The impact of fiscal policy on non-oil GDP in Saudi Arabia

Figures & data

Table 1. Descriptive statistics of the variables

Statistic	y	gcu	gca	l	k
Mean	13.51	12.76	11.04	8.84	14.12
Maximum	14.21	13.46	12.52	9.42	15.08
Minimum	12.91	12.16	8.07	8.37	13.20
Standard Deviation	0.45	0.43	1.20	0.35	0.60
Coefficient of Variation	0.03	0.03	0.11	0.04	0.04

Source: Authors’ own calculation.

Figure 1. Time profiles of the variables.

Panel A: The log levels

Panel B: The growth rates of the variables

Source: Authors’ own calculation.

Figure 2. Oil revenues and the price of Arab Light crude.

Table 2. Long-run elasticities using ARDL, FMOLS, DOLS and CCR

Methods	$gcu$	$gca$	$l$	$k$	c
	Coef. (P-value)	Coef. (P-value)	Coef. (P-value)	Coef. (P-value)	Coef. (P-value)
ARDL	0.21 (0.00)	0.02 (0.00)	0.56 (0.00)	0.25 (0.00)	2.07 (0.00)
FMOLS	0.27 (0.00)	0.02 (0.00)	0.43 (0.00)	0.30 (0.00)	1.77 (0.00)
DOLS	0.25 (0.06)	0.03 (0.09)	0.61 (0.02)	0.20 (0.03)	1.79 (0.00)
CCR	0.28 (0.00)	0.02 (0.02)	0.42 (0.00)	0.30 (0.00)	1.72 (0.00)

Notes: The dependent variable is $y$; c indicates the intercept term; coef = coefficient; estimation period: 1992–2018.

In the ARDL estimation, we set the maximum lag order to 3 based on the data-dependent rule, and the Schwarz information criterion selects ARDL(3,0,0,2,0) as the optimal specification. This specification passes all the diagnostics and misspecification tests. It also indicates a cointegrating relationship among the variables even after using Narayan (2005) critical values as a small sample bias correction. This supports the finding from the Johansen test. The results are available from the authors upon request.

Note that we included a deterministic time trend in the long-run estimations above, but it appeared statistically insignificant in all the estimations. This is most likely because the trend component of non-oil GDP is captured and cancelled out by the trend components of the explanatory variables, as the residuals of the long-run equations for non-oil GDP are not trending, they are stationary, and a cointegrated relationship is established. Hence, there is no need for a deterministic trend in the estimations.

Table 3. Final ECM specification and test results

Panel A: Estimated coefficients of the final ECM specification
Regressor	$EC{T}_{t-1}$	$d{\left(y\right)}_{t-2}$	$d{\left(gcu\right)}_t$	$d{\left(gca\right)}_t$	$d{\left(l\right)}_t$	$d{\left(l\right)}_{t-1}$	$c$
Coefficient	−0.55	0.69	0.13	0.01	0.46	−0.43	−0.01
p-value	0.00	0.00	0.00	0.01	0.11	0.06	0.21
Panel B: Residual diagnostics, misspecification, over-identification and endogeneity tests results
Statistic	${\chi }_{SC}^2$	${\chi }_{HET}^2$	${\chi }_{ARCH}^2$	$J{B}_N$	$F_{FF}$	$J$	$J_{DIF}$
Sample value	1.50	8.58	2.71	0.45	0.00	0.26	0.03
p-value	0.47	0.20	0.10	0.80	0.99	0.88	0.86

Notes: Dependent variable is $d{\left(y\right)}_t$; ${\chi }_{SC}^2,{\chi }_{HET}^2$ and ${\chi }_{ARCH}^2$ denote chi-squared statistics to test the null hypotheses of no serial correlation, no heteroscedasticity and no autoregressive conditioned heteroscedasticity (ARCH) in the residuals, respectively. $F_{FF}$ is the F statistic of the Ramsey RESET test for the null hypothesis of no functional form mis-specification; $J{B}_N$ indicates the Jarque–Bera test for the null hypotheses of the normal distribution of the residuals; $J$ is the J-statistic for testing whether the null hypothesis of over-identification is valid; $J_{DIF}$ is the difference in the J-statistics from the Durbin–Wu–Hausman  test for testing the null hypothesis of exogeneity of a given variable (in our case this is $d{\left(l\right)}_t$). Estimation period: 1992–2018.

Narayan, P. K. 2005. “The Saving and Investment Nexus for China: Evidence from Cointegration Tests.” Applied Economics 37 (17): 1979–1990. doi:https://doi.org/10.1080/00036840500278103.

 Web of Science ®Google Scholar