RDP 2007-11: Global Imbalances and the Global Saving Glut – A Panel Data Assessment 7. Sensitivity Analysis

7.1 The Exchange Rate

Dooley et al (2004a) argue that some countries (such as China) have undervalued their exchange rate in order to promote growth in the traded sector, and view the substantial increases in foreign exchange reserves as a consequence of this policy.^[25] This has not been accounted for explicitly in our model, although the financial crisis dummy variables may partially capture the increase in reserves. We experimented with adding the real exchange rate to the model as a deviation from a statistical trend estimate, and included one lag (see Appendix C for the model results).^[26] The performance of the model appears to improve for several regions for particular periods (compare Figures 6 and 7) – such as the major oil exporters during 1996, China from 2003 onwards, and the US from 1999–2001. However, the overall improvement to the fit of the model from including the real exchange rate is small.^[27]

7.2 Including China in the Asian Crisis

The dating of financial crises, described in Section 4.1.1, does not include China in the Asian crisis. However, China's reserve assets as a share of GDP declined slightly from 1997 to 2000, which may have reflected pressure on its fixed exchange rate due to contagion from other crisis countries (Figure 1). After 2000, China's reserve assets increased strongly, possibly as authorities attempted to increase investor confidence. Consequently, as the Asian crisis appears to have influenced China's foreign reserve accumulation, we re-estimate the model assuming that the Chinese economy responded as if it had been one of the crisis economies in 1997. The impact of this change on the coefficient estimates is minimal (see Appendix C), and for the fitted values is limited mainly to China itself during the crisis (Figure 8). The estimated total capital outflow from crisis-affected countries during the Asian crisis does increase, but remains small overall.

Footnotes

Bergin and Sheffrin (2000) extend the intertemporal model to include the real exchange rate. [25]

We use a Hodrick-Prescott (1997) filter to isolate the cyclical component. We set the smoothing parameter to 400, which yields a very smooth trend estimate. Similar results were obtained with lower values. [26]

The real exchange rate may be endogenous. If we instrument it with longer lags (which may be weak instruments), the fitted values display the same broad trends as in Figure 7, but are more volatile. [27]