Growth effects of a comprehensive measure of globalization with country-specific time series data

Abstract

Many studies have estimated the growth effects of globalization where globalization was measured with a few economic variables, ignoring its social and political dimensions. Recently, Dreher (2006) has developed a comprehensive measure of globalization with several variables from the economic, political and social sectors. He showed, with the panel data methods, that globalization has positive growth effect implying that countries with higher globalization grow faster. We argue that 5-year average growth rates, used in many panel data studies, are inadequate proxies for the unobservable Steady State Growth Rate (SSGR). Using the Dreher indices, we extend the Solow (1956) model to derive country-specific estimates of SSGRs for Singapore, Malaysia, Thailand, India and the Philippines. Our results show that countries with higher levels of globalization have higher SSGRs but the growth effects on SSGRs are smaller than in many studies.

Acknowledgements

The authors are grateful to two referees of this journal and their many valuable suggestions.

Notes

¹ Panel data methods are broadly of two kinds, namely those that ignore the stationarity properties of the variables (because they frequently use 5-year averages in the panels) and those that use methods for nonstationary variables. Classical methods of estimation, e.g. the generalized least squares, seemingly unrelated regressions and generalized method of moments are popular in the former. The Pedroni (1999, 2004) method is popular in the latter. While many studies on the effects of globalization have used the former type, there are relatively few studies with the Pedroni method. More recently Mark and Sul (2003) and Breitung (2006) have developed alternatives to the Pedroni method. Some widely used software packages, in both approaches, are EViews, STATA, RATS, TSP and GAUSS.

² Another important issue is whether or not globalization alleviates poverty; see, for example, Dollar and Kraay (2004) and Dreher (2006).

³ Although these observations were made by Easterly et al. (2004) in the context of the aid–growth relationships, they are equally applicable to other specifications.

⁴ These indices can be downloaded from http://globalization.kof.ethz.ch/

⁵ For some methodological views on the relative merits of country-specific time series estimates of growth models, see Greiner et al. (2004). More recently Luintel et al. (2008) also observed that country-specific time series studies are more useful.

⁶ Edwards (1998) provides an excellent survey of empirical studies prior to 1992.

⁷ Recently, Subasat (2003) demonstrates that the index developed by Dollar (1992) has fundamental flaws and therefore has no relevance to the debate on trade orientation and should be abandoned.

⁸ Dreher (2006) is not the first to construct a comprehensive measure of globalization. Kearney, and Andersen and Herbertsson (2005) have also developed such indices. The Kearney measure combines indicators of trade, finance, political engagement, information technology and personal contact to form a comprehensive measure of globalization for 62 countries. This index starts from 2000 and used by Foreign Policy magazine to determine the annual rakings of countries on the basis of the Kearney index. The Andersen and Herbertsson index is developed for 23 Herbertsson index is developed for 23 Organization for Economic Cooperation and Development (OECD) countries for the period 1979 to 2000. Nine economic variables are used to develop their globalization index with the factor loading technique. Like the Kearney index, the Andersen and Herbertsson index is used to rank the countries on the basis of their globalization index. This work has also a good critique of the Kearney index. A weakness of the Kearny index is that the weighting scheme is somewhat arbitrary in which they are not adjusted for the size of the country on the basis of its population. In contrast to these two studies, the Dreher index combines many economic, political and social indicators with the widely available technique of the principal components method in software packages like EViews and TSP, etc. Furthermore, it is not possible to use the Kearney and Andersen and Herbertsson indices in time series regressions whereas the Dreher index is most suitable for this purpose.

⁹ Winters (2004) also recognized that 5-year average growth rates are inadequate to measure the unobservable SSGRs. However, he suggests that 5-year growth rates are a pragmatic option to capture the transitional growth rates. We disagree with Winter's view because of two reasons. Firstly, globalization (or any other variable) is found to have transitional growth effects, by definition its permanent growth effects are nil. Secondly, even if the only aim is to estimate the transitional growth effects, then the degrees of freedom and efficiency of the estimates can be vastly improved with annual observations of the variables. This latter criticism also applies to other panel data works in growth.

¹⁰ Edwards (1998) has used an alternative method which is particularly useful for estimates with panel data. In his approach TFP is computed as the residual from the growth accounting exercises for each country. Their averages over 10-year panels were used as the dependent variable. Using alternative measures of trade openness he found that they all have significant effects on TFP. However, we have reservations on his short lengths of panels.

¹¹ Rodriguez and Rodrik (2000) and Winters (2004) have argued in favour of additional conditionality variables because globalization is often measured partially with a few economic variables. Let this conditionality variable be Z. The extended specification based on Equation 4 will be

¹² Following a suggestion of a referee, we have also used the Johansen Maximum Likelihood (JML) method to estimate for Singapore (for which our results are more robust) the cointegrating equations between ln y, ln k and T × GLO. Although there is a single cointegrating equation with a restricted trend, all the estimated coefficients are insignificant. Furthermore, the coefficient of T × GLO was negative and the coefficient of trend was implausibly high at 0.09. This may be, as this referee pointed out, due to the near perfect correlation of 0.99 between trend and T × GLO. Next, we tested for cointegration with trend and no T × GLO and then with T × GLO and no trend. In both cases there was a well-defined single cointegrating equation and all the estimated coefficients were significant. When only trend is retained in a restricted form, its coefficient is 0.03 and the share of profits is somewhat higher at 0.489. When only T × GLO is retained, with no trend, the share of profits increased to 0.674 and the coefficient of T × GLO was 0.243E⁻³. These estimates are qualitatively similar to those in Table 1 but not close enough. Since results with JML are not very impressive for Singapore, we did not proceed further with JML. For the merits of GETS over some alternative methods of cointegration techniques, see Rao (2007) and Rao et al. (2009), keeping in mind that this controversy is a methodological issue.

¹³ The computed 5 and 10% critical (absolute and sample size adjusted) values are 4.1063 and 3.7012, respectively. The cointegration test statistic is the absolute value of the t-ratio of the adjustment coefficient λ. The null is that there is no cointegration.

¹⁴ The average rates of growth of output per worker for Singapore are 5.6% (1971–1974) and 2.5% (2000–2004). The corresponding rates for Malaysia are 6.6 and 2%, respectively.

¹⁵ Estimates for india are for the period 1974 to 2003.

¹⁶ It may be argued that there is no need for the cointegration tests in the GETS because all the variables are I(0) in the specification. Therefore, the classical methods of estimation are valid and there is no need to estimate GETS specifications with the time series methods; see Rao et al. (2009). However, since these are methodological views, Ericsson and McKinnon (2002) have developed cointegration tests to make GETS consistent with the cointegration approach.

¹⁷ It is possible to estimate separate slope coefficients for each or a group of countries by introducing slope dummy variables. However, this is possible if the sample size is very large which in fact is not true in the empirical growth models. We are not aware of any empirical work that has introduced slope dummy variables although intercept dummies are common.