Bi-demographic and current account dynamics using SVAR model: evidence from Saudi Arabia

Abstract

The study explores the impacts of the bi-demographic structure on the current account and gross domestic product (GDP) growth. Using structural vector autoregressive modeling (SVAR), we track the dynamic impacts on these underlying variables. New insights about the dynamic interrelation between bi-population age dependency rate, current account, and GDP growth have been developed. In the short and medium terms, the reactions of GDP growth to both shocks of native and immigrant working-age populations move unsteadily in opposite directions. However, in the long run, both effects become moderately positive. Additionally, the positive long-run contribution of immigrant workers to the current account growth largely compensates for the negative contribution of the native population. We find a negative hump-shaped reaction of Saudi Age Dependency Rate to immigration policy shocks during a generation. When the shocks emanate from immigrants’ working age, there is a complex mechanism from the complementarity process to the substitutability process between immigrants and the Saudi workforce. In the short and medium term, the immigrant workers are more complement than substitutes for native workers.

Appendices

Appendix 1

See Tables 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12.

Table 2 Correlation: CA-to-GDP and age population sharing from 1974 to 2016

Table 3 Descriptive statistics with 42 observations (1974–2016)

Table 4 KPSS stationarity tests

Table 5 Zivot–Andrews (ZA) unit root tests

Table 6 Bounds test for cointegration with break points (k = 3)

Table 7 Narayan critical values (k = 3, n = 45)

Table 8 Diagnostic tests

Table 9 VAR lag-order selection criteria

Table 10 VAR residual serial correlation LM tests

Table 11 Residual variance–covariance matrix

Table 12 Residual correlation matrix

Notes on ARDL Bound test

The ARDL approach introduced and revised by Pesaran et al. (2001) is mostly used in empirical research when the regressors have a different order of integration or are all I(1) or all I(0). The ARDL bound model is based on the conditional error correction (CEC) model, which is formulated as follows (PSS 2001):

$$\Delta y_{t} = a_{0} + a_{1} t + a_{2} D_{t} + b_{0} y_{{t - 1}} + \sum\limits_{{i = 1}}^{k} {b_{i} } x_{{i,t - 1}} + {\text{ }}\sum\limits_{{i = 1}}^{{p - 1}} {\psi _{i}^{\prime } } \Delta z_{{t - i}} + \omega ^{\prime } \Delta x_{t} + u_{t}$$

where $y_{t}$ is a random scalar process that is conditionally modeled given the k-vector of random variables $x_{t}$ and the past values $z_{t - i} = \left( {y_{t - 1} ,{ }x_{t - i}^{^{\prime}} } \right)^{^{\prime}}$, and $u_{t}$ is the error term. The deterministic components are the intercept and the trend $t$, which are associated to a drift parameter $a_{0}$ and a time trend parameter $a_{1}$, respectively. When the deterministic components contribute to the error correction term, they are implicitly projected onto the span of the cointegrating vector, implying that $a_{0}$ and $a_{1}$ in the CEC model must be restricted.

Narayan (2005) does not consider the critical value (CV) bounds of the t Statistic for the cases 1, 3 and 5. For the case 5, PSS (2001) provide that at 5% of significance, the lower stationary bound and the upper bound are − 3.410 and − 4.160, respectively; and at 1% of significance − 3.960 and − 4.730, respectively. If the computed F-Statistic is higher than the upper bound of CV, then the null of no cointegration is rejected. The lower bound is based on the assumption that all of the variables are I(0), while the upper bound that all of the variables are I(1).

Appendix 2

Figures 4, 5, 6, 7, and 8.