【文章內(nèi)容簡介】 can be seen from Figures 4(a–c), the three series appear to be nonstationary in level form. Therefore, we investigate the stationarity of the first difference of the three series by testing for unit roots. The ADF tests are performed on both the level and first differenced observations by estimating the following three models: ? No constant and no trend model: tki ititt yryy ?? ????? ?? ?? 11 （ 1） ? Constant and no trend model: titki itt yryy ??? ?????? ??? ?110 （ 2） ? Constant and trend model: titki itt yryty ???? ?????? ??? ?1120 （ 3） The results of the ADF test are shown in Table 2. They show that the null hypothesis of a unit root is: (a) accepted for the level series of FDI in all three models。 (b) rejected for the level series of DI in model (3), and (c) rejected for the level series of GDP in model (1). 6 Based on the first differenced data, the results indicate that all three series are stationary. Therefore, we conclude that the three time series are all integrated of order 1, I(1). b. Testing for Cointegration of Variables Now, the cointegration test is performed to investigate any longterm equilibrium relationships among the three variables of FDI, DI and GDP. After a careful search and trial, a model with six lags, constant and centred seasonal dummy variable was chosen. The result of the Johansen cointegration rank test is summarised in Table 3, which indicates the presence of two cointegrating vectors at 1 per cent and 5 per cent levels of significance, respectively (. The null hypotheses of no cointegration is rejected for the rank of zero and less than or equal to 2). This means that there exists a longterm relationship among the three variables. c. The Error Correction Model To analyse the causal relationship between the three variables FDI, DI and GDP, we use an error correction model (ECM) of the following VAR system: When applied to the Chinese data, the VAR system performs quite well. As reported none of the diagnostic statistics are significant at the 95 per cent critical value. Therefore, there is nothing to suggest that the system model is incorrectly specified. Based on the Schwarzz (1978) and Akaike (1974) information criteria, the number of lags is chosen as six. d. Innovation Accounting and theGranger Causality Test The innovation accounting (variance deposition and impulse response function) technique can be utilised to examine the relationships among economic variables . Using this technique, Kim and Seo (2021) explored the plementary or substitution relationship between FDI and domestic investment, and analysed the impact of FDI on economic growth in South Korea. On the other hand, the forecast error variance deposition allows us to make inferences about the proportion of movements in a time series due to its own shocks 7 versus shocks to other variables in the system (Enders, 1995, p. 311). These results suggest that the strength of the relationships between FDI, domestic investment and economic growth are different. FDI plays an important role in China’s economic growth but its influences are less than that of domestic investment ( per cent versus per cent). GDP shows stronger influences on China’s domestic investment than FDI does ( per cent versus per cent). The influences of DI and GDP on FDI are relatively low ( per cent and per