《Construction Investment, Other Investment and GDP in China.doc》由会员分享，可在线阅读，更多相关《Construction Investment, Other Investment and GDP in China.doc（12页珍藏版）》请在三一文库上搜索。
1、Construction Investment, Other Investment and GDP in ChinaSubmitted to the Seventh AsRES Conference, Seoul, Korea, July 2002BySiqi Zheng and Hongyu LiuInstitute of Real Estate Studies, Tsinghua UniversityBeijing, ChinaConstruction Investment, Other Investment and GDP in ChinaZheng Siqi, Liu Hongyu (
2、Institute of Real Estate Studies, Tsinghua University, Beijing 100084)AbstractThis paper examines the interactions between construction investment (CI) and GDP, other investment (OI) and GDP respectively, based on Granger causality analysis. Firstly the Augmented Dickey Fuller (ADF) technology is us
3、ed to study the integration of these three time series and the cointegration relationships between CI and GDP, OI and GDP are examined. With Error Correction Model (ECM), Granger causality test is then performed to find the causality relationships between GDP and CI as well as GDP and OI. We find th
4、at CI, OI and GDP are all have unit roots and there exists cointegration relationships between CI and GDP, OI and GDP. We also find that CI while not OI Granger-causes GDP in short run; GDP Granger-causes CI and OI in long run. We also study the effects of housing investment and non-housing investme
5、nt on GDP, respectively. These interesting empirical results will benefit the policy markers in China. Key Words: GDP; Construction Investment; Other Investment; Cointegration; Error Correction Model1. BACKGROUNDNow the economy in China is still in its embryo period, which has a typical characterist
6、ic of element driving. Both the economic and final consumption growths rely on the increase of investment. Although investment and consumption are two main elements that promote the economic growth, the effect of investment apparently is more important in China presently. So the rise of investment c
7、an drive the growth of national economy. However, high-speed development of investment cannot sustain without corresponding economy growth, because persistent and healthy growth of economy is the powerful support of investment. Therefore interaction between investment and national economy exists.Gen
8、erally speaking, investment in construction is a powerful driving force of national economy. It has been used as an effective way to stimulate economy in its depressed period by almost all the governments. Expanding the scale of investment in construction may have several positive effects, not only
9、providing new job opportunities, increasing domestic demand and the national income, but also prompting the economy growth and providing basis for the next cycle. The mainstream literature also demonstrated that investment in construction plays a very important role in economy growth, especially in
10、the developing countries whose construction investment even occupy a great portion as 20 percents of its whole investment（Kessedes,1995）. In China, a number of studies have discussed the impacts of construction investment on national economy, but they still remained as qualitative analysis. Little h
11、as been studied quantitative relationship among the construction investment, other investment and GDP. Therefore, the attempt of this study is to highlight the quantitative relationships between construction investment (CI) and GDP, other investment (OI) and GDP respectively from long run and short
12、run aspects.2. METHODOLOGY1.1 VariablesThe variables used in this research include GDP, construction investment (including installation investment) and other investment. GDP refers to the final products of all resident units in a country (or a region) during a certain period of time; therefore, we u
13、se it here as a parameter reflecting the development of national economy.Investment in fixed assets refers to the volume of activities in construction and purchases of fixed assets in monetary terms, which may be divided into three parts, i.e. construction and installation, purchase of equipment and
14、 instrument, and other expenses. In this article we use construction and installation investment (CI) as the variable to measure the construction activities, which refers to the construction of various buildings and infrastructures and installation of equipments and/or instruments on them. The rest
15、two parts of investment in fixed assets are defined as other investment (OI) in this paper.1.2 Literature ReviewWhat we try to explore here are the causality relationships between CI and GDP, OI and GDP. In econometrics, Granger definition of causality is the most widely accepted definition of causa
16、lity. According to Granger (1969), Y is said to “Granger-cause” X if and only if X is better predicted by using the past values of Y than by not doing so with the past values of X being used in either case. In short, if a scalar Y can help to forecast another scalar X, then we say that Y Granger-cau
17、ses X. If Y causes X and X does not cause Y, it is said that unidirectional causality exists from Y to X. If Y does not cause X and X does not cause Y, then X and Y are statistically independent. If Y causes X and X causes Y, it is said that feedback exists between X and Y. Essentially, Grangers def
18、inition of causality is framed in terms of predictability.Granger (1969) originally suggested the Granger test, which was improved by Sargent (1976). To implement the Granger test, we assume a particular autoregressive lag length k (or p) and estimate Equation (1) and (2) by OLS: (1) (2)F test is ca
19、rried out for the null hypothesis of no Granger causality where F statistic is the Wald statistic for the null hypothesis. If the F statistic is greater than a certain critical value for an F distribution, then we reject the null hypothesis that Y does not Granger-cause X (equation (1), which means
20、Y Granger-causes X.A time series with stable mean value and standard deviation is called a stationary series. If d differences have to be made to produce a stationary process, then it can be defined as integrated of order d. Granger (1981, 1983) proposed the concept of cointegration, and Engle and G
21、ranger (1987) made further analysis. If several variables are all I(d) series, their linear combination may be cointegrated, that is, their linear combination may be stationary. Although the variables may drift away from equilibrium for a while, economic forces may be expected to act so as to restor
22、e equilibrium, thus, they tend to move together in the long run irrespective of short run dynamics. The definition of the Granger causality is based on the hypothesis that X and Y are stationary or I(0) time series. However, it is now widely recognized that many macro-economic series appear to conta
23、in a (or at least a) unit root in their autoregressive representations, and there are plenty of empirical evidences to indicate that macro-economic series often appear to be I(1). So we can not apply the fundamental Granger method for variables of I(1).The classical approach to deal with integrated
24、variables is to difference them to make them stationary. Hassapis et al. (1999) show that in the absence of cointegration, the direction of causality can be decided upon via standard F-tests in the first differenced VAR. the VAR in the first difference can be written as: (3) (4)However if both Yt an
25、d Xt are truly I(1) and cointegrated, the bivariate dynamic relation between Y and X will be misspecified if the researcher just simply differences both Y and X. According to Engle and Granger (1987), the test must be carried out with error-correction models (ECM). They proved that any cointegrated
26、series must have an error correction representation, and the converse is also the truth, which cointegration is a necessary condition of error correction models (ECM). Here are the error correction representations: , (5), (6)Where (i=1, 2) is error-correction (EC) term(s). and are called coefficient
27、s of adjustment and one of them must not be equal to zero according to Engle and Granger (1987). In Equation (5) and (6), all series are I(0) processes.An ECM representation is really a restricted VAR with co-integration specification. So it is designed for the non-stationary series known to be co-i
28、ntegrated. The parameters in the ECM have clear interpretations. In Equation (5), the coefficient of Y in the EC term () is the long-run elasticity of X with respect to Y. Conversely, in Equation (6), the coefficient of X in the EC term () is the long-run elasticity of Y with respect to X. and clear
29、ly reflect the immediate response of X to changes in Y and the immediate response of Y to changes in X respectively. They are therefore the short-run elasticities (Thomas, 1997). In Equation (5), the larger the parameter f1, the faster adjustment of X to the previous periods deviation from long-run
30、equilibrium. At the opposite extreme, very small values of f1 imply that X is unresponsive to the last periods equilibrium error. The same condition exists in equation (6). Since the ECM and cannot at the same time equal to zero as the result of the presence of the cointegrating relationship, there
31、must exist one direction of long-term causality between Y and X. Standard t test can be used to test the significance of and .We use augmented Dickey-Fuller (ADF) method to test the order of the series, and Johansens method to test if cointegrating relationship exists between the series.In this rese
32、arch we use the path as shown in Figure 1 to test the existence and direction of causality between X and Y:Are X and Y both stationary?X and Y are all I (0)X and Y are all I (1)Use equations (1) and (2) to test the causality relationshipX and Y are not cointegrated.X and Y are cointegrated.Use equat
33、ions (3) and (4) to test the causality relationshipUse equations (5) and (6) to test the causality relationshipFigure 1. Granger Causality Tests3. DATA3.1 Data in TableThe data used in the research comes from the National Bureau of Statistics of China. All three time series (GDP, CI, OI) are estimat
34、ed at constant market prices of 1981. Yearly series are applied in the paper from 1981 to 2000, so result in 20 observations for each variable (Table 1).In addition, all series are transformed to natural logarithms, which has two reasons. Firstly, in an economy with high-speed growth, it is more con
35、venient to use natural logarithms values to estimate the model; secondly, the first difference of natural logarithms value becomes the proportional rate of growth, which is because:, therefore .Table 1GDP, CI and OI in China（From year 1981 to year 2000, the original units of GDP, CI and OI are all 1
36、00 million RMB）YearLnGDPLnCILnOIYearLnGDPLnCILnOI19818.489 6.536 5.603 19919.415 7.516 6.889 19828.576 6.751 5.865 19929.547 7.802 7.231 19838.680 6.867 6.046 19939.674 8.127 7.606 19848.821 7.043 6.361 19949.793 8.186 7.641 19858.948 7.262 6.638 19959.893 8.368 7.232 19869.032 7.417 6.754 19969.984
37、 8.285 7.624 19879.142 7.531 6.899 199710.069 8.290 7.775 19889.249 7.583 6.955 199810.144 8.433 7.904 19899.289 7.383 6.634 199910.212 8.498 7.967 19909.327 7.357 6.667 200010.289 8.582 8.076 Notes: 1. All data are in constant market prices (1981). 2. Data source: China Statistical Yearbook, 2001.3
38、.2 Tests of IntegrationGenerally speaking, economic series of flow variables estimated in constant price are always I(1). So intuitively we can make an assumption that LnGDP, LnCI and LnOI are all I(1). The results of ADF tests prove it. As an example, we only list the test for LnOI.Estimate the fol
39、lowing equation by OLS: （7）The t statistic value of is 2.927, whose absolute value is not greater than the critical value of 3.60. Let , the t statistic value of is 0.5, whose absolute value is smaller than the critical value of 2.85, so there does not exist the time term. Now estimate the following
40、 equation by OLS: （8）The t statistic value of is 3.070, whose absolute value is greater than the critical value of 3.00. So we can reject the null hypothesis that has a unit root, then is I (1). The test of LnGDP and LnCI are similar with this, and they are all I (1).3.3 Test of Cointegration3.3.1 L
41、nGDP and LnCIFirstly we use OLS to get the equation of long-run equilibrium: （9） (8.438) (22.092) Adjusted R2=0.962 F=488.035So error correction term is . Then we can test if ecm is stationary.Estimate the equation: （10）The t statistic of is 3.810, whose absolute value is greater than the critical v
42、alue of 3.60. The null hypothesis that has a unit root then is rejected which indicates that is stationary. So LnGDP and LnCI are cointegrated. 3.3.2 LnGDP and LnOISimilarly we use OLS to get the equation of long-run equilibrium: （11） (12.349) (16.274) Adjusted R2=0.933 F=264.856Error correction ter
43、m is . Then we test if ecm is stationary:Estimate the equation: （12）The t statistic of is 3.630, whose absolute value is greater than the critical value of 3.60. The null hypothesis is rejected and is stationary. So LnGDP and LnOI are also cointegrated.Since LnGDP and LnCI, LnGDP and LnOI are both c
44、ointegrated, we are able to use equation (5) and (6) to test the existence and direction of causality between them.4. MODELS AND RESULTS4.1 GDP and CIFirst we need to determine the lag length. Because of the limitation of macroeconomic series in China, Lag Length is determined as 2 in this paper. We
45、 estimate the equations (13) and (14) using OLS, in which is from equation (9): （13） （14）The results are showed in Table 2:Table 2: Statistic Results of Causality Test between GDP and CICIGDP，Equation（13）GDPCI，Equation（14）F4.142Sig.0.023F4.188Sig.0.022Independent VariableCoefficientstSig.Independent
46、 VariableCoefficientstSig.0.2460.7660.4600.6142.1470.055-0.698-2.2460.0460.4301.5040.1610.1081.6510.1271.1520.8350.4220.1021.5410.152-0.484-0.3560.7290.0050.0570.956-1.129-3.0080.0124.1.1 CIGDPFrom Table 2, we can see that equation (13) can hold under the confidence level of 95%, so the influence of CI to GDP is remarkable, and CI Granger-causes GDP.Then we should explore whether the influence of CI to GDP is in short run or long run. It can be seen that and are both prominent (can hold under the confidence level of 85%) and the coefficient of is approximately zero, and probability of fai