Does Capital Account Liberalization Lead to Growth

Supplemental Appendix

Quinn, Dennis, Martin Schindler & A. Maria Toyoda (forthcoming), “Assessing Measures of Financial Openness and Integration,”IMF Economic Review.

Methodological Note on the IMF Tables

In the text, we note three drawbacks of table-based indicators. The first is that the switching point has never been defined by the IMF. By switching point, we mean that to assign a binary code of 0 or 1, the coder must have a threshold level, or degree of restrictiveness in mind at which point a country’s code switches from 0 to 1 or vice versa. One way of calculating the implied switching point is by comparing it to another, more finely gradated index, such as CAPITAL. The value of CAPITAL that provides the best fit in terms of dividing countries into restricted and unrestricted compared with IMF_Binary ‘drifts’ over time. This suggests that a given coding of IMF_Binary does not necessarily have the same meaning at different points in time. For example, Hungary (1993), Poland (1995), and Romania (1995) ‘switched’ to open in the IMF Table at a time when the corresponding scores in CAPITAL (discussed below) were 25, 37.5, and 50 (out of 100), respectively. KA, KAOPEN and CAPITAL do not score Hungary open until 2001, and FOI until 2002. In contrast, the Thatcher 1979 and Muldoon (New Zealand) 1983 liberalizations caused a switch to open when the scores on CAPITAL were 100 and 75 respectively. (See the discussion in IMF 2001).

A second caution is that the data series for indicators based on the Tables suffer from a structural break between 1995 and 1996. As noted, the Table from volume 1997 has properties incommensurable with those in prior editions.[1] The first break is that “restrictions on payments for current transactions,” which contains information about import payments and non-tariff barriers, is replaced with a narrower indicator – “control on payments for invisible transactions and current transfers.” (See IMF 1997, pp. 1-2 for a discussion of the categories.) Fifty-nine countries in 1995 are categorized as having current account restrictions versus 105 countries that are categorized as having invisibles restrictions in 1996 (IMF 1996, pp. 548-551; IMF 1997, pp. 946-952.) Paradoxically the narrower indicator shows many more instances of restriction than the broader one, even though other financial current account indicators show liberalization in 1996 (and in 1995 and 1994).[2]

A second break, already noted from Figures 1 and 2, is the elimination of the binary capital account indicator and its replacement with 12 or 13 categories. In 1995, 131 countries showed restrictions on IMF_Binary. The range for the 12 indicators listed in 1996, is from 54 countries restricting “liquidation of direct investment” to 131 countries with capital restrictions on commercial banks (IMF 1996, pp. 548-551; IMF 1997, pp. 946-952.) These structural breaks can affect the constructed measures. For instance, Chinn and Ito follow Mody and Murshid (2005) and use the invisibles restrictions category in place of the broader current account one to extend their measure past 1995 (2007, 169, fn. 14).[3] This simple change produces a large drop in the average values for KAOPEN between 1995 and 1996 in the range of 10%, which does not recover until 2002. (Mody and Murshid also note the drop in average values for their indicator between 1995 and 1996.) Other de jure indicators, in contrast, show increasing openness between 1995 and 1996.

A third caution is that data in the Table are ‘point in time’ measures. In volume 2008, for instance, that point in time for two thirds of the 185 countries and three territories was 31 December 2007. For the remaining one third of the cases (62 out of 188), the point-in-time was sometime in 2008, as late as 31 July in a few cases (e.g., Australia). In general, in annual data, the window of error is up to seven months, or roughly 60% of the temporal range. In five year averaged data, the seven month window is roughly 10% of the temporal range of five years.In what appears to be a systematic pattern, the data for larger and more economically significant emerging market countries are reported at later dates. In volume 2008 for the year 2007, Brazil (March), Chile (April), China (March), India (March), Mexico (June), and Vietnam (April) were among the countries with data occurring in 2008. Eleven EU members and the U.S. (May) also had later data. This will matter to investigators using time to identify models. If investigators use the AREAER indicators contemporaneously as independent variables, the best outcome is that the X variable as of 31 December of, e.g., 2007 is matched to a Y variable for 2007. In a third of the cases, however, the X Table variable will be measured in 2008 and used to explain Y for 2007.

Figures

1

1

Table A1. Five year non-overlapping averages, GMM system estimations

Conditioning information from QT RFS(2008)

Notes: The dependent variable is per capita economic growth PPP-adjusted. No serial correlation is indicated in GMM-SYS models when, in second stage analysis, the ABm2 test for second-order serial correlation is not significant, and the AR1 test shows evidence of significant negative serial correlation in the differenced residuals. For a discussion, see Doornik and Hendry (2001, 69). Standard errors are listed below the coefficients. * p-value < .10; ** p-value < .05; *** p-value < .01

Table A2. Five year panels with GMM_system estimations, Matched Samples

Conditioning information from QT RFS 2008

Notes: The dependent variable is per capita economic growth PPP-adjusted. The adjusted R2 is defined as 1-RSS/TSS with the standard adjustments. No serial correlation is indicated in GMM-SYS models when, in second stage analysis, the ABm2 test for second-order serial correlation is not significant, and the AR1 test shows evidence of significant negative serial correlation in the differenced residuals. For a discussion, see Doornik and Hendry (2001, 69). Standard errors are listed below the coefficients. * p-value < .10; ** p-value < .05; *** p-value < .01

Table A3. Five year panels with GMM_system estimations, Matched Samples

Conditioning information from BHL JFE 2005

Notes: The dependent variable is per capita economic growth PPP-adjusted. No serial correlation is indicated in GMM-SYS models when, in second stage analysis, the ABm2 test for second-order serial correlation is not significant, and the AR1 test shows evidence of significant negative serial correlation in the differenced residuals. For a discussion, see Doornik and Hendry (2001, 69). Standard errors are listed below the coefficients. * p-value < .10; ** p-value < .05; *** p-value < .01

1

[1] As Mody and Murshid note, their “financial integration index from 1996 onward is not entirely comparable to earlier years” (2005, p. 225, fn. 7).

[2] FIN_CURRENT and FOI assess financial current account openness. Both indicators show a liberalization of the global averages of the financial current account in 1996, as well as similar increases in 1994, 1995, and 1997.

[3] The details of the procedures are not reported in either paper.