Health care expenditures and GDP in Latin American and OECD countries: a comparison using a panel cointegration approach

Abstract

This paper provides empirical evidence of the existence of a long-run causal relationship between GDP and health care expenditures, for a group of Latin American and the Caribbean countries and for OECD countries for the period 1995–2014. We estimated the income elasticity of health expenditure to be equal to unity for both groups of countries, that is, health care in Latin American and OECD countries is a necessity rather than a luxury. We did not find evidence of a causal effect in the opposite direction, i.e. from changes in health expenditure to GDP. We present conclusive evidence of the cross-country dependence of the analyzed series, and consequently we used panel unit root tests, panel cointegration tests, and long-run estimates that are robust to such dependence. Specifically, we use the CIPS panel unit root test and the panel Common Correlated Effects estimator. We also show that the results obtained by mistakenly using methods that assume cross-section independence are unstable.

Appendix

Health care expenditures as percentage of GDP in the world, and in selected groups of countries

See Figs. 1 and 2.

Fig. 1

Health care expenditures as percentage of GDP between 1995 and 2014. Notes “LA” is the group of 33 Latin American countries. “OECD” is the group of 35 Organisation for Economic Co-operation and Development member countries. “ALL” includes 192 countries for which HE data is available in the Global Health Observatory of the WHO

Fig. 2

Source: Global Health Observatory Map Gallery, WHO

Health care expenditures as percentage of GDP in 2014

Trends in health care expenditures and GDP

See Figs. 3, 4, 5 and 6.

Fig. 3

Health care expenditures trends, LA countries. Notes Natural logarithm of health care expenditures per-capita, in constant dollars of 2010

Fig. 4

Health care expenditures trends, OECD countries. Notes Natural logarithm of health care expenditures per-capita, in constant dollars of 2010

Fig. 5

GDP trends, LA countries. Notes Natural logarithm of GDP per-capita, in constant dollars of 2010

Fig. 6

GDP trends, OECD countries. Notes Natural logarithm of GDP per-capita, in constant dollars of 2010

Sensitivity to country exclusion: CIPS panel unit root test

See Tables 12 and 13.

Table 12 Second-generation panel unit root tests on HE and GDP excluding one country at a time. LA countries

Table 13 Second-generation panel unit root tests on HE and GDP excluding one country at a time. OECD countries

Individual country CCE estimates

See Table 14.

Table 14 CCE estimates by country. LA and OECD countries

Sensitivity to country exclusion: CCE estimates

See Table 15.

Table 15 Robustness of CCE estimates to country exclusion. LA and OECD countries

Westerlund (2007) cointegration test

The following description of the test was taken from the help file that accompanies the Stata command xtwest coded by Persyn and Westerlund (2008).

The panel cointegration tests developed by Westerlund (2007) contrast the absence of cointegration by determining whether there is error correction for individual panel members or for the panel as a whole. Consider the following error correction model, where all variables in levels are assumed to be I(1):

$$\begin{aligned} \Delta y_{it}= & {} c_i + a_{i1}*\Delta y_{it-1} + a_{i2}*\Delta y_{it-2} + \cdots + a_{ip}*\Delta y_{it-p} \\&+ b_{i0}*\Delta x_{it} + b_{i1}*\Delta x_{it-1} + \cdots + b_{ip}*\Delta x_{it-p} \\&+ a_i(y_{it-1} - b_i*x_{it-1}) + u_{it} \end{aligned}$$

where $a_i$ provides an estimate of the speed of error-correction towards long-run equilibrium $y_{it} = - (b_i/a_i) * x_{it}$ for the series i.

The Ga and Gt test statistics contrast $H_0: a_i = 0$ for all i against $H1: a_i < 0$ for at least one i. These statistics start from a weighted average of the individually estimated $a_i$’s and their t-ratio’s respectively. Rejection of $H_0$ should therefore be taken as evidence of cointegration of at least one of the cross-sectional units.

The Pa and Pt test statistics pool information over all the cross-sectional units to test $H_0: a_i = 0$ for all i vs $H_1: a_i < 0$ for all i. Rejection of $H_0$ should therefore be taken as evidence of cointegration for the panel as a whole.

If the cross-sectional units are suspected to be correlated, robust critical values can be obtained through bootstrapping.

Controls

See Tables 16 and 17.

Table 16 Descriptive statistics of covariates, by country. LA countries

Table 17 Descriptive statistics of covariates, by country. OECD countries