4 Although numerous papers have suggested multiple shocks and economic structures to explain the high S–I correlation, many have been limited to simulation or vector autoregression methods and have not offered strong empirical support. Moreover, their results depend on the specification of the model, particularly the structure of investment and the specification of shocks.
3
Refer to Coakley et al. (1998) for a detailed survey on this literature. See for example, Mendoza (1991), Backus et al. (1995), Baxter and Crucini (1993), Taylor (1994) and Glick and Rogoff (1995). 4
282
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
3. Empirical methodology Most earlier empirical papers applied a cross-sectional analysis with time-averaged data and related the S–I correlation to long-term capital movement. Such an application averaged out high-frequency fluctuations in the data. However, a cross-sectional analysis is not appropriate in examining Asian countries since the capital market liberalization process was quite rapid and concentrated in a short period of time, and since the timing of liberalization was different among countries. Time-averaging removes these characteristics from the data. Therefore, a time-series analysis is more appropriate, as it reflects short-term capital movements over time. A time-series analysis does pose several econometric problems. First, S–I correlations from country-by-country time-series regressions are insignificant in many cases and too different across countries to identify any systematic relationship with capital mobility. Second, both saving and investment data are subject to cyclical shocks, which can create an endogeneity problem in which regressors are correlated with error terms. Third, time-series data of saving and investment are usually non-stationary, which can result in a spurious regression. We overcome these problems by adopting a panel data analysis and by controlling business cycle effects caused by exogenous shocks. The use of a panel data set provides a more efficient estimator compared to the case of a single time-series estimation. It is difficult to reflect a crosscountry correlation in the cross-sectional regression and without considering cross-country correlation, the inference on coefficient estimates may not be consistent. The panel data regression allows us to reflect a cross-country correlation and therefore avoid this inconsistency problem. We are also able to specifically incorporate country-specific effects in the panel regression. In order to remove cyclical properties from the data, we run the regressions of saving and investment on business cycle shocks and use the residuals from these regressions to derive the S–I correlation. We avoid the spurious regression problem by using the first differenced data.5 For estimation of the panel data system, we apply the GLS estimation calculated by iterating the SUR system using newly computed covariances and system equation estimates. Though we set a maximum of 20 iterations, all of the results converge before reaching this limit. We test if there is cross-country heteroscedasticity and a cross-country correlation in the error term using the Lagrange Multiplier tests. Test statistics indicate that strong evidence exists for both. In addition, we use the t-test for the within-country autocorrelation in the error term assuming AR(1) error structure. The estimated AR(1) coefficient is 0.1 for the whole sample and we can reject the null hypothesis of AR(1) error structure. Therefore, we adopt the GLS estimation assuming an i.i.d. error structure. 4. Empirical results First, we describe the general statistical properties of savings and investment data in East Asia. Then, we report the estimation results of the S–I correlation before and after controlling the relevant factors, cyclical shocks and country differences. Three shocks are considered in the regression: productivity, fiscal and terms of trade shocks. A number of previous studies
5 We use the Augmented Dickey–Fuller unit root test to check the order of integration of the data. The test results show that both saving and investment are I(1) processes for most countries, which guide us to use the first differenced data for regression analysis.
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
283
Table 1 Properties of saving and investment (percentage, period average) 1980–2002
1980–1989
1990–2002
S/Y
I/Y
S/Y
I/Y
S/Y
I/Y
China Hong Kong Indonesia Japan Korea Malaysia Philippines Singapore Taiwan Thailand
37.9 32.3 30.2 30.6 33.5 37.8 18.9 44.9 30.3 30.8
37.2 28.6 27.2 29.0 32.4 32.4 22.0 37.1 23.1 31.6
34.8 33.5 31.0 31.7 31.0 33.2 23.0 41.6 34.1 26.0
35.4 28.1 28.8 29.8 30.3 30.7 22.2 42.4 23.7 29.4
40.3 31.5 29.6 29.8 35.5 41.4 15.7 47.4 27.4 34.5
38.5 28.9 26.0 28.4 34.1 33.7 21.9 32.9 22.6 33.3
Total Big 3 ASEAN Greater China
32.7 34.0 32.5 33.5
30.1 32.9 30.1 29.6
32.0 32.5 31.0 34.1
30.1 31.8 30.7 29.1
33.3 35.2 33.7 33.1
30.0 33.6 29.6 30.0
Note: Big 3 (Korea, Japan, China), ASEAN (Indonesia, Thailand, Malaysia, Singapore and the Philippines), Greater China (Hong Kong, Taiwan, China).
confirmed that these shocks have significant effects on both saving and investment.6 Finally, we report S–I correlation using regional data. A detailed explanation of the data set is in Appendix A. 4.1. Statistical properties of saving and investment Table 1 reports the major statistical properties of saving and investment (as a ratio of GDP) in East Asia. The first column reports the statistics of the whole period (1980–2002). Average savings and investment rates of the 10 countries are 33 and 30 percent, respectively. Singapore exhibits the highest savings rate (about 45 percent), while China shows the highest investment rate of 37 percent. The Philippines have the lowest savings and investment rates, 19 and 22 percent, respectively. The average savings rate is higher than the average investment rate in all countries except for the Philippines and Thailand. Next, we divide the whole sample period into two: 1980–1989 and 1990–2002. Four countries exhibit higher savings and investment rates in the second period: China, Korea, Malaysia and Thailand. In particular, the savings rate in China, Malaysia and Thailand increase by 6–7 percent over this period, while investment rates go up by 3–4 percent. The table also reports average statistics of sub-group countries: the big three (China, Japan and Korea), ASEAN (Indonesia, Thailand, the Philippines, Malaysia, and Singapore), and greater China (Hong Kong, Taiwan and China). Both the savings and investment rates of the big three are the highest among the three sub-groups. In particular, the average investment rate of the big three is about 33 percent, while it is around 30 percent in the other two groups.
6
See Kim (2001) for details.
284
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
Table 2 Saving–investment correlation (I/Y)i = a + b(S/Y)i
OLS
Cross-sectional regression Whole period 1980–1989 1990–2002 1980–1994 1981–1995 1982–1996 1983–1997 1984–1998 1985–1999 1986–2000 1987–2001 1988–2002
0.65 0.85 0.51 0.76 0.75 0.72 0.70 0.67 0.63 0.58 0.55 0.53
(0.14) (0.26) (0.12) (0.21) (0.20) (0.19) (0.18) (0.15) (0.14) (0.13) (0.12) (0.13)
DIit = ai + bDSit
OLS
GLS
Panel regression Asian countries OECD (1960–1992)
1.06 (0.054) 0.91 (0.041)
0.88 (0.040) 0.70 (0.019)
Note: Numbers in the parentheses are S.E.
4.2. Cross-sectional and panel data estimation of S–I correlation First, we report the cross-sectional regression results of saving and investment rates using the OLS estimation method. Cross-sectional data are constructed by taking the averages of saving and investment rates (as of GDP) over different time periods. In the first panel of Table 2, we report the coefficients b from the following regression7: ðI=YÞi ¼ a þ bðS=YÞi :
(1)
Regression of the whole period data produces an S–I correlation of 0.65. The subperiod analysis shows that the S–I correlation decreases from 0.85 in the 1980–1989 period to 0.51 in the 1990– 2002 period. We also calculate the rolling S–I correlation in a 15-year window. The results show that the S–I correlation consistently decreases over time from 0.76 in the 1980–1994 period to 0.53 in the 1988–2002 period. These results confirm that the S–I correlation decreased significantly over time in East Asia. To measure the S–I correlation using panel data, we run the following regressions: DI it ¼ ai þ bi DSit ;
(2)
where ai denotes the country-specific intercept. The second panel of Table 2 reports the panel data regression results assuming that bi is common across countries. Estimation results with GLS show that the S–I correlation is approximately 0.88 in the entire sample period. Compared to the 7 We name b saving–investment correlation for convenience, even though it is not correlation coefficient by definition. Other terms have been used to name this coefficient such as saving ratios coefficient, saving–investment relation, saving retention coefficient, etc.
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
285
Table 3 Saving–investment correlation (country-specific coefficient) (DIit = ai + biDSit) Country
OLS
China Hong Kong Indonesia Japan Korea Malaysia Philippines Singapore Taiwan Thailand
0.88 0.66 1.23 1.07 1.39 0.82 1.01 1.15 0.63 2.04
Average
1.09
GLS (0.16) (0.75) (0.48) (0.06) (0.41) (1.15) (1.47) (1.28) (0.83) (0.86)
1.66 0.73 0.73 1.00 0.04 0.45 1.02 0.45 0.22 1.63
(0.12) (0.07) (0.05) (0.07) (0.20) (0.16) (0.13) (0.11) (0.10) (0.15)
0.79
results of the 19 OECD countries (0.70 taken from Kim (2001)), the S–I correlation in East Asia is higher. The OLS estimation produces slightly higher coefficients (1.06 and 0.91 for East Asian and OECD countries, respectively) and larger standard errors than GLS. However, the main result remains the same—the correlation coefficient of the East Asian sample is higher than that of the OECD countries. Table 3 presents the S–I correlation estimated without equality restrictions on the coefficient. Although the S–I correlation coefficients differ across countries, they are statistically significant in most cases. The average coefficient is 0.79. Results show that China and Thailand exhibit a high correlation, approximately 1.6, while other countries show a correlation between 0.4 and 1, with the exception of Korea and Taiwan. For Korea, S–I correlation is near zero, while it is around 0.2 for Taiwan. With the OLS estimation, the S–I correlation generally increases compared to GLS estimation. 4.3. Exogenous shocks and S–I correlation In order to control for cyclical shocks, we use the residuals from the following panel regressions for saving and investment: DSit ðDI it Þ ¼ ai þ b0 shockit þ b1 shockit-1 þ b2 shockit-2 þ residualsit ;
(3)
where coefficients bs are assumed to be constant across countries. We set the lag length up to 2 for shocks because the coefficients from the lag length 3 and above are insignificant in most cases. Note that we do not specify any prior structure of the estimation model. In order to avoid multicollinearity problem resulting from the use of three shocks in the same regression, we first check if these shocks are systematically correlated. If the correlations among shocks are low, then we do not need to concern ourselves with the multicollinearity problem in regressions with multiple shocks. The results demonstrate that the average pair wise correlations of the three shocks are low: 0.13 (prod/fiscal), 0.07 (prod/TOT) and 0.04 (Fiscal/TOT) with standard errors of 0.19, 0.25, and 0.36, respectively. The low correlation coefficients confirm that we can use multiple shocks in the same regression.
286
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
Table 4 Effects of shocks on aggregate variables Lag
Prod shock
Fiscal shock
TOT shock
Output
0 1 2
0.164 (0.015) 0.142 (0.015) 0.019 (0.014)
0.024 (0.016) 0.184 (0.016) 0.194 (0.016)
0.024 (0.011) 0.03 (0.011) 0.04 (0.011)
Consumption
0 1 2
0.015 (0.007) 0.017 (0.007) 0.001 (0.007)
0.01 (0.015) 0.046 (0.015) 0.031 (0.015)
0.01 (0.007) 0.014 (0.007) 0.019 (0.007)
Gov spending
0 1 2
0.008 (0.002) 0.012 (0.002) 0.011 (0.002)
0.075 (0.003) 0.009 (0.003) 0.001 (0.002)
0.001 (0.001) 0.003 (0.001) 0.009 (0.001)
Saving
0 1 2
0.026 (0.010) 0.013 (0.010) 0.001 (0.010)
0.096 (0.018) 0.038 (0.019) 0.073 (0.018)
0.016 (0.009) 0.005 (0.009) 0.005 (0.009)
Investment
0 1 2
0.066 (0.016) 0.022 (0.016) 0.004 (0.015)
0.099 (0.028) 0.05 (0.028) 0.006 (0.027)
0.004 (0.010) 0.027 (0.010) 0.066 (0.010)
In Table 4, we report the coefficients from the panel regressions of DY, DC, DG, DS, and DI on each shock with a lag structure.8 The numbers in parentheses are standard errors. The signs of the coefficients are as expected in most cases. First, an increase in productivity initially has positive and significant effects on output, consumption, and investment, but these effects diminish over time. Saving also increases because productivity shocks increase output more than consumption (consumption smoothing). Second, a positive fiscal shock (an unexpected increase in government spending) has a positive impact on output but ambiguous effects on consumption. Saving initially decreases due to an initial increase in government spending (S = Y C G). However, effects on saving become positive over time as output rises. Investment initially increases but rapidly decreases over time. In theoretical models, the effects of fiscal shocks on saving and investment are ambiguous since the effects depend on the specification of the shocks. The empirical results in the table also indicate ambiguous effects of fiscal shocks. Third, an increase in TOT has a positive effect on output with a time lag, an initial decrease followed by a significant increase over time. This observation is consistent with the well-known J-curve effects of the terms of trade: it takes time for the production sector to reflect the benefit from an improvement in TOT. Consumption increases over time following an initial downturn. Government spending is not affected by a change in TOT. Saving initially increases but then decreases over time. Investment decreases for the first 2 years followed by a significant increase in the third year showing a strong J-curve effect. The results in Table 4 provide additional findings. First, productivity shocks have the largest impact on all aggregate variables among the three shocks, which is consistent with the fact that productivity shocks are the most important source of business cycles in the economy. Second, the effects of lagged shocks are quite weak in
8 The shocks are all multiplied by the mean of each country’s real GNP over the estimation period. We can infer, therefore, that the coefficients are the change of left-hand side variables as a percent of average GNP in response to a one percent increase in each shock.
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
287
Table 5 S–I correlation after controlling for shocks Lag
Prod shock
Fiscal shock
TOT shock
All shocks
Benchmark value: 0.88 0 0.82 (0.04) 1 0.82 (0.04) 2 0.82 (0.04)
0.86 (0.05) 0.87 (0.05) 0.83 (0.05)
0.92 (0.04) 0.9 (0.04) 0.85 (0.04)
0.83 (0.04) 0.61 (0.05) 0.4 (0.03)
Benchmark value: 0.70 (OECD) 0 0.75 1 0.7 2 0.64
0.7 0.7 0.65
0.68 0.7 0.6
0.64 0.58 0.42
Note: All the correlation coefficients are significant with the 1 percent level in the OECD case.
the case of productivity shocks, while fiscal and TOT shocks have prolonged effects, especially on investment. Third, in most cases, the effects on consumption are less than the effects on output, which supports the consumption-smoothing property. Fourth, the effects of productivity and TOT shocks on government spending are significantly weak compared to the effects on other variables, Y, C, and I. This suggests that the government spending may not respond to the current shocks to the economy but does respond to other targets such as income, inflation, current account, or government debt of current and previous periods. In Table 5, we report the S–I correlations after controlling the cyclical properties caused by exogenous shocks in the economy. The correlation coefficients are derived from the regression of the residuals of investment on the residuals of saving where the residuals are derived from the regressions on exogenous shocks, as in (3), with different number of lags. In the first row, we report the S–I correlations after controlling each shock setting the number of lags at 0–2. Compared to the benchmark value of 0.88, the correlations decrease only by a small number. However, when taking all three shocks at the same time, the S–I correlation decreases to 0.83, 0.61 and 0.4 as the number of lags increases. Table 5 also reports the S–I correlation of OECD countries after controlling for cyclical shocks (Kim, 2001). The OECD countries’ results share some common properties with the results from the Asian sample: the S–I correlation decreases only by a small amount when controlling individual shocks (from benchmark of 0.7 to approximately 0.6) but when including all three shocks with the lag length 2, the correlation decreases to 0.42. In most instances, the S–I correlation for the OECD countries is smaller than that of the Asian countries, which is consistent with the case before controlling shocks. Table 6 presents the S–I correlation estimated without equality restrictions after controlling for all shocks. As before, the S–I correlation significantly differs across countries. Among the 10 countries, the S–I correlation decreases after controlling for all shocks except for two countries: Korea and Taiwan. Note that these two countries show the lowest S–I correlation among the 10 countries before controlling for shocks. These differences in coefficients are mainly due to the lack of data points. In individual country analysis, the number of observation is only 20. 4.4. Country differences and S–I correlation Panel data analysis enables us to examine the effects of country differences on the S–I correlation. Unlike cyclical shocks, we analyze the effects of country differences by running separate regressions with sub-country data sorted by each category, because these variables affect
288
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
Table 6 S–I correlation after controlling for shocks (country-specific coefficient) Country
OLS
GLS
No control
Control
No control
Control
China Hong Kong Indonesia Japan Korea Malaysia Philippines Singapore Taiwan Thailand
0.88 0.66 1.23 1.07 1.39 0.82 1.01 1.15 0.63 2.04
0.82 0.46 1.26 1.07 1.01 0.61 0.80 1.04 0.44 2.10
1.66 0.73 0.73 1.00 0.04 0.45 1.02 0.45 0.22 1.63
0.02 0.27 0.26 0.71 0.47 0.11 1.07 0.08 0.39 0.68
Average
1.09
(0.16) (0.75) (0.48) (0.06) (0.41) (1.15) (1.47) (1.28) (0.83) (0.86)
0.96
(0.16) (0.89) (0.70) (0.09) (0.49) (1.40) (1.18) (1.55) (1.13) (1.20)
0.78
(0.12) (0.07) (0.05) (0.07) (0.20) (0.16) (0.13) (0.11) (0.10) (0.15)
(0.11) (0.05) (0.03) (0.07) (0.19) (0.06) (0.12) (0.09) (0.08) (0.19)
0.41
Note: ‘Control’ denotes the case when all shocks up to lag length 2 are controlled.
the S–I correlation cross-sectionally or over relatively long horizon. We focus on two variables: country size and the size of the non-traded sector. First, a country with a large share of the world output is likely to have a relatively large share of the world’s total saving and investment. Small countries take the world interest rate as given, while changes in the investment and saving behavior of large countries will have an impact on the world interest rate. For example, an increase in the national saving of a large country lowers the world interest rate, which increases the investment of all countries. Therefore, the saving– investment correlation of a large country tends to be high.9 In Table 7, we sort 10 Asian countries into three groups according to the size of the real GDP denominated in constant US dollar, and run separate regressions with each country group.10 Theory predicts that the S–I correlation increases with country size. However, in both models, with and without controlling for cyclical effects, the S–I correlation does not have a positive relationship with country size. Both OLS and GLS estimations produce similar results. Second, as long as domestic residents consume traded as well as non-traded goods, an increase in saving leads to an increase in wealth and in the future consumption path. However, consumption in non-traded goods rises only if production of non-traded goods increases, which necessitates a rise in investment in the non-traded goods sector. Therefore, as long as non-traded goods exist, we should expect to find a correlation between saving and investment, even if capital is perfectly mobile across countries (Wong, 1990; Tesar, 1993). The second panel of Table 7 shows the effects of the non-traded sector on the S–I correlation. We sort 10 Asian countries into three groups using the ratio of (import/GNP) in the year 2002 as the proxy of the size of the nontraded sector.11 We assume that the smaller the relative size of the import sector, the larger the
9
See, for example, Murphy (1984), Dooley et al. (1987), and Bayoumi and Rose (1993). Large countries are China, Japan and Korea. Medium countries are Thailand, Taiwan, Singapore and Hong Kong, while small countries include Indonesia, Malaysia and the Philippines. 11 We used the same proxy that Wong (1990) used for estimating the size of the non-traded sector. Countries with large non-traded sectors (i.e., have small share of import) are China, Japan and Indonesia. Countries with medium non-traded sectors are Thailand, Taiwan, Korea and the Philippines, while countries with small non-traded sectors include Singapore, Hong Kong and Malaysia. 10
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
289
Table 7 S–I correlation and country differences OLS
GLS
No control
Control
No control
Control
1.06 (0.10) 1.07 (0.16) 1.16 (0.13)
0.93 (0.11) 0.9 (0.19) 1.32 (0.17)
0.98 (0.09) 1.01 (0.13) 0.83 (0.08)
0.78 (0.09) 0.82 (0.13) 0.72 (0.09)
Size of non-traded sector Large 1.05 (0.08) Medium 1.35 (0.28) Small 0.8 (0.20)
0.80 (0.10) 1.49 (0.30) 0.72 (0.19)
1.08 (0.08) 1.04 (0.13) 0.58 (0.16)
0.83 (0.08) 0.98 (0.13) 0.45 (0.15)
Size of GDP Large Medium Small
Table 8 Aggregate saving–investment correlation All Big 3 ASEAN Greater China
1.25 1.02 1.73 1.11
(0.20) (0.20) (0.26) (0.21)
non-traded sector. The regression results before controlling for shocks are consistent with the theoretical predictions: the larger the non-traded sector is, the higher the S–I correlation. However, after controlling for cyclical effects, this positive relationship disappears. 4.5. Aggregate S–I correlation Table 8 reports the S–I correlation using aggregate saving and investment data of the sample Asian countries. Aggregate data are constructed by taking the sum of each country’s variables (adjusted for currency unit and inflation). Regressions based on the aggregate data may indicate how much investment of a certain region is financed by regional savings. That is, we treat the whole Asian region as one country and examine the S–I correlation. Note that this aggregate level correlation may not reflect the true degree of international capital mobility because we do not consider intra-regional borrowing and lending. Regressions with regional savings and investment data show high S–I correlation, suggesting that the Asian investment is largely financed by savings within the region. The table also shows that the S–I correlation is larger in the ASEAN countries compared to those in the big three or in greater China. This suggests that investment in the ASEAN countries is more likely to be financed by regional savings (less capital mobility), while investment in big three countries are relatively more financed by savings outside of the region (more capital mobility). 5. Conclusion In this paper, we have explored the S–I correlation of East Asian countries in relation to international capital mobility. Generally, the direction of changes in the S–I correlation over time is consistent with the changes in the degree of capital mobility; the S–I correlation decreases as
290
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
capital mobility increases over time. Additionally, the S–I correlation in East Asia is always less than that in the OECD countries in all periods. This is consistent with the fact that capital mobility in East Asia is lower than that among the OECD countries. On the other hand, even after controlling cyclical shocks, the S–I correlation remains positive, which may imply that the absolute degree of capital mobility is still low. Despite these findings, it is difficult to directly infer the degree of capital mobility of individual countries because the S–I correlation varies greatly across countries in a countryspecific analysis. This cross-country distinction may reflect either the presence of other countryspecific factors that are not considered in this paper or the small sample size. For example, if a government maintains a current account targeting policy, then the current account would not fluctuate to the optimal level and in turn, saving and investment would be highly correlated even in the presence of perfect capital mobility. Therefore, the S–I correlation may provide indirect evidence of a government’s stance on the current account policy, whether governments target the current account balance as their policy goal or simply allow for the current account as a residual of economic activity. Acknowledgement We are grateful for financial support from the Korea Institute of International Economic Policy. Appendix A Ten Asian countries are analyzed in this paper: China, Hong Kong, Indonesia, Japan, Korea, Malaysia, the Philippines, Singapore, Thailand, and Taiwan. We use annual data from 1980 to 2002. Saving is defined as the GDP minus government consumption and private consumption. Investment is gross fixed capital formation plus changes in stocks. Most of the national income accounting data is taken from the International Financial Statistics, IMF. The Asian Development Bank, Bank of Korea, Ministry of Finance in Japan and Taiwan are additional sources of data. We convert local currency data into the US dollar using period average exchange rates and nominal values into real values using GDP deflator. Productivity shocks are defined as annual percentage changes in productivity. For productivity measure, we use Solow residuals. The share of labor in manufacturing output is assumed to be 0.6. Industrial production data has been used for output and the employment data for labor input. Fiscal shocks are defined as percentage changes in unexpected government spending. The use of government spending data for fiscal shocks may cause an endogeneity problem with other national income accounting variables. We construct the data for the unexpected government spending by assuming that policymakers determine the growth rate of government spending at the start of a fiscal year considering the predicted annual GDP growth rate and the net government debt at the start of the year. Government debt data are taken from the IFS (line 80, net government surplus/deficit in national currencies). We assume that policymakers have perfect foresight and run the country-by-country OLS regression of the growth rate of real government spending at time t on the real GDP growth rates and the real net government debt at the start of the year. We use the residuals of these regressions for the unexpected government spending data. Terms of trade (TOT) shocks are defined as the percentage changes in the TOT—export price/import price. Export and import prices are unit values of exports and imports taken from IFS (lines 74, 75), respectively. For Indonesia and Malaysia, we use only export prices as import price index data are
S. Kim et al. / Japan and the World Economy 19 (2007) 279–291
291
not available. For Malaysia, export price index is constructed by taking the weighted average of three main export goods’ prices, rubber, tin and palm. References Backus, D., Kehoe, P., Kydland, F., 1995. International business cycles: theory and evidence. In: Cooley, T. (Ed.), Frontiers of Business Cycle Research. Princeton University Press, Princeton, NJ. Baxter, M., Crucini, M., 1993. Explaining saving–investment correlations. American Economic Review 83, 416–436. Bayoumi, T., 1989. Saving-investment correlations: immobile capital, government policy or endogenous behavior? IMF Working Paper 66. Bayoumi, T., Rose, A., 1993. Domestic savings and intra-national capital flows. European Economic Review 37, 1197– 1202. Coakley, J., Kulasi, F., Smith, R., 1998. The Feldstein–Horioka puzzle and capital mobility: a review. International Journal of Finance and Economics 3, 169–188. Dooley, M., Frankel, J., Mathieson, D., 1987. International capital mobility: what do saving–investment correlations tell us? IMF Staff Papers 34, 503–529. Feldstein, M., 1983. Domestic saving and international capital movements in the long run and the short run. European Economic Review 21, 129–151. Feldstein, M., Horioka, C., 1980. Domestic saving and international capital flows. Economic Journal 90, 314–329. Feldstein, M., Bacchetta, P., 1991. National saving and international investment. In: Bernheim, B., Shoven, J. (Eds.), National Saving and Economic Performance. University of Chicago Press, Chicago, IL. Glick, R., Rogoff, K., 1995. Global versus country-specific productivity shocks and the current account. Journal of Monetary Economics 35, 159–192. Golub, S., 1990. International capital mobility: net versus gross stocks and flows. Journal of International Money and Finance 9, 424–439. Kim, S.H., 2001. The saving–investment correlation puzzle is still a puzzle. Journal of International Money and Finance 20, 1017–1034. Mendoza, E., 1991. Real business cycles in a small open economy. American Economic Review 81, 797–818. Murphy, R., 1984. Capital mobility and the relationship between saving and investment in OECD countries. Journal of International Money and Finance 3, 327–342. Obstfeld, M., 1995. International capital mobility in the 1990s. In: Kenen, P. (Ed.), Understanding Interdependence: The Macroeconomics of the Open Economy. Princeton University Press, Princeton, NJ. Sinn, S., 1992. Saving–investment correlations and capital mobility: on the evidence from annual data. Economic Journal 102, 1162–1170. Taylor, A.M., 1994. Domestic saving and international capital flows reconsidered. NBER Working Paper No. 4892. Tesar, L., 1991. Savings, investment and international capital flows. Journal of International Economics 31, 55–78. Tesar, L., 1993. International risk sharing and non-traded goods. Journal of International Economics 31, 55–78. Wong, D., 1990. What do saving–investment relationships tell us about capital mobility? Journal of International Money and Finance 9, 60–74.