Does Financial Openness Reduce Poverty?

by Shona Carter

Poverty is has been a global challenge for many years as leaders around the world work together to combat its devastating effects. Although success in poverty reduction has been seen globally, there still remains persistent challenges to combating the issue. Poverty remains a big problem in Latin America. As globalization spreads throughout the world, partnerships regarding politics and economics aim to reduce poverty because it can increase shared prosperity. Financial globalization has been considered a way of improving struggling economies and thereby reducing poverty.  I hypothesize that more financial integration would lead to lower levels of poverty. I will assess the impact of financial openness on poverty in Latin America because poverty remains an issue in the region and there strong data exists regarding financial integration in the region. The independent variable was obtained from a financial openness index produced by economists Menzie Chinn and Hiro Ito. The index measures the extent to which a country’s capital account is open or has unrestricted capital account transactions. Previous measures of financial openness were mostly binary variables that tell if a country is “open” or “closed”, without accounting for capital accounts that are partially open or partially closed. To overcome this, the Chinn-Ito index uses previous measures and 4 binary variables that indicate a country’s financial openness: the presence of multiple exchange rates, restrictions on current account transactions, restrictions on capital account transactions, and export proceeds surrender requirements. These variables together measure the intensity of capital controls better than a binary variable that only considers whether a capital account is open or closed. 

The variable for poverty, which was obtained from the World Bank’s listing of economic indicators, is a very scarce variable that is defined by the percentage of people living on less than $1.25 per day. Due to the scarcity of the poverty variable, I created a new variable using factor analysis. Two additional variables, life expectancy and infant mortality were used to construct a new measure of poverty due to their high correlation with the original poverty variable.  I will use this new variable to run a regression with financial openness variable and poverty in order to get a meaningful idea of the correlation between financial openness and poverty in Latin America. Due to the negative relationship between infant mortality and life expectancy, the factor variable has a negative relationship to the original poverty variable.

I will control for social and political indicators of globalization, as they may influence the degree to which a country is financially integrated.  The social globalization variable was made using information regarding international connections, such as tourism, also the flow of information, such as internet users. It also uses cultural proximity which could be defined using the number of international businesses per capita. Social and Political globalization variables come from a Quality of Governance dataset. Political globalization measures the degree to which a country has a relationship with other countries on a government to government level. This variable includes information such as the number of embassies a country has and its memberships within international institutions. I use cross section time series analysis to create multple regressions using the STATA software. I also account for issues that arise with panel data such as autocorrelation. 

Cross Section Time Series Analysis

Methods and Tables:

The statistical method that I used required me to set up the data by the country and year so that I can make the regressions.  I ran a hausman test which helps determine whether I should used fixed effects or random effects in my regression. These effects would help me to obtain a valid statistical inference when there is unobserved heterogeneity ( varying unobserved  components of the effect of my estimation) The hausman test supports my using fixed effects in my regressions.

Table 1 shows 5 models. The main independent variable, financial integration, and two additional independent/control variables, social and political globalization, are regressed with the original poverty variable. They are also regressed with the factor variable, as well as the two dependent variables, infant mortality and life expectancy, which are highly correlated with poverty.  In the second model, only the observations for the poverty variable are regressed with the factor variable.

Table 1: Financial Integration’s Impact on Poverty in Latin America (year effects)

	(1)	(2)	(3)	(4)	(5)
VARIABLES	Poverty	Factor(for Poverty)	Factor	Infant Mort.	Life Expect.
Financial Integration	-4.633**	0.0195	-0.117***	2.654**	-0.913***
	(1.967)	(0.0744)	(0.0328)	(1.183)	(0.239)
Political Globalization	-0.00791	0.0102***	0.00347***	-0.173***	0.0122
	(0.0667)	(0.00252)	(0.00129)	(0.0471)	(0.00954)
Social Globalization	-0.304**	0.00725	0.0150***	-0.434***	0.137***
	(0.128)	(0.00485)	(0.00215)	(0.0777)	(0.0157)
Constant	17.70**	-0.188	-1.296***	89.19***	56.46***
	(7.392)	(0.358)	(0.128)	(4.632)	(0.940)
Observations	148	148	997	1,021	1,008
Number of id1	14	14	27	27	27

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1 

The first table shows the regressions with year fixed effects and the second table shows the regressions with year and country effects.  Table 1 and 2 produce similar results with the coefficients aligning in the same directions.

Table 2: Financial Integration’s Impact on Poverty in Latin America (country effects)

	(1)	(2)	(3)	(4)	(5)
VARIABLES	Poverty	Factor (for poverty)	Factor	Infant Mortality	Life Expectancy
Financial Integration	-4.173**	0.655***	-0.178***	6.702***	-1.307***
	(1.819)	(0.0550)	(0.0463)	(1.623)	(0.323)
Political Globalization	-0.0681		0.0225***	-0.674***	0.143***
	(0.0566)		(0.00161)	(0.0509)	(0.0106)
Social Globalization	-0.139		0.0474***	-1.656***	0.374***
	(0.0971)		(0.00233)	(0.0749)	(0.0156)
Constant	25.68***	-0.0624**	-2.885***	138.5***	46.03***
	(3.550)	(0.0287)	(0.0787)	(3.439)	(0.814)
Observations	148	1,189	997	1,021	1,008
R-squared	0.203	0.109	0.623
Number of id1	14	32	27	27	27

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Table 3: Financial Integration’s impact on Poverty in Latin America (lagged independent variables)

	(1)	(2)
VARIABLES	Factor	Factor
Financial Integration	0.168	0.0432
	(0.103)	(0.0312)
Lag Social Globalization	0.0118**	0.0137***
	(0.00558)	(0.00178)
Political Globalization	0.00987***	0.00481***
	(0.00254)	(0.000848)
Lag Political Globalization	0.0135***	0.00488***
	(0.00251)	(0.000865)
Social Globalization	0.0343***	0.0131***
	(0.00543)	(0.00171)
Lag Financial Integration	-0.364***	0.0261
	(0.102)	(0.0302)
Constant	-2.854***	-1.011***
	(0.0791)	(0.178)
Standard Error/Auto. Adjustment?	No	Yes
Observations	972	972
R-squared	0.635	0.605
Number of id1	27	27

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

In these regressions for table 3, I decided to lag the independent variables in order to get a more accurate estimate of the true effect of financial integration. By lagging the independent variable, the models consider previous values of X for the effect on Y, and could produce the most meaningful of all the models.  Tables 1, 2, and 3 show a progression of obtaining the most accurate and meaningful results.

Results:

Based on the most controlled and accountable regression in table 3, financial integration has a weak or non-significant effect on poverty.

The simpler regressions in the first 2 tables that do not include lagged variables or account for auto regression, show a significant yet inconsistent effect on the scarce poverty variable. Keeping in mind that the factor variable and poverty should have a negative relationship, these regressions do not show the expected results. For example, since financial integration is associated with a decrease in poverty in model 1 on the first 2 tables, then financial integration should lead to a positive coefficient in the factor variable in model 3. The expected direction of the factor variable is only shown in model 2 where it is regressed solely with the same units for the poverty variable. The fact that the coefficients for the factor variable and poverty go in the same direction in models 1 and 3 in the first two tables, means that those results are inconsistent.   

In the third table I set out to get the most accurate regression to account for the discrepancies in tables 1 and 2 by using lagged variables with standard error adjustments.  In model 1 of this table with lagged variables, I used the same process for working with panel data that I used in the previous two models. In model 2 of this table, I used a different process to calculate panel corrected standard error estimates that help with the issue of auto correlation.

The results from the first tables show that financial integration has a significant effect on the scarce variable for poverty, loosely showing that financial integration leads to a decrease in poverty, however since the poverty variable is scarce and the factor variable produces inconsistent results, model 1 is not enough to allow me to reject the null hypothesis. Therefore I would fail to reject the null hypothesis that financial integration has no relationship to poverty in Latin America. The 2 other independent variables however, show a significant relationship to life expectancy and infant mortality in the expected directions.  

Conclusion: 

The regressions produce inconsistent results despite high levels of significance. After making the proper adjustments to the regression over time as I have gradually done from tables 1 through 3, the effects of financial integration remain volatile. This tells me that either there is likely more to this analysis than what these particular variables contribute or financial integration is not a strong predictor of poverty levels. It does not mean that financial integration is definitely not associated with reductions in poverty, but rather my regression does not produce a stable result that controls for the issues associated with panel data.

Appendix:

Summary Statistics

Variable		Mean	Std. Dev.	Min	Max	Observations
poverty	overall	14.1763	15.71471	0.03	68.49	N =     165
	between		15.20486	1.62	61.75917	n =      14
	within		5.382453	2.527136	36.82714	T = 11.7857
Factor	overall	1.20E-10	1	-4.004508	1.487515	N =    1684
	between		0.6822429	-2.025807	0.9966946	n =      36
	within		0.741246	-2.262154	1.790473	T = 46.7778
financial integration	overall	0.4315984	0.3442466	0	1	N =    1313
	between		0.2168614	0.1247907	1	n =      34
	within		0.2723456	-0.245791	1.130454	T-bar = 38.6176
Political Globalization	overall	51.74875	19.25492	10.19148	94.72778	N =    1273
	between		17.81613	18.27359	85.22592	n =      33
	within		9.09341	25.41777	78.00189	T = 38.5758
Social Globalization	overall	39.82963	11.57027	12.46371	66.42154	N =    1273
	between		9.886109	14.93896	63.53885	n =      33
	within		6.339785	25.87805	66.74578	T = 38.5758
infant mortality	overall	45.07957	34.22633	5	202	N =    1806
	between		22.89857	13.7125	116.6556	n =      36
	within		25.74855	-20.65191	136.4481	T-bar = 50.1667
Life expectancy	overall	67.26714	7.067833	42.159	79.70502	N =    2027
	between		5.206262	52.75044	77.01164	n =      43
	within		5.132249	51.33151	79.94593	T-bar = 47.1395

  
***************************************************************************




Graph: Financial Integration and Poverty

A strong effect would show a clear positive relationship between financial integration and the factor variable which is the same as a negative relationship between financial integration and poverty.  

Sources:

http://qog.pol.gu.se/data/datadownloads/qogstandarddata

http://data.worldbank.org/indicator/SI.POV.DDAY

http://web.pdx.edu/~ito/Chinn-Ito_website.htm