Capital flows, economic growth and the real effective exchange rate: Evidence from China

Abstract

This paper examines the Granger causal relationship between capital flows and economic growth in China over the period 1998Q1–2019Q2, allowing for real effective exchange rate (REER) effects. As parameter instability tests indicate structural changes, we use bootstrap rolling window causality tests, which suggest that the causal nexus between capital flows and GDP growth is time-varying. We find that the causal links between foreign direct investments (FDIs) and GDP growth are hardly affected by the REER, whereas the REER plays a more important role in affecting the causal connections between portfolio investments and other investments and GDP growth. Our results suggest that cumulative portfolio inflows and cumulative other investment inflows harm GDP growth, whereas cumulative portfolio outflows and cumulative other investment outflows positively affect GDP growth.

Keywords:

• Capital flows
• real effective exchange rate
• GDP growth rate
• bootstrap Granger causality test
• parameter instability test
• time-varying causality

Acknowledgements

We thank Jan Jacobs for econometric guidance and helpful discussions and three anonymous referees for very helpful comments on a previous version of the paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Data source: http://www.safe.gov.cn/en/BalanceofPayments/index.html, and the authors’ calculations.

2 Data source: https://data.worldbank.org/indicator/NY.GDP.MKTP.KD.ZG?locations=CN, and the authors’ calculations.

3 We also tested window lengths of 30 and 40 quarters; the results (available upon request) are similar.

4 Impulse response analysis is an important tool in the VAR system analysis. However, most types of capital flows are I (0) and the REER and GDP growth rate are I (1) in levels. Conflicting time-series properties hamper the implementation of impulse responses. Therefore, impulse responses are not included in the paper.

5 The quality of Chinese data has been questioned. However, Chow (2006) finds that official Chinese data are reasonably accurate and reliable for use in many macroeconomic analyses. Sinclair (2019) reports that revisions to the GDP level and growth in official Chinese data are similar to GDP revisions in the US.

6 The QFII programme allows licensed global institutional investors to invest in China’s Shanghai and Shenzhen stock markets.

7 A panda bond is a RMB-denominated bond issued in China’s mainland by a foreign entity.

8 The QDII programme allows certain domestic investors to invest in foreign securities markets.

9 For more information about the Bond Connect, please see https://www.chinabondconnect.com/en/index.html.

10 The weights are derived from manufacturing trade flows and capture both direct bilateral trade and third-market competition by double-weighting (Klau and Fung 2006).

11 Please see the PBC’s announcement at http://www.pbc.gov.cn/english/130721/2941603/index.html.

12 However, the result changes when we allow for a structural break based on the method suggested by Perron (1989). The null of Perron (1989) is that the GDP growth rate has a unit root with no break. Our results reject the null at the 5% significance level when the changes are assumed to take place gradually (the Innovational Outlier Model) and when changes are assumed to take place rapidly (the Additive Outlier Model). Therefore, we conclude that China’s GDP growth rate is stationary with one break. The break occurred around 2007–2008, suggesting that the structural break in the Chinese GDP growth rate is associated with the 2007–2008 global financial crisis. However, previous research has not addressed the issue of how stationary series with structural breaks affect the Granger causality. Our methodology is based on the Toda and Yamamoto (1995) method, which can better deal with a VAR system with non-stationary series. We employ bootstrap techniques to produce critical values that are much more accurate than the asymptotic ones regardless of the order of integration (Hacker and Hatemi 2006). Given the existence of a structural break, the relationship between capital flows and the GDP growth rate may show instability across different subsamples. We therefore apply subsample rolling window Granger causality tests, which allows the Granger causality to vary over time.

13 The null hypothesis can only be rejected at the 10% significance level.

14 Full results are available upon request.