4.1 Johansen cointegration test results
Since all the variables are stationary in the first difference, we run the Johansen cointegration test to determine the presence of any cointegration or long run relationship among the variables. However, before running the cointegration test, we need to run the VAR model first to determine the optimal lag length based on the minimum Akaike Information Criterion (AIC). Due to the limited number of observations, the maximum lag has been set to two in the lag length selection process. The optimal number of lags is two based on the AIC results.
After having determined the number of lags, we proceed with the cointegration test for the model. Additional file
2: Table S2 shows that there are five cointegration equations based on the trace test, and three cointegration equations based on the maximum eigenvalue test. In other words, the results indicate the existence of more than one long run relationship among the variables in the system comprising lnIO, lnGFCFI, lnOP, lnMX, GPOP and lnAO.
Since there are more than one cointegration relationship among the variables lnIO, lnGFCFI, lnOP, lnMX, GPOP and lnAO, there is the problem of selecting the appropriate cointegration relationship. However, since the main objective of this study is to examine the main determinants of industrial output of Syria, we have identified the first cointegrating equation to be normalized with respect to the real industrial output variable as it meets the theoretical, a priori expectations with respect to the signs of the coefficients. Additional file
3: Table S3 shows the normalized cointegrating vector.
From Additional file
3: Table S3, the long-run equation for industrial output is as follows:
$$\ln{\text{IO}} = - 56.32602 + 0.626022 \, \ln {\text{GFCFI}} - 0.627859 \, \ln {\text{OP + }} 0.402981 \, \ln {\text{MX}} + 0.525604{\text{ GPOP}} + 2.244837 \, \ln {\text{AO}}$$
(2)
The cointegration equation given by Eq. (
2) above shows that lnIO is positively related to lnGFCFI, lnMX, GPOP and lnAO, and negatively related to lnOP in the long run.
The positive coefficient of 0.626 for lnGFCFI indicates that over the long run when the gross fixed capital formation of industry increases by one percent, industrial output will increase by 0.63 percent, ceteris paribus. It is noteworthy that when the amount of capital invested in industrial activities increase, industrial production in the country will increase because it can support these activities with more and better machines and spare parts. The Syrian government has worked to make the industrial sector the locomotive of economic growth in the country by encouraging local and foreign investments in the industrial activities, as well as establishing the Industrial Bank of Syria in order to develop the industrial sector by providing credit facilities and loans to finance the industrial activities.
The negative coefficient of 0.628 for lnOP indicates that when the oil price increases by one percent, industrial output will decrease by 0.63 percent, ceteris paribus. This outcome is as expected since production requires fuel for transportation of raw materials to the factories, and the finished products to the markets. Oil is also an input in the chemical industries. Therefore, any increase in oil prices affects the cost of industrial production, and that affects negatively the industrial production in the country. Moreover, the rise in the cost of transportation and production activities after oil price increases will increase the prices of industrial products in the global market, which will lead to a decline in the international competitiveness and external demand for these products. This will drive producers to reduce their production, which has a negative impact on the industrial production in the country.
Our result seems to agree with Lee and Ni (
2002). Their findings indicated a reduction in the demand for non-oil-intensive industries because of oil price shocks. However, it contradicts with the findings of Farzanegan and Markwardt (
2007) for Iran, and Mehrara and Sarem (
2009) for Iran, Saudi Arabia and Indonesia. The latter two studies found that oil price has a positive effect on the industrial production of Iran and Saudi Arabia, because they are oil-producing and exporting countries, and oil industries have an important role in their industrial activities. For Indonesia, oil price has no significant effect on industrial production due to the successful diversification of the real sector to reduce the harmful impact of oil booms. In Syria’s case, the oil sector has a vital role in the economy. However, oil production has been declining continuously since 1996, and the government has been trying to reduce its dependence on the oil sector, by promoting industrial diversification and making the industrial sector the locomotive of economic growth.
The positive coefficient of 0.403 for lnMX indicates that when manufactured exports increase by one percent, industrial output will increase by 0.40 percent, ceteris paribus. It is noteworthy that producers can utilize the returns from manufacturing exports to increase and improve their production. Moreover, exporting to global markets increases the degree of competition, which leads producers to be more enthusiastic to improve the quality of their production. By the end of the 1980s, Syrian exports to the Soviet Union and Eastern European markets have increased. Furthermore, the new Syrian government’s strategy to diversify Syrian exports, open up the Syrian economy to foreign trade, and reduce the percentage share of raw materials in total Syrian exports have motivated the Syrian producers to increase and improve their production in order to achieve higher profits from exporting abroad. Our finding is in line with that of Mamun and Nath (
2005); Uddin and Noman (
2011); Akpan et al. (
2012) who argued that exports have a positive effect on industrial production.
The positive coefficient of 0.527 for GPOP indicates that for every 0.1 percentage point increase in population growth, industrial output will increase by 5.3 %, ceteris paribus. This suggests that population growth has an important role in improving and increasing the industrial production in the country. When the population increases, domestic consumption and demand for different goods and services will rise in the country, which will induce producers to increase and improve their production to meet the increase in local demand. Furthermore, most of the industrial activities in Syria are labor-intensive, so increases in population growth can play a crucial role in providing the necessary manpower for the industries. Besides that, the Syrian government has worked to upgrade the skill of workers in the industrial sector to improve and increase the industrial production in the country.
The positive coefficient of 2.245 for lnAO indicates that when agricultural output increases by one percent, industrial output will increase by 2.25 %, ceteris paribus. This result is as expected, as the agriculture sector is one of the main sources of raw materials and semi-finished products for different industries such as the food, clothing, and textile industries. The output of the food and textile industries has the largest percentage share of total manufacturing output in Syria, where it accounts for about 51–61 % of the total manufacturing output in the country. When agricultural production increases, inputs available for industrial activities will also increase, which motivate industrial producers to increase their production. This result clearly shows the importance of the agricultural sector in supplying these industries with inputs needed in their production activities. Hence, increasing agricultural output will boost the industrial production in Syria, while low agricultural production leads to a decline in industrial production in the country. This result concurs with that of Gollin et al. (
2002) who found that industries are dependent on agricultural production.
4.3 Statistical diagnostic tests results
In order to check for model adequacy, the VECM is subjected to a number of diagnostic tests, namely the normality, serial correlation, heteroscedasticity (BPG and ARCH) and Ramsey RESET tests to ascertain its statistical adequacy. A 5 % level of significance is used in all these tests.
Additional file
5: Table S5 reports the results of the diagnostic tests. The lnIO, lnOP and lnAO equations in the VECM passed the normality, heteroscedasticity and Ramsey RESET tests, except the serial correlation test. However, the lnGFCFI, lnMX and GPOP equations passed all the four tests. The serial correlation problem may be due to the insufficient number of lags in the VECM. However, even after increasing the lag length, the serial correlation problem persists. Given the limited number of observations, it is not possible to increase further the lag length. Therefore, the serial correlation problem has been corrected using the Newey-West HAC standard errors before proceeding with the t and F tests for long-run and short-run Granger causality.
4.5 Variance decomposition (VD) analysis results
The variance decompositions (VD) for 1-year to 10-year forecast horizons indicate the amount of information each variable contributes to the other variables in the VAR.
Additional file
6: Table S6 gives the decomposition of the forecast error variance of the variables in the industrial output model. In the 1st year, the error variance of industrial output is exclusively generated by its own innovations and its’ contribution has been decreasing since then for the various forecast horizons. However, at the 10th year forecast horizon, its own shocks contribute about 63 % of the forecast error variance. On the other hand, GPOP, lnGFCFI, lnMX, lnOP and lnAO shocks explain 11.4, 6.7, 4.7, 7.1 and 7.5 %, respectively to the forecast error variance of industrial output. Furthermore, the contributions of lnOP and lnAO in explaining lnIO forecast error variance have increased during the 10-year forecast period, but there are no significant changes in the contribution of lnMX. The relative importance of lnGFCFI, however, has decreased at the 2- to 6-year forecast horizon and then increased at the 7- to 10-year horizon. The contribution of GPOP shocks in explaining lnIO error variance at first increased from the 2- to 5-year horizon, but it started to decline in the 6- to 10-year horizon; however, it remained the highest contributor after lnIO.