Resolved: On balance, economic globalization benefits worldwide poverty reduction 3

Globalization reduces poverty

Our dataset covers the 1983–2007 period, with poverty data available from 1988, averaged over four five-year periods: 1988–92, 1993–97, 1998–2002, and 2003–07. Five-year averages are used both because we lack yearly data and to minimize the impact of measurement errors. The panel is unbalanced but includes information for 114 countries, and the efficient sample consists of more than 300 observations meeting baseline specifications. An absolute majority of these observations refers to conditions in countries classified as low- or lower middle-income countries with a 2008 GNI per capita of USD 3,855 or less. Table 12 in the Appendix presents descriptive statistics and sources for all variables used in the analysis, and Table 13 in the Appendix presents details about the country sample.

The main dependent variable is the percentage of the population in a country living on less than one dollar per day (PPP adjusted 1993). This absolute measure comes from the Povcal database (World Bank, 2010) and is derived from household surveys. Our globalization measure is the KOF Index (Dreher et al., 2008), as described in Section 2. We use the index both as a composite measure (KOF), in which the three dimensions of globalization are equally weighted together, and in a disaggregated format (KOF1 and KOF2). Moreover, we use the sub-components for the economic and the social globalization index (flows and restrictions, and information flows, personal contact, and cultural proximity, respectively). The index takes values between 0 and 100 with higher values indicating more globalization. As can be seen in Figure 2, the cross-country correlation between the aggregate globalization index and poverty is clearly negative.

Our baseline is a panel data setting that should capture potential short-run effects on poverty. We specify an equation that relates globalization to poverty and to a set of control variables as follows (countries being indicated by i and time by t):

In Eqn. (1), the brackets indicate that the quadratic term of globalization may be excluded. Glob is a vector for different types of globalization, X includes control variables, δ_i corresponds to a country-fixed effect that captures stable differences in poverty between countries, and ρ_t is a period-fixed effect capturing the influence of shocks that affect poverty in multiple countries at the same time. Finally, ɛ_it is an error term assumed to be normally distributed.

In the baseline regression, globalization is lagged one period. For example, average globalization in 1983–87 is used to explain average poverty in 1988–92. With this specification, if the J-curve hypothesis is correct, a linear globalization term should have no or possibly a negative sign, whereas a quadratic specification is likely to fit the data better.¹¹

We begin by estimating a relatively parsimonious baseline, controlling only for country log real GDP per capita (PPP adjusted) collected from the World Bank (2010). To maximize comparability across specifications including the same indicator of globalization, we let the sample contain the same countries. The number of observations might, however, vary across index-specific estimations.¹² The null hypothesis of no country effects is rejected in all estimations and using a Hausman test, the random-effect model is rejected against the fixed-effect model. Moreover, time dummies are jointly significant in a majority of baseline specifications, suggesting they should be included in the model. All specifications consequently include country-fixed and time-fixed effects. Table 2 presents baseline results using panel regressions.

As expected, the linear specifications in columns 1, 4, and 7 fit worse than the remaining quadratic specifications: As shown in columns 2 and 3, aggregated globalization negatively associates with absolute poverty with decreasing marginal effect, consistent with the inverted J-curve hypothesis. Note that the aggregate index includes political globalization that we do not analyze separately.¹³ Results for the different types of globalization (columns 5, 6, 8, and 9) suggest that the poverty-decreasing effect holds for both economic and social globalization. Interestingly, the size of the association decreases only marginally when controlling for income (columns 3, 6, and 9). This result suggests that globalization tends to decrease poverty, but not mainly via the income channel—at least not in the short run captured by the panel.

Table 3 examines what components of economic and social globalization that explain the negative association with absolute poverty in Table 2. Including trade flows and trade restrictions separately and together (columns 1–6) suggests that the significant coefficient on economic globalization comes from restrictions rather than from flows.¹⁴ A similar exercise for social globalization (columns 7–14) points to the importance of information flows for reducing poverty, while cultural proximity actually seems to have a small poverty-increasing effect when all components of social globalization are included simultaneously.

To further check the robustness of our baseline findings using panel regressions, Table 14 in the Appendix presents the results from the following sensitivity tests: changing the specification to a random effects model, using globalization in period t rather than t-1, excluding observations with extreme values of globalization and poverty, using alternative measures of poverty, and excluding various geographical regions. ¹⁵ Overall baseline results are robust to these changes, suggesting that globalization is good for the poor. In particular, more liberal trade restrictions and larger information flows correlate with less absolute poverty.