This study has formulated the following simple regression model to examine the effects of income growth, domestic revenue, and public debt on public health expenditure in India for the period from 1980-81 to 2015-16.
We have taken Public health expenditure (PHE) as our dependent variables which include both medical & public health and family welfare expenditure in both current and capital accounts of the general government (i.e. central and state) in India. We have taken Economic growth (EG), Total Revenue (REV), Total Debt (DEBT) as our independent variables. EG is calculated as the annual percentage change in Gross Domestic Product (GDP). REV includes the revenue generated from both tax and non-tax sources. Tax revenue includes both direct taxes and indirect taxes. Direct taxes include taxes on personal income, property, and capital transactions whereas indirect taxes include taxes on sales of goods & services. Non-tax revenue includes revenue generated from fees & fines, interest receipts from commercial enterprises, royalties from natural resources. Total debt (DEBT) includes both domestic and external liabilities. In Eq. (1) PHE is measured as a percentage of GDP; REV is measured as a percentage of GDP; DEBT is measured as a percentage of GDP; is a disturbance error term; is time; ln is the natural log. The data has been collected from the Handbook of Statistics on the Indian Economy published by RBI (2019). All the data are from combined government finance[1] which includes aggregate public health expenditure, aggregate revenue, and aggregate debt. All variables are in constant (real) prices at base 2004-05 and converted into a natural logarithm except EG for the empirical analysis.
Table 1 represents the descriptive statistics and pairwise correlation results of variables. We have found that the mean percentage of revenue (i.e. as a ratio of GDP) and the mean percentage of public debt (i.e. as a ratio of GDP) is 21% and 68% respectively in India. Similarly, the mean annual growth of GDP and mean percentage of PHE (i.e. a ratio of GDP) is 6% and 1.27% respectively. The gap between minimum and maximum range values of variables – DEBT and REV is larger but there is a very little gap between minimum and maximum values in PHE. The pair-wise correlation result shows that there is a positive association between PHE and REV while a negative association between PHE and DEBT. The simple correlation analysis could not be produced the strength of the association between variables. Therefore, we have employed advanced econometric methods to examine the short-run and long-run relationships between PHE and other macro-fiscal factors – EG, REV, and DEBT for the last 36 years of the Indian economy.
Table 1. Descriptive statistics and pairwise correlation
Variables
|
Definition
|
Mean
|
Std. Dev.
|
Min
|
Max
|
Pairwise correlation
|
|
|
|
|
|
|
PHE
|
EG
|
REV
|
BOR
|
PHE
|
Public Health Expenditure (as a percent of GDP)
|
1.272
|
0.102
|
1.135
|
1.549
|
1
|
|
|
|
EG
|
Economic Growth (annual growth rate of GDP)
|
6.350
|
2.068
|
1.430
|
10.159
|
-0.177
|
1
|
|
|
REV
|
Public Revenue (as a percent of GDP)
|
20.586
|
1.237
|
17.870
|
23.372
|
0.331
|
0.188
|
1
|
|
DEBT
|
Public Debt (as a percent of GDP)
|
67.833
|
8.577
|
47.936
|
83.228
|
-0.293
|
0.134
|
0.262
|
1
|
Note: GDP: Gross Domestic Product; All variables are constant 2004-05 base year prices.
Source: Author’s estimation from the Handbook of Statistics on Indian Economy, RBI (2019).
2.1 The Autoregressive Distributive Lag (ARDL) approach to Cointegration
To estimate Eq. (1), we have applied the Autoregressive Distributive Lag (ARDL) approach to cointegration, proposed by Pesaran et al. (2001).
Eq. (2), ∆ denotes the first difference operator of the respective variables and non-stochastic drift parameter. To find out whether there is a long-run cointegrating relationship among variables - PHE, EG, REV, and DEBT, we test the null that: H0: β1 = β2 = β3 = β4 = 0 and the alternative hypothesis, Ha: β1 ≠ β2 ≠ β3 ≠ β4 ≠ 0 by following a non-standard F-test statistics. If it rejects the null hypothesis of no cointegration in Eq. (2) statistically, we can say that there is a long-run association exists among the variables.
After getting a long-run association among variables using the ARDL bounds testing approach to cointegration, we can estimate the short-run and long-run elasticity of public health expenditure using the following unrestricted Error-Correction Model (ECM):
where λ is the speed of adjustment parameter and ECT (Error Correction Term) is the residuals from the estimated model in Eq. (3).