Foreign capital and economic growth

Brookings Papers on Economic Activity

Volumn , Issue 1, 2007, Pages 153-209

  Prasad, Eswar S ^a   Rajan, Raghuram G ^b   Subramanian, Arvind ^c  

^a Department of Obstetrics Gynecology

^b UNIVERSITY OF CHICAGO   (United States)

^c PETERSON INSTITUTE FOR INTERNATIONAL ECONOMICS   (United States)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 34548213934     PISSN: 00072303     EISSN: None     Source Type: Journal    
DOI: None     Document Type: Conference Paper

References (71)

1. Keynes (1920, p. 11).
2. Keynes (1933).
3. A current account surplus has to equal the sum of the following: net private and official outflows of financial capital (this includes debt and nongrant aid, but not remittances, which should properly be reflected in the current account itself); net errors and omissions (a positive number could, for instance, represent capital flight through unofficial channels); and net accumulation of international reserves by the government (typically the central bank). Thus the current account surplus summarizes the net amount of capital flowing out of the country in a given period or, equivalently, the excess of domestic saving over domestic investment in that period; correspondingly, a current account deficit summarizes net capital flowing in or, equivalently, the excess of domestic investment over domestic saving.
4. See Obstfeld and Taylor (2004) for example.
5. See, for example, Bernanke (2006).
6. Lucas (1990).
7. Excluding the oil-exporting countries does not alter the basic patterns in figure 2 (not shown). We also constructed similar graphs using initial (1970) relative income, rather than relative income in each period, in order to take out the effects of income convergence. This, too, makes little difference to the shapes of the plots.
8. Alfaro, Kalemli-Ozcan, and Volosovych (2005).
9. Hsieh and Klenow (2003); Caselli and Feyrer (2007).
10. Gertler and Rogoff (1990); Reinhart and Rogoff (2004).
11. Of course, more-rapid growth could imply greater factor employment and even a lower marginal productivity of capital. However, there is a positive cross-sectional correlation between GDP growth and the Bosworth-Collins (2003) measure of total factor productivity growth (based on the updated version of their dataset that goes through 2003) for the nonindustrial countries in our dataset. Caselli and Feyrer (2007) have constructed a measure of the marginal product of physical capital that corrects for the share of natural capital (land) in the total capital stock of each country and for differences in the relative price of capital across countries. For the countries that are common to our dataset and theirs, average GDP growth is strongly positively correlated with the Caselli-Feyrer measure. This suggests that high-growth countries do have more attractive investment opportunities.
12. Gourinchas and Jeanne (2006a); the same authors also provide evidence of a negative correlation between capital inflows and investment rates.
13. Henry (2006) argues correctly that the financing provided by foreign capital can have permanent effects on the level of income but only temporary effects on its rate of change. But for the not-so-long horizons examined in this paper, and given how far developing countries are from their steady states, transitional and permanent effects are probably indistinguishable in the data, making the growth effects from additional investment a reasonable focus of inquiry.
14. The sample differs from that of Bosworth and Collins in that it omits Bangladesh, Guyana, and Taiwan; the countries are listed in appendix table A-1.
15. A more negative current account balance indicates larger net inflows of foreign capital. A positive current account balance indicates a net outflow of capital.
16. The Lowess procedure estimates a locally weighted regression relationship between the dependent variable and the explanatory variable. It thus allows us to estimate a smoothed, nonparametric relationship between the two.
17. See Edwards (2005) and Glick, Guo, and Hutchison (2006).
18. Kose, Prasad, and Terrones (2006).
19. Henry (2006) and Kose and others (2006) provide surveys.
20. Houthakker (1961), Modigliani (1970), and Carroll and Weil (1994).
21. Aghion, Comin, and Howitt (2006).
22. Gourinchas and Jeanne (2006b).
23. Aizenman, Pinto, and Radziwill (2004).
24. This discussion draws upon Rodrik (2006).
25. Kose and others (2006) review these measures and argue that, since de jure ones cannot capture the enforcement and effectiveness of capital controls, they may not be indicative of the true extent of financial integration. Actual capital flows may be more relevant for examining the role of foreign capital in the growth process.
26. Bosworth and Collins (2003). Ourr work is also related to their earlier paper (Bosworth and Collins, 1999).
27. Sachs and Warner (1995).
28. These stock measures have been constructed by Lane and Milesi-Ferretti (2006).
29. See Bosworth and Collins (2003), who argue that growth in the capital stock is a better measure than the investment-GDP ratio for the purposes of growth accounting and regressions.
30. We test in appendix table A-2 whether there is a relationship between financial integration and growth, using the measures of integration that have conventionally been used in the literature. We find, consistent with Kose and others (2006), no relationship, in our sample of countries, either between GDP growth and the level of financial openness, whether measured by stocks or by flows, or between GDP growth and changes in these measures. There is weak evidence that FDI, which is qualitatively different from other flows in bringing in technology, is positively correlated with growth (see Borensztein, De Gregorio, and Lee, 1998). We also tested whether the trade balance (as opposed to the current account balance) is the prime driver (results are available from the authors). It turns out that the trade balance, defined as net exports of goods and nonfactor services, is positively correlated with growth, but not statistically significantly so, and the magnitude of the correlation is smaller than that between the current account balance and growth. Clearly, there are elements in the current account balance (including factor incomes and transfers) that add to its explanatory power. For nonindustrial countries, these items can be quite large.
31. Abiad, Leigh, and Mody (2007) find that current account balances are negatively correlated with growth among European countries, including a small group of transition countries. Their work is useful in pointing out that the correlation for transition economies is different from that for other nonindustrial economies, a fact we verify above.
32. These are growth spurts that occurred after 1970 and were followed by sustained growth, as identified by Hausmann, Pritchett, and Rodrik, (2005).
33. This is not to say that all forms of foreign finance fall during growth spurts. Indeed, the average ratio of FDI to GDP rises from an annual average of 0.2 percent in the five years before the initiation of a growth spurt to 0.7 percent in the five years after. Similarly, using the episodes of growth decelerations identified by Jones and Olken (2005), we find that the average FDI-GDP ratio falls from 1.7 percent in the five years before the deceleration to 1 percent in the five years after. But even these increases and decreases are small compared with the changes in domestic saving following a growth spurt or deceleration.
34. One version of the life cycle model applied to countries has implications for the evolution of current account balances (see the discussion in Chinn and Prasad, 2003). According to this theory, poor countries that open up to foreign capital early in the development process should run current account deficits as they import capital to finance their investment opportunities. Eventually, these countries would become relatively capital rich and begin to run trade surpluses, in part to pay off the obligations built up through their accumulated current account deficits.
35. GMM estimators come in two flavors. There is the difference-GMM estimator of Arellano and Bond (AB; 1991 ) and the system-GMM estimator of Blundell and Bond (BB; 1998). In both, identification relies on first-differencing and using lagged values of the endogenous variables as instruments. In the AB estimator, lagged levels are used to instrument for the differenced right-hand-side variables, whereas in the BB estimator, the estimated system comprises the difference equation instrumented with lagged levels as in the AB estimator as well as the level equation, which is estimated using lagged differences as instruments. Each estimator has its limitations. The AB estimator often leads to a weak-instruments problem because lagged levels are typically not highly correlated with their differenced counterparts. So, in what follows, we present estimations based on the BB estimator. All specifications include time effects to control for common shocks.
36. One methodological point bears mentioning. GMM procedures allow a fair amount of freedom, especially in specifying the lag structure for the instruments. There is a trade-off: the greater the lags, the more the information that is used. But greater lags can lead to overfitting and weak instrumentation. Two key diagnostics to use in checking for these problems are the Hansen test for overidentifying restrictions and the Arellano-Bond test for serial correlation. When we used the second lag, our results were stronger than reported in the text, but there were occasional problems of overfitting, reflected in very large p-values for the Hansen test. We therefore report results using the third and fourth lags, which are more reassuring in relation to these two diagnostics.
37. We cannot include data for the transition countries in the panel regressions, as our estimation procedure requires data for at least four time periods for a country to be included in the sample.
38. Bernanke and Gürkaynak (2002) report a positive correlation between productivity growth and saving in a broad sample of countries - they do not break their sample out into different groups of countries based on income.
39. Carroll and Weil (1994), for instance, show that habit persistence may be one way to reconcile the strong positive correlation between saving and growth, a correlation that runs counter to the predictions of the standard life cycle or permanent income hypothesis. Jappelli and Pagano (1994) build a model showing how financial market imperfections that limit the ability to borrow against future income could generate a correlation between saving and growth in a fast-growing economy with a low level of financial development.
40. Wurgler (2000) provides evidence that underdeveloped financial sectors are unable to reallocate resources to their highest-productivity uses, leading to a mismatch between productivity increases and investment.
41. See Rajan and Zingales (1998).
42. Caballero, Farhi, and Gourinchas (2006).
43. In truth, many developing country households (for example, in China) have been accumulating domestic financial assets in the form of bank deposits. The final holder of foreign assets is often the government, not households. One could argue that households are willing to hold bank deposits only because banks hold central bank paper, which is eventually a claim on foreign bonds, but this seems a tenuous line of reasoning.
44. Glick and Rogoff (1995) showed that country-specific productivity shocks tend to generate investment booms and larger current account deficits (or smaller surpluses) in what were then the Group of Seven leading industrial countries.
45. Gourinchas and Jeanne (2006a).
46. Kraay and Ventura (2000). Their argument is based on the intuition that the marginal portfolio allocation decision (how to invest the extra saving generated by income shocks) will resemble the average decision (reflected in the existing net liability stock) unless investment risk is low and domestic investment is highly subject to diminishing returns.
47. Rajan and Zingales (1998).
48. Rajan and Zingales (1998) describe how they calculate the number for the period 1980-89. We calculate a similar number using U.S. corporate data between 1990 and 1998 (after 1998, normal financing behavior would be contaminated by the equity bubble). In computing each industry's dependence on finance for 1990-98, we first compute the dependence on finance of each firm in the industry over the period, truncate outlier firms at the 10th and 90th percentiles, and then average across all firms. We then take the average of the industry's dependence for the 1980s and the 1990s to get our final measure.
49. Rajan and Zingales (1998).
50. Chinn and Ito (2006).
51. To reduce the effect of data errors, all variables are "winsorized" at the 99 percent and the 1 percent level. Standard errors are robust, and we report the estimates when we cluster by country. Results are qualitatively similar when we cluster by industry. These results are available from the authors upon request.
52. Kose and others (2006).
53. The index was constructed by De Nicoló, Laeven, and Ueda (2006).
54. See Chinn and Ito (2006) and Alfaro and Hammel (2007).
55. Relative to the earlier specification, we drop the industry's initial share of manufacturing and the interaction of industry dependence on finance with the country's corporate governance index. The initial share of manufacturing should be absorbed in the industry x country indicator, and the interaction is not meaningful since neither the corporate governance index nor dependence on finance varies across time. Note that in this panel specification the openness to capital flows varies across time and countries, whereas dependence on external finance varies across industries, which, in the presence of industry-country fixed effects, allows identification within country, within industry, and across time.
56. The coefficient on the interaction in the panel is negative also for countries with above-median levels of financial development, unlike in the cross-sectional results. One interpretation of this is that the benefits of foreign capital accrue even to financially well-developed countries only in the medium run.
57. Detragiache, Tressel, and Gupta (2006) show that, in poor countries, a stronger foreign bank presence is robustly associated with less credit to the private sector in both cross-sectional and panel tests. In addition, in countries with more foreign bank penetration, credit growth is slower and there is less access to credit. By contrast, they find no adverse effects of foreign bank presence in more advanced countries. Tressel and Verdier (2007) show that, in countries with weak institutions, financial integration leads to greater investment by politically connected firms, with a loss of efficiency. Our findings are not inconsistent with these results.
58. This argument does not, of course, detract from the possibility that foreign capital has large indirect benefits, including on financial development itself. Some authors point to the beneficial effects of equity market liberalization on growth (for example, Bekaert, Harvey, and Lundblad, 2005, and Henry, 2006). In addition to the problem of timing that the literature notes - such liberalization is typically part of broader macroeconomic reforms that affect outcomes - the countries that liberalize might be the same ones that are typically able to reap the benefits from foreign finance, in part because they have stronger financial sectors. For this reason, our findings need not be inconsistent with the more positive tone of the equity market liberalization literature.
59. i). We average this measure for each country over the relevant period. This measure is also used by Rajan and Subramanian (2005).
60. We could run the same regression in a panel context, but there is more reason to expect the real exchange rate to be decoupled from capital flows in the short run; countries can use sterilized intervention, fiscal policy, and other measures to retain influence over the real exchange rate. Unless we can control for these short-run policies, it would be difficult to identify the effect of flows on overvaluation.
61. One qualification to this result is that, when we use the current account-GDP ratio in place of private capital inflows, we do not find a statistically significant relationship with our measure of overvaluation, either in the cross section or in the panel. There is a huge endogeneity problem in such regressions, of course, which could explain this in the context of non-industrial countries. Systematic undervaluation could stimulate speculative inflows through unofficial channels when there are selective capital controls in place; similarly, overvaluation may lead to capital flight. (Both these unofficial inflows and outflows would be reflected in the errors and omissions category of the balance of payments.) This is why measures of private capital inflows may be more relevant for understanding the effects of net flows on exchange rates. There is an endogeneity problem in this case as well, but it should drive the correlations that we report in table 6 negative (more overvaluation reduces inflows of private inflows through official channels). Hence the positive correlations that we find are still interesting.
62. Although this particular specification is sensitive to the inclusion of Mauritius, in others, where the Africa dummy is dropped, the result is more robust.
63. Alternative lag structures yield a significant coefficient on the overvaluation term. 64. Since the overvaluation term is instrumented in the panel, reverse causation should be less of a concern. See also Razin and Collins (1999).
64. Again, as identified by Hausmann, Pritchett, and Rodrik (2005).
65. It is less easy to run these regressions in a panel context because the exportability index exhibits virtually no time variation, and the overvaluation variable is also quite persistent across the two decades. So there is very little time variation to enable identification.
66. Rajan and Subramanian (2005).
67. See, for example, Dooley, Folkerts-Landau, and Garber (2004a, 2004b). Why, for example, would Korea or Taiwan be comforted, when making direct investments in China, by the fact that China holds enormous amounts of U.S. government securities?
68. Of course, if development helps countries absorb foreign capital better, why is the correlation between current account balances and growth for nonindustrial countries getting stronger over time, as figure 5 suggests? This is an important question for future research.
69. See, for example, Rajan and Zingales (2003), Mishkin (2006), and Kose and others (2006).
70. The Chinese approach of trying to spur banking reform by committing to open up the country's banking sector to foreign competition in early 2007, as part of their World Trade Organization accession commitments, can be seen in this light. Prasad and Rajan (2005) suggest an alternative strategy for dealing with the potential adverse effects of inflows through controlled liberalization of outflows (essentially by securitizing inflows), which would allow countries experiencing large capital inflows to develop their domestic financial markets and simultaneously mitigate appreciation pressures associated with those inflows.
71. For instance, capital account openness means more than just opening up to inward flows; it also means allowing outward flows. Outward flows could well relieve incipient appreciation pressures on the national currency, but they could also be a source of fragility, especially if the financial sector is underdeveloped. The fragility associated with the exit of capital could be attenuated if an economy is more open to trade (see Calvo, Izquierdo, and Mejia, 2004, and Frankel and Cavallo, 2004); trade openness could also mitigate the adverse effects of crises.