According to Ramirez, inwards FDI to developing nations have the chance to raise their capital stock as well as technological know-how, and in turn, increases the recipient country's output level, labor productivity, and tax revenues. Nonetheless, FDI flows may harm the growth prospects of a nation if they result in huge reverse flows like profit and dividends remittances and/or if the MNCs gain large tax concessions from the host nation. The negative impact will be continuously compounded if the expected positive spillover effects from the technological transfer are fully minimized or eliminated due to the over restrictive intellectual property rights (IPRs) and/or because the transferred technology is inadequate for the recipient country's factor proportions (e.g., too capital intensive).
Following De Mello (1997) and Zhang (2001), the spillover (positive /negative) related to the incorporation of FDI stock can be generally modeled through an augmented Cobb-Douglass production function in the following:
Y = Af(L, Kp, E) = ALα Kβ E1-α-β
(1)
where Y represents real output, Kp represents the private capital stock, L represents labor, and E is the externality created by the additions to the FDI stock. α and β are the shares of local labor and private capital respectively, and A captures the efficiency of production. Besides, α and β are assumed to be less than one. In other words, there are diminishing returns to the labor and capital inputs.
The externality, E, can be captured by the following type of a Cobb-Douglas function:
E = (L, Kp, )θ,
(2)
where γ and θ are, respectively, the marginal and the inter-temporal elasticities of substitution between private and foreign capital. Let γ > 0, such that a greater FDI stock creates a positive externality to the economy. If θ > 0, inter-temporal complemetarity exists and, if θ < 0, additions to the FDI stock crowd out capital over time and deteriorate the growth potential of the host nation.
Combining equations (1) and (2), the following can be obtained:
Y = ALα+ θ (1-α-β)
(3)
A standard growth accounting equation can be derived by taking logarithms and time derivatives of equation (3) to generate the following dynamic production function:
gy = gA + [α + θ(1 - α- β)]gL + [β+ θ(1- α - β)]gKp + [γ θ(1 - α - β)]gKf ,
(4)
where gi is the growth rate of i = Y, A, L, Kp, and Kf. Equation (4) states that (provided γ and θ > 0) additions to the FDI stock will augment the elasticities of output with respect to labor and capital by a factor θ(1 - α - β).
In terms of FDI spillovers, MNCs are crucial drivers of technology and skills, management methods, and training that serve great stimulation of economic growth and development in recipient nations. Investment made by MNCs in developing nations can result in the transfer of best practices, marketing skills, as well as knowledge to domestic firms (Musonera, 2007). Then, all these could result in quality and productivity growth, as well as in other positive externalities.
Foreign investment affects the recipient nation's economic growth through different channels. One of them is: MNCs bring their best technologies, management techniques, and marketing expertise to the domestic market, and could affect GDP growth in a favorable manner. In addition, because MNCs intend to maximize their profits by competing and attempting to outperform domestic and global competitors, they also stimulate competition in the recipient nation. This increased competition is usually followed by spillover effect in the manufacturing industries. Lastly, MNCs could enhance the productivity of domestic firms via providing technical and training assistance to improve a supplier's product quality and other condition (Smarzynska, 2004). The knowledge or technological transfer (Caves, 1974) and the training of labor (Buckley & Casson, 1976) raise manufacturing productivity. MNCs also assist in increasing the productivity and product quality of domestic suppliers, and hence contribute in better use of economies of scale for local firms (Ozawa, 1975). Ozawa also indicated that MNCs bring knowledge and technology skills, which serve as catalyst for economic development and social enhancement, and nations could mutually benefit from interactions with each other in the aspect of trade and investment opportunities.
However, based on Dunning's (1981) informal framework, there are three mandatory conditions needed to attract FDI namely the ownership of knowledge-based assets; the presence of locational advantage (access to consumers, low input prices, transport costs, tariffs, quotas); as well as some advantages to producing internally instead of via licensing arrangements (incomplete contracts, asset specificity, bounded rationality, imperfect information, transaction costs, corporate governance). Markusen (1995) have formalized part of these concepts by emphasizing that knowledge-based assets could serve as a joint factor across plants, giving economies of scale at the level of the firm instead of at the plant level.
Knowledge-based assets consist of firm-specific processes and products, as well as intangibles like managerial and marketing skills. These are all channels through which FDI can provide productivity enhancement. Knowledge-based assets increase the risks related to serving a market via licensing, because potential competitors may gain access to technical secrets, and product quality may not be maintained as the licensee can be a free-rider on the firm's reputation (Barrell & Holland, 2000). Thus, if a company's competitive advantage relies on knowledge-based assets, it is more probably to invest itself than license (Barrell & Pain, 1997). This is crucial if we were to move from a production based on mass assembly of identical products to one that is more customized. Therefore, FDI has become an increasingly vital channel in terms of knowledge transferring.
Specifically, FDI could affect domestic manufacturing firms in two ways namely the output (manufacturing) growth, and wage growth. For the first case, FDI bring in benefits that cannot be completely captured by the firm, like technology. Although technology transfer exists via many different channels, FDI can play a crucial in a few contexts. New technology may not be commercially available and R&D firms may be reluctant to sell their technology through licensing agreements. Therefore, cooperating with R&D firms or closely linked to them may be the best way to learn new technology. Besides, FDI may also stimulate competition mandatory to spur technology diffusion, especially if domestic firms are protected from the import-substitution policy. FDI may also provide a type of employee training which cannot be conducted by local firms or buying overseas, like managerial skills. Technology diffusion could exist via labor turnover when local labors switch from MNCs to local firms (Harrison, 1994). All these will in turn result in the improvement in the labor productivity, and hence the manufacturing output growth.
In addition, the procedure of penetrating into overseas markets is very hard. In order to export, firms have to get information regarding foreign preferences and form distribution channels in overseas market. A clear means for firms to learn about export markets is to observe other exporters that are already experienced in selling overseas. These exporters could be other local firms, or MNCs. MNCs brings information regarding export markets to domestic producers, providing them the path to penetrate overseas market (Harrison, 1994). As result, the local producers can expand their production (economies of scale) and hence gain in terms of manufacturing output growth.
On the other hand, through what channel FDI influences host country's wage levels? Theories suggest two ways namely the 'wage differential effect' and the 'wage spillover effect'. The 'wage differential effect' is that wages offered by MNCs may be substantially different from those of domestic firms and hence the varying proportion of foreign firms in the domestic economy may influence domestic wage levels. MNCs may pay higher wages than domestic firms as they are bigger, more capital-intensive, operate in a more skill-intensive industry or hire more educated labors than domestic firms. There are also a few reasons why MNCs tend to pay higher wages than domestic ones for the same quality of worker. Firstly, MNCs may face segmented worker markets and greater labor costs because of various rules in the recipient country. There may be asymmetric information on domestic labor markets and have to pay higher wages to determine and induce labors with quality. Secondly, there may be a complicated set of issues to do with comparability of pay across nations. Internal fairness policies in MNCs may avoid huge wage-gaps between workers based on different place and hence increase worker payment in regions with low wage levels. Thirdly, MNCs may offer higher wages to compensate for specific disadvantages of multinational employment. For instance, employees may prefer domestic firms and the employment uncertainty may be greater than that in indigenous firms. Fourthly, multinational labors may have access to the 'knowledge capital' of MNCs and these technology or management skills will leak out to local competitors when workers switch jobs. To reduce leakage, MNCs tend to pay higher wages to decrease employee turnover. Besides, it is possible that MNCs pay lower wages than domestic firms if an MNC creates a monopsonist buyer of worker in a domestic labor market after crowding out the domestic firms.
There is strong empirical evidence reinforcing the claim that oversea firms pay higher wages than local firms, even after controlling for scale, labor quality, industry and regional features. The predicted wage inequality between MNCs and domestic firms are around 10 to 15 percent in the US (Lipsey, 1994); about 6 to 26 per cent in the UK (Girma, Greenaway, & Wakelin, 2001); etc.
While the 'wage spillover effect' is that the existence of MNCs may influence the wage level of domestic firms. Firstly, the competition of MNCs in both product and input markets may cause the wage levels of domestic firms to fluctuate. The competition may raise worker demand and this leads domestic firms to raise wages in order to induce workers with quality. However, if the competition drives domestic firms under the minimum efficiency scale, or even crowds out them, then the wage spillover effect is negative. Secondly, MNCs could create a positive knowledge spillover to domestic firms who may adopt new technologies introduced by MNCs via imitation. The labors previously work for the MNCs may change to domestic firms and hence transfer information to them. MNCs may transfer technology to firms that are potential suppliers of intermediate goods or purchasers of their own products. These technological spillovers may positively influence the productivity of domestic firms and hence enhance their wage levels.
However, the wage spillover effect has somewhat mixed empirical evidence. Aitken, Harrison and Lipsey's (1996) results showed a lack of wage spillover, but substantial wage differentials between oversea and local firms in Mexico and Venezuela. In the US, there is a small wage differential, but some evidence reinforcing the wage spillover effect. In addition, Girma et al. (2001) based on UK, failed to find overall wage spillover effect on wage levels, but a negative effect on wage growth. In contrast, Driffield and Girma's (2003) result showed positive wage-spillover effect, but this effect is confined to the region where FDI take places.
Policy makers may be more interested in the aggregate effect of FDI on regional average wage levels. Combining the 'wage spillover effect' and the 'wage differential effect', there is uncertain prediction on the direction of the domestic wage effect of FDI. There are only a few empirical researches available on the aggregate effect of FDI on regional average wage levels. For instance, Aitken et al. (1996) found that FDI tended to increase average industry wages in Mexico and Venezuela. Figlio and Blonigen (2000) studied country-industry data from South Carolina and found that foreign investment had increased domestic industry wages much more than local investment. Generally, the evidence is in favor of the positive effect of FDI on average wage levels.
Nonetheless, the direction of the causality relationship between FDI and recipient nation wages is ambiguous. The causality connection between FDI and recipient nation wages may go in inverse direction. Firstly, recipient wage levels may influence the location decision of FDI. The theory of MNCs suggests that the motive of vertical FDI is to exploit the international difference in input prices and to produce in a location with low input costs, suggesting that domestic wage levels may be one crucial factor of FDI. Empirical research of the role of worker costs in FDI location choice show mixed evidence (see: Wheeler & Mody, 1992; Billington, 1991; Head, Ries & Swenson, 1999). Nevertheless, as these research do not control for the difference in labor productivity, a positive impact of wages on FDI may still be tallied with MNCs being attracted by low costs for a certain labor productivity. After controlling for labor productivity, there is evidence that high labor costs significantly and negatively influence FDI (see: Culum, 1988; Friedman, Gerlowski, & Silberman, 1992).
Secondly, domestic firms with high wages are more probably to become 'foreign' if MNCs 'cherry pick' the most efficient domestic firms. A few empirical research reinforce this statement. For instance, Harris and Robinson (2002) found that MNCs take over the most efficient plants previously operated by UK enterprises and that productivity deteriorate after acquisition. The positive connection between FDI and recipient nation wages may merely because of MNCs' acquisition of high-wage (productivity) firms, while FDI has no effect on recipient nation wages.
Finally, foreign investors tend to invest in high wage industries in developed nations (Harrison, 1994). Within those industries, there is only a small difference in wages paid by MNCs and local firms. Furthermore, MNCs tend to be relatively huge, and big firms conventionally pay greater wages than the small ones. However, the wage inequality cannot be depicted by the notion that foreign FDI locate in high wage industries. The wage inequality between foreign firms and local firms is great even within the same industry. It was hypothesized (by Harrison) that MNCs simply employ all the best labors away from their local competitors. This implies that even if wages are higher in MNCs average wages do not increase with FDI inflows. Indeed, his results indicated that average wages do increase as a result of rises in FDI, suggesting that MNCs are not only employing the best labors. The higher wages paid by MNCs implies that they bring in new technology and ideas, increasing the productivity of their employees, and hence increases their wages level. Besides, Harrison also stated that MNCs tend to maintain their workers from leaving, particularly after making investment in human asset specificity. Higher wages is a mean of ensuring the loyalty of the workers. As a whole, it seems obvious that nations which attract FDI gain from at least one aspect, namely higher wages for workers of MNCs.
In short, the presence of FDI increases manufacturing output growth through technological diffusion and firms' learning from exporters (some are MNCs). The presence of FDI also increases manufacturing wage growth via 'wage differential effect' and 'wage spillover effect'. As a result, it is expected that FDI positively influences manufacturing output growth and wage growth.
3.3 Model Specification
FDI inflows from various sourcing nations may carry different effect on recipient nations' productivity due to the differences in FDI type and technological gap. Based on MIDA's annual media conference in 2012, it is noticed that Korea, Singapore and Saudi Arabia are among the largest FDI investors towards Malaysia and they are evidenced to have a beneficial effect on Malaysia's (especially in terms of creating job opportunity). However, how the FDI originate from technological leading nations like Japan and the US contribute to Malaysia's manufacturing performance is less examined although their investment in Malaysia is increasing recently. To evaluate the potential differences in the productivity effect of FDI between Japan and the US, this study incorporates the United States FDI (US FDI) and Japanese FDI (JPFDI) in number ii of both manufacturing output growth and manufacturing wage growth model.
On one hand, as explained in the previous subsection (3.2), FDI is expected to be positively affecting growth. On the other hand, LLB, LLBP and LCAP are incorporated in the model as controlled variables. Again, all three have positive expected priori sign.
3.3.1 Econometric model
In order to achieve the objectives of this study, two models were employed, namely the manufacturing output growth model and the manufacturing wage growth model.
Manufacturing output performance model
In the first equation, LFDI was incorporated as an independent variable in affecting the dependent variable (LMOG), controlled by LLB and LCAP. In the second equation, LUSFDI and LJPFDI were incorporated as explanatory variables in affecting the dependent variable (LMOG), controlled by LLB and LCAP. Note that LMOG represents natural log of manufacturing output growth; LFDI represents natural log of total FDI inflows in Malaysia; LUSFDI represents natural log of US's FDI into Malaysia; LJPFDI represents natural log of Japanese FDI into Malaysia; LLB represents natural log of number of labor employed; while LCAP represents natural log of total capital investment.
LMOG = f(LFDI,LLB,LCAP)
LMOG = f(LUSFDI, LJPFDI, LLB, LCAP)
Wage growth Model
In the first equation, LFDI was incorporated as an independent variable in influencing the dependent variable (LWG), controlled by LLBP and LCAP. In the second equation, LUSFDI and LJPFDI were incorporated as explanatory variables in affecting the dependent variable (LWG), controlled by LLBP and LCAP. Note that LWG represents natural log of manufacturing wage growth; LFDI represents natural log of total FDI inflows in Malaysia; LUSFDI represents natural log of US's FDI into Malaysia; LJPFDI represents natural log of Japan's FDI into Malaysia; LLBP represents natural log of manufacturing labor productivity; LCAP represents natural log of total capital investment.
LWG = f(LFDI, LLBP, LCAP)
LWG = f(LUSFDI, LJPFDI, LLBP, LCAP)
3.4 Empirical Methodology
After the models are being specified, stationarity tests of order of integration were conducted to avoid spurious regression. In this case, unit root tests such as the Augmented Dickey Fuller (ADF) test and the Philip-Perron (PP) test were employed to identify the stationarity property of the data or to confirm the order of integration I(d). Note that the linear combination of all variables must be I(d-b). The null hypothesis of both the ADF as well as PP tests is similar: the series are non-stationary. Besides, the null hypothesis can be rejected in favor of stationary if the t-statistic is greater than the critical value; otherwise do not reject the null hypothesis.
3.4.1 Estimation Methods
Next, estimation methods used in this study will be described. This study employs time series estimation approach where the length of period range from 1980 to 2010 (31 years), and it is intended to look at the overall effect of total FDI inflows, US FDI and Japanese FDI on the Malaysian manufacturing output growth and wage growth. By using time series, one can decompose a trend, a seasonal, a cyclical and an irregular component. The most significant purpose of time series method is to have estimation based on economic data. This research intends to test on how total FDI inflows, US's FDI and Japan's FDI affect manufacturing sector performance in Malaysia. There will be a further explanation on the model and Bound testing approach in the next section.
Bound testing approach
To examine any possible existence of long run relationship among the variables, autoregressive distributed lag (ARDL) bound test approach which introduced by Pesaran, Shin & Smith (2001) is employed. There are several benefits in employing the bounds testing procedure as compared to Jahansen and Juselius (JJ) multivariate cointegration test. First of all, it allows testing for the presence of a cointegrating relationship between variables in level form regardless of the variables' order of integration. Besides, it is applicable for small sample size research, unlike the JJ test. ARDL also can test for both short run and long run relationship among the variables. The following regression is the construction of Vector Auto-Regression (VAR) of order p (VAR (p)) from Pesaran et al. (2001) for the function explained previously.
(5)
The notation of is the vector for both and , where is the endogeneous variable. It includes manufacturing output growth and wage growth. On the other hand, refers to exogeneous variable in the model. It includes FDI, USFDI, JPFDI, LB, LBP and CAP. also serves as the vector matrix of explanatory variable. σ = [ , ]', time variable is express in t, and is the matrix of VAR parameter for lag i. must be I(1) variable or must be consist of 1 unit root, but the independent variables can be either I(1) or I(0) (Pesaran et al., 2001).
The VAR (p) model can be rewritten in vector error correction model (VECM) form as:
(6)
Where (difference operator). The long run multiplier matrix conformably with as below:
(7)
According to Pesaran et al. (2001), the first assumption of the roots of =0 are either outside the unit circle or satisfy Z=1. Followed by the third assumption, k-vector and the fourth assumption is the matrix has rank (r), 0≤ r ≤ k. The matrix is transform to
(8)
The diagonal elements of the matrix are unrestricted, so the selected series can be either I(0) or I(1). Provided that , then y is I(1), but if then y will be I(0).
To implement the bounds testing approach, it is important to examine whether the long run relationship exist in equation 1, and further proceeding to unrestricted error correction model (UECM) is needed.
(9)
The long run relationship can be estimated through the model above, and hence the statistical test to be used under the bound testing approach is F-test. The null hypothesis and alternative hypothesis are expressed as bellow:
Null hypothesis indicates that there is no long run levels relationship in the model; in contrast the alternative hypothesis shows that the existence of long run levels relationship in the model. The decision rule is reject the null hypothesis when the computed F-statistic (Wald test) is greater than the upper bound critical value; otherwise do not reject the null hypothesis (when F-statistic is lower than the lower bound critical value). According to Pesaran et al., the lower bound critical value assume that the independent variables are integrated of order zero, or I(0), while the upper bound critical value assume that the independent variables are integrated of order one, or I(1). As a result, if the computed F-statistic falls between the lower bound and upper bound critical values, it can be concluded that the result is inconclusive.
There are two sets of critical value namely generated from Pesaran et al. (2001) and Narayan (2005). Due to the small sample size in this study (only 31 observations), employing the critical value generated by Narayan (2005) is more appropriate.
In order to test the effect of various FDI (namely total FDI inflows, US's FDI, Japan's FDI) on manufacturing performance, the following unrestricted error correction models (UECM) of ARDL model is estimated:
Manufacturing Output Growth Model
(10a)
(10b)
Manufacturing Wage Growth Model
(11a)
(11b)
where represents disturbance term of ADRL model. The null hypothesis for testing long run relationship is: =0, and for equation 10 and 11 respectively. In contrast, the alternative hypothesis (which contradicts with the null hypothesis) indicates that at least one variable is not equal to zero.
Besides, a conditional error-correction model (ECM) is the dynamic adjustment of the ideas generation process, which can be used to test the existence of long run relationship using the ARDL bound test (Pesaran et al, 2001) and also test of the short run relationship. The ECM equation can be express as below:
Manufacturing Output Growth Model
(12a)
(12b)
Manufacturing Wage Growth Model
(13a)
(13b)
The ECT (error-correction term) in both model explained the speed of adjustment of toward the equilibrium, the rule of thumb of ECT must be in negative and significant in order to explain the short run dynamics toward the equilibrium. Furthermore, the equation of 14 and 15 is the long run ARDL level model (P,Q,R,S,T).The optimum lag length for the ARDL level model is based on the SBC information criteria.
Manufacturing Output Growth Model
(14a)
Manufacturing Wage Growth Model
(15a)
(15b)
The long run elasticity for both models can be formulated as bellow:
(16)
3.5 Data Description
For the comparative analysis, FDI (inflows) data from the US and Japan were collected. There are two reasons for choosing these two countries. First of all,
Variables
Definition
Source of data
MOG
Manufacturing output growth in Malaysia (manufacturing value added)
World Bank
WG
Manufacturing wage growth in Malaysia (wage)
Department of Statistic (DOS) Malaysia
FDI
Total foreign direct investment inflows of Malaysia, divided by nominal GDP
UNCTAD, world bank
USFDI
Malaysia's FDI inflows from the United States
Malaysian Industrial Development Authority (MIDA)
JPFDI
Malaysia's FDI inflows from Japan
Malaysian Industrial Development Authority (MIDA)
LB
Number of employed person in Malaysia manufacturing sector
Department of Statistic Malaysia (DOS)
LBP
Manufacturing labor productivity in Malaysia (value added divided by labor)
World Bank, Department of Statistic (DOS) Malaysia
CAP
Total Capital Investment in manufacturing sector
Malaysian Industrial Development Authority (MIDA)
CHAPTER 4: RESULTS AND INTERPRETATION
4.1 Introduction
This chapter reports the estimated results of the linkage among total foreign direct investments (FDI), US FDI, Japanese FDI, labor (and labor productivity), capital and Malaysia's manufacturing performance. First of all, unit root tests such as the Augmented Dickey-Fuller (ADF) and Philips Perron (PP) tests were used (as shown in Table 1) to test the data properties of the time series. The null hypothesis for both tests is similar, that is 'the series are non stationary' (at level form). And the decision rule is as the following; the null hypothesis can be rejected if the t-statistic is higher than the critical value. Hence, one can conclude that the variables are stationary. However, if one fails to reject the null hypothesis in the level form, then he or she has to proceed to the first differenced form. Besides, to avoid the autocorrelation problem in the ADF test and PP test, the Schwarz Information Criteria (SIC) and the New-West Banwidth criteria were referred to select the optimal lags length respectively. Note that robustness of results can be achieved when the error terms are ensured to be non-correlated.
Next, the ARDL bounds testing approach as depicted in table 2, 3, 4 and 5 was used to determine the short run and long run relationship for both empirical models. Finally, a summary regarding the empirical results serves as a closing session for this chapter.
4.2 Unit root test
Table 1 reports the results of Augmented Dickey Fuller (ADF) test and Philips-Perron (PP) test at both the level and first differenced forms. Both tests take into consideration the condition of constant with trend and constant without trend. As mentioned previously, the optimal number of lag length is determined by referring to the SIC (Schwarz Information Criteria) for the ADF test and New-West Banwidth criteria for the PP test. Note that PP test serves as a complementary test for the ADF test. In other words, it normally reinforces the results of the ADF test.
As shown in Table 1, the t-statistics in both the ADF test and the PP test (level form) for the variables LMOG, LFDI, LWG and LLB are statistically insignificant to reject the null hypothesis of one unit root at 5% significance levels. This indicates that they are non-stationary at level form, and it is suspected that they contain one or more unit root. Therefore, the proceeding to first differenced form is needed. In this case, the t-statistics for both the ADF test and the PP test showed that the null hypothesis of two unit roots has to be rejected and it is concluded that LMOG, LFDI, LWG and LLB have 1 unit root or integrated of order one I(1). Note that both dependent variables namely LMOG and LWG are I(1). This conforms to one of the criterions for the ARDL method: the dependent variable must be in one order of integration, though the independent variables can be either I(0) or I(1).
Table 4.1: Results of the Unit Roots Tests
Variable
ADF Test
PP
Level
First difference
Level
First difference
Constant
With Trend
Constant
No Trend
Constant
With Trend
Constant
No Trend
Data Period (1980-2010)
LMOG
LFDI
LUSFDI
LJPFDI
LWG
LCAP
LLB
LLBP
-0.4492(0)
-1.9789(0)
-3.7012(0)**
-5.4799(0)***
-1.1973(0)
-4.8269(0)***
-0.4607(1)
-5.1296(0)***
-4.6177(0)***
-5.8318(0)***
-7.8816(0)***
-9.2815(0)***
-5.8956(0)***
-7.3705(0)***
-5.7987(0)***
-7.2953(0)***
-0.4066(2)
-2.0107(4)
-3.7141(2)**
-5.4800(0)***
-1.2187(2)
-4.8190(4)***
-1.0021(1)
-5.4158(4)***
-4.5849(2)***
-5.8353(4)***
-8.97734(7)***
-18.1453(19)***
-5.8724(2)***
-17.6838(28)***
-5.8113(2)***
-10.1440(9)***
Note : The null hypothesis is that the series is non-stationary, or contains a unit root for ADF test and PP test. The rejection of null hypothesis for the stationary test is based on MacKinnon (1991) critical values.
Figures in parentheses ( ) refer to the selected lag length. The number of lags was selected based on Schwarz Information Criteria (for ADF test) and New-West Banwidth (for PP test) in order to avoid the problem of autocorrelation, that is to ensure that the error terms are uncorrelated and enhance the robustness of the results.
***, ** and * indicates the rejection of the null hypothesis of non-stationary at 1%, 5% and 10% significance level respectively.
On the other hand, the t-statistics in both the ADF test and the PP test (level form) for the variables LUSFDI, LJPFDI, LCAP and LLBP are statistically significant to reject the null hypothesis of one unit root at 5% significance levels. This indicates that they are stationary at level form, and it is concluded that they are I(0) variables. Therefore, the proceeding to the first differenced form is unnecessary.
In short, the result of the PP test is consistent with the one in the ADF test which indicates that the variables LMOG, LFDI, LWG and LLB contain one unit root(stationary at first differenced form) while the variables LUSFDI, LJPFDI, LCAP and LLBP contain no unit root (stationary at level form). In other words, the former group are I(1) while the latter group are I(0).
4.3 Bound Test (ARDL approach):
The bounds test was conducted after the stationary test to determine whether there is cointegration among the time series. It is the most adequate econometric technique to the long run and short run relationships among the total FDI, US FDI, Japan FDI, and the Malaysian manufacturing performance, given the relative tiny sample size in the study (Pesaran et al., 2001). As the data's length of period for this study is 1980-2010 (only 31 observations) which is less than 100 observations, the study will refer to the critical values computed by Narayan (2005) instead of the one computed by Pesaran et al. (2001). According to Pesaran and Shin (1995) as well as Pesaran et al., (2001), when the order of ARDL is identified, the the estimation and identification can be estimated using Ordinary Least Square (OLS). In addition, the asymptotic distribution of the F-statistic is non-standard under the null hypothesis of no cointegration among the variables under studied, regardless of the independent variables are purely I(0) and I(1), or a mixture of I(0) and I(1). One of the benefits of employing ARDL approach is that it is not a must to pretest the identification of the order of integration. In others words, it allows the I(0) and I(1) variables as regressors (independent variables).
Model 1: Manufacturing Output Growth
4.3.1: The Impact of total FDI on Manufacturing Output Growth (Model 1, Equation 1)
Table 2 shows the ARDL model based on Model 1, equation 1, using the Hendry's general to specific method (Pesaran et al., 2001). The value of adjusted R-squared and standard error of regression is equal to 0.3645 and 0.0284 respectively. Besides, several diagnostic checkings like Breusch-Godfrey LM test, ARCH test, Jacque-Bera test and Ramsey RESET test were conducted to ensure the validity of the model which free from the problem of serial autocorrelation, heteroscedasticity, non-normality and model specification correspondingly. Generally, probabiblity value (p-value) for all tests shows insignificancy, suggesting that the model is less (or even free from) associated with those problem mentioned above.
Table 4.2: The Estimated ARDL Model based on Model 1, Equation 1
(Total FDI)
Variable
Coefficient
t-statistic
Probability
C
LMOG(-1)
LFDI(-1)
LCAP(-1)
LLB(-1)
D(LMOG(-1))
D(LMOG(-2))
D(LFDI)
D(LFDI(-1))
D(LFDI(-2))
D(LCAP)
D(LCAP(-1))
D(LLB(-2))
D(LLB(-3))
D(LLB(-4))
Adjusted R-squared
Standard Error of Regression
F-statistic
Prob(F-statistic)
1.3326
-0.1833
0.1939
-0.1308
0.0172
-0.2887
-0.0213
0.2007
-0.0311
-0.1160
-0.1690
0.0084
-0.3050
0.0275
0.3245
2.5652
-1.2770
3.6574***
-1.5168
0.0857
-1.2127
-0.0739
3.4015*
-0.5913
-1.9285*
-1.4805
0.2487
-1.2059
0.1210
1.3695
0.3645
0.0284
2.0241
0.1226
0.0263
0.2279
0.0038
0.1575
0.9332
0.2506
0.9424
0.0059
0.5663
0.0800
0.1668
0.8082
0.2531
0.9059
0.1982
II. Diagnostic Checking
Autocorrelation (Breusch-Godfrey Serial Correlation LM test)
F(2)= 0.6792[0.5312] F(4)=1.2807[0.3627]
Heteroscedasticity ARCH Test:
F(2)=1.4563 [0.2557] F(4)=0.9320 [0.4688]
Jacque-Bera Normality Test : χ2 (2)= 1.3813[0.5013]
Ramsey RESET specification Test :
F-statistic = 1.1765[0.3516] Number of fitted terms=2
Note: ***, **and * denote significancy at 1%, 5% and 10% significance levels.
Figures in squared parentheses [ ] refer to marginal significance level.
For both Breusch-Godfrey LM test and ARCH test, we are testing for serial
correlation and heteroscedasticity at the significance level ranging from the first to the fourth order.
With respect to the first step (bounds test), the result of bound cointegration test for model 1, equation 1 was shown in Table 3. The test clearly demonstrates that the null hypothesis of β11= β12= β13=β14=0 against the alternative hypothesis of β11≠β12≠β13≠β14≠0 is rejected at 10% significance level. The computed F-statistic (Wald test) 4.2371 is greater than the upper bound critical value (4.150). Therefore, based on the test results, it is concluded that there exists a cointegration among manufacturing output growth, total FDI inflows, capital, and labor, suggesting that these variables cannot deviate substantially from each other and indirectly, it can be concluded that disequilibrium between them is just a short run phenomenon.
In the second step (ARDL level relation), both the short run and long run elasticities of manufacturing output growth with respect to total FDI inflows, capital, and labor were reported in Table 4.
Table 4.3: Bound Test based on Model 1, Equation 1
Computed F-statistic: 4.2371*
Null hypothesis: No Cointegration
Critical value
(K=3)
Critical Value
(K=4)
Lower Upper
Lower Upper
1% significance level
5.333 7.063
4.768 6.670
5% significance level
3.710 5.018
3.354 4.774
10% significance level
3.008 4.150
2.752 3.994
Note: The critical values were taken from Narayan (2001), Table CI (iii):
Unrestricted intercept and no trend.
***, ** and * denote significancy at 1%, 5% and 10% significance level.
In the short run (second column), it was found that LFDI (0.0519) is the only explanatory variable that has a significant effect on manufacturing output in the short run. As consistent with the expected priori sign, it has a positive sign. However, LCAP (-0.0237) and LLB (0.1790) do not play any role in influencing the dependent variable.
In addition, the error correction term (ECT) is significant (in 1% significance level) and has a negative sign. This is in line with the conventional criteria in the ARDL short run estimation where the ECT must be significant and is negative in value for the variables to be converged as a result of a deviation from the long run. The coefficient of the ECT, also known as the speed of adjustment is -0.1619, meaning that when the variables deviate from the fundamental (long run), they will converge by 16.19% yearly (note : this is an annual data based study). As a result, they need 6.18 years to fully converge (100% convergence).
In the long run, it was found that two regressors significantly and positively influence Malaysia's manufacturing output growth namely LFDI and LLB, with the respective estimated coefficients of 0.3203 and 1.1056. The result (especially for LFDI) is consistent with results of some previous empirical research (Chandran & Krishnan, 2008; Al-Zu'bi et al., 2012). The estimated coefficients imply that in the long run, a 1% increase in total FDI inflows and number of labor employed in manufacturing sector will lead to a 0.3203% and 1.1056% respective increase in Malaysia's manufacturing output growth. However, it was found that LCAP does not have a significant impact on Malaysia's manufacturing value added.
Table 4.4: Long Run Coefficients using the ARDL Approach Model 1,
Equation 1 (1,0,0,0)
Variables
Coefficient
Standard Error
T-Ratio[Prob]
LFDI
0.3203*
0.1742
1.8384[0.078]
LCAP
-0.1462
0.1766
-0.8278[0.416]
LLB
1.1056**
0.4444
2.4878[0.020]
INPT
2.4678
2.0689
1.1928[0.245]
The Short Run Dynamics using the ARDL Approach
dLFDI
0.0519**
0.0250
2.0714[0.049]
dLCAP
-0.0237
0.0251
-0.9418[0.356]
dLLB
0.1790
0.1081
1.6564[0.111]
ECM(-1)
-0.1619***
0.0550
-2.9438[0.007]
dINPT
0.3995
0.2864
1.3947[0.176]
Diagnostic checking
Serial Correlation=1.3305[0.293]
Heteroscedasticity=0.2471[0.908]
Note: ***, **, and * refer to significant level at 1%, 5% and 10% respectively.
Table 5 depicts the ARDL model based on Model 1, equation 2, using the Hendry's general to specific method (Pesaran et al., 2001). The value of adjusted R-squared and standard error of regression is equal to 0.7359 and 0.0181 respectively. Besides, several diagnostic checkings like Breusch-Godfrey LM test, ARCH test, Jacque-Bera test and Ramsey RESET test were conducted to ensure the validity of the model which free from the problem of serial autocorrelation, heteroscedasticity, non-normality and model specification correspondingly. Generally, probabiblity value (p-value) for all tests shows insignificancy, suggesting that the model is less (or even free from) associated with those problem mentioned above.
4.3.2: The Impact of US's and Japan's FDI on Manufacturing Output Growth (Model 1, Equation 2)
Table 4.5: The Estimated ARDL Model based on Model 1, Equation 2
(USFDI & JPFDI)
Variable
Coefficient
t-statistic
Probability
C
LMOG(-1)
LUSFDI(-1)
LJPFDI(-1)
LCAP(-1)
LLB(-1)
D(LMOG(-1))
D(LMOG(-2))
D(LUSFDI(-1))
D(LUSFDI(-2))
D(LUSFDI(-3))
D(LJPFDI)
D(LJPFDI(-1))
D(LJPFDI(-2))
D(LJPFDI(-3))
D(LCAP)
D(LCAP(-1))
D(LCAP(-2))
D(LCAP(-3))
D(LLB())
D(LLB(-1))
Adjusted R-squared
Standard Error of Regression
F-statistic
Prob(F-statistic)
-1.5962
-0.7143
-0.3589
-0.0452
-0.2228
1.9730
-0.9791
-1.8979
0.2204
0.1210
0.0662
0.0719
0.2283
0.1433
0.0527
-0.0753
0.1273
0.0194
-0.0344
0.4363
-0.7875
-3.5103**
-4.5336***
-4.9614***
-0.8082
-2.0149*
6.0995***
-3.9699***
-5.4627***
3.6434**
2.3880*
2.1822*
2.2945*
3.6933**
3.2470**
2.2124*
-2.0111*
2.0262*
0.4936
-1.5473
2.4652**
-2.4465**
0.7359
0.0181
4.6231
0.0329
0.0127
0.0040
0.0025
0.4498
0.0905
0.0009
0.0074
0.0016
0.0108
0.0542
0.0718
0.0616
0.0102
0.0175
0.0689
0.0910
0.0891
0.6392
0.1728
0.0488
0.0500
II. Diagnostic Checking
Autocorrelation (Breusch-Godfrey Serial Correlation LM test)
F(2)= 0.4862[0.6471] F(4)=0.5152[0.7425]
Heteroscedasticity ARCH Test:
F(2)=1.4091 [0.2656] F(4)=0.6805 [0.6144]
Jacque-Bera Normality Test : χ2 (2)= 0.3756[0.8287]
Ramsey RESET specification Test :
F-statistic = 0.7470[0.5301] Number of fitted terms=2
Note: ***, **and * denote significancy at 1%, 5% and 10% significance levels.
Figures in squared parentheses [ ] refer to marginal significance level. For both Breusch-Godfrey LM test and ARCH test, we are testing for serial correlation and heteroscedasticity at the significance level ranging from the first to the fourth order.
With respect to the first step (bounds test), the result of bound cointegration test for model 1, equation 2 was shown in Table 6. The test clearly demonstrates that the null hypothesis of β21= β22= β23= β24= β25=0 against the alternative hypothesis of β21≠β22≠β23≠β24≠β25≠0 is easily rejected at 1% significance level. The computed F-statistic (Wald test) 7.9173 is greater than the upper bound critical value (6.670). Therefore, based on the test results, it is concluded that there exists a cointegration among manufacturing output growth, US's FDI, Japan's FDI, total capital, and number of labor employed, suggesting that these variables cannot deviate substantially from each other and indirectly, it can be concluded that disequilibrium between them is just a short run phenomenon.
Table 4.6: Bound Test based on Model 1, Equation 2
Computed F-statistic: 7.9173***
Null hypothesis: No Cointegration
Critical value
(K=3)
Critical Value
(K=4)
Lower Upper
Lower Upper
1% significance level
5.333 7.063
4.768 6.670
5% significance level
3.710 5.018
3.354 4.774
10% significance level
3.008 4.150
2.752 3.994
Note: The critical values were taken from Narayan (2001), Table CI (iii):
Unrestricted intercept and no trend.
***, ** and * denote significancy at 1%, 5% and 10% significance level.
In the second step (ARDL level relation), both the short run and long run elasticities of manufacturing output growth with respect to the US's FDI, Japan's FDI, total capital, and number of labor employed were reported in Table 7.
In both the short run and long run, the empirical result showed that there is only one variable namely the LLB has significant impact on manufacturing output growth, with the respective elasticities 0.2794 and 1.8851. As consistent with the expected priori sign, the coefficient is positive in value. The 1.8851 long run coefficient implies that a 1% increase in number of labor employed in manufacturing sector will lead to a 1.8851% increase in Malaysia's manufacturing output growth. However, all other independent variables do not play any role in influencing manufacturing wage growth both in the long run and short run.
Furthermore, the error correction term (ECT) is significant (at 5% significance level) and has a negative sign. This is in line with the conventional criteria in the ARDL short run estimation where the ECT must be significant and is negative in value for the variables to be converged as a result of a deviation from the long run. The coefficient of the ECT is -0.1482, meaning that when the variables deviate from the fundamental (long run), they will converge by 14.82% yearly. As a result, they need 6.75 years to fully converge (100% convergence).
Table 4.7: Long Run Coefficients using the ARDL Approach Model 1
Equation 2 (1,0,0,0,0)
Variables
Coefficient
Standard Error
T-Ratio[Prob]
LUSFDI
-0.1643
0.1667
-0.9855[0.335]
LJPFDI
-0.0859
0.1473
-0.5835[0.565]
LCAP
0.0171
0.1662
0.1031[0.919]
LLB
1.8851***
0.4526
4.1652[0.000]
INPT
-0.5405
2.0117
-0.2687[0.791]
The Short Run Dynamics using the ARDL Approach
dLUSFDI
-0.0243
0.0228
-1.0697[0.296]
dLJPFDI
-0.0127
0.0218
-0.5833[0.565]
dLCAP
0.0025
0.0250
0.1031[0.920]
dLLB
0.2794**
0.1158
2.4120[0.024]
ECM(-1)
-0.1482**
0.0586
-2.5273[0.019]
dINPT
-0.0811
0.3037
-0.2638[0.794]
Diagnostic checking
Serial Correlation=1.6532[0.215]
Heteroscedasticity=0.1273[0.881]
Note: ***, **, and * refer to significant level at 1%, 5% and 10% respectively.
Model 2: Manufacturing Wage growth
4.3.3: The Impact of Total FDI Inflows on Manufacturing Wage Growth (Model 2, Equation 1)
Table 8 shows the ARDL model based on Model 2, equation 1, using the Hendry's general to specific method (Pesaran et al., 2001). The value of adjusted R-squared and standard error of regression is equal to 0.4490 and 0.0389 respectively. Moreover, several diagnostic checkings like Breusch-Godfrey LM test, ARCH test, Jacque-Bera test and Ramsey RESET test were conducted to ensure the validity of the model which free from the problem of serial autocorrelation, heteroscedasticity, non-normality and model specification correspondingly. Generally, probabiblity value (p-value) for all tests shows insignificancy, suggesting that the model is less (or even free from) associated with those problem mentioned above.
Table 4.8: The Estimated ARDL Model based on Model 2, Equation 1
(Total FDI)
Variable
Coefficient
t-statistic
Probability
C
LWG(-1)
LFDI(-1)
LCAP(-1)
LLBP(-1)
D(LWG(-1))
D(LWG(-2))
D(LFDI)
D(LFDI(-1))
D(LCAP)
D(LCAP(-1))
D(LCAP(-2))
D(LCAP(-3))
D(LCAP(-4))
D(LLBP(-1))
D(LLBP(-2))
Adjusted R-squared
Standard Error of Regression
F-statistic
Prob(F-statistic)
-0.2172
-0.1789
0.1033
0.5321
-0.6643
-0.7693
-0.4751
0.1410
0.0906
0.2004
-0.2512
-0.1277
-0.0618
0.0109
-0.0702
0.2331
-0.2827
-1.3944
1.8442*
2.7490**
-2.4429**
-3.2595***
-2.0407*
1.8826*
1.4087
3.5682***
-2.0965*
-1.4286
-1.0124
0.2556
-0.2445
0.7376
0.4490
0.0389
2.3582
0.0871
0.7832
0.1934
0.0949
0.0205
0.0347
0.0086
0.0686
0.0891
0.1892
0.0051
0.0624
0.1864
0.3352
0.8035
0.8118
0.4777
II. Diagnostic Checking
Autocorrelation (Breusch-Godfrey Serial Correlation LM test)
F(2)= 0.7122[0.5192] F(4)=1.4197[0.3334]
Heteroscedasticity ARCH Test:
F(2)=0.7697 [0.4758] F(4)=0.5711 [0.6972]
Jacque-Bera Normality Test : χ2 (2)= 3.5265[0.1715]
Ramsey RESET specification Test :
F-statistic = 3.4226[0.1268] Number of fitted terms=6
Note: ***, **and * denote significancy at 1%, 5% and 10% significance levels.
Figures in squared parentheses [ ] refer to marginal significance level. For both Breusch-Godfrey LM test and ARCH test, we are testing for serial correlation and heteroscedasticity at the significance level ranging from the first to the fourth order.
With respect to the first step (bounds test), the result of bound cointegration test for model 2, equation 1 was shown in Table 9. The test clearly demonstrates that the null hypothesis of γ11= γ12= γ13= γ14=0 against the alternative hypothesis of γ11≠γ12≠γ13≠γ14≠0 is rejected at 5% significance level. The computed F-statistic (Wald test) 5.4780 is greater than the upper bound critical value (5.018). Therefore, based on the test results, it is concluded that there exists a cointegration among manufacturing wage growth, total FDI inflows, total capital, and number of labor employed, suggesting that these variables cannot deviate substantially from each other and indirectly, it can be concluded that disequilibrium between them is just a short run phenomenon.
Table 4.9: Bound Test based on Model 2, Equation 1
Computed F-statistic: 5.4780**
Null hypothesis: No Cointegration
Critical value
(K=3)
Critical Value
(K=4)
Lower Upper
Lower Upper
1% significance level
5.333 7.063
4.768 6.670
5% significance level
3.710 5.018
3.354 4.774
10% significance level
3.008 4.150
2.752 3.994
Note: The critical values were taken from Narayan (2001), Table CI (iii):
Unrestricted intercept and no trend.
***, ** and * denote significancy at 1%, 5% and 10% significance level.
In the second step (ARDL level relation), both the short run and long run elasticities of manufacturing output growth with respect to total FDI inflows, total capital, and number of labor employed were reported in Table 10.
In the short run (second column), it was found that LFDI (0.0624) is the only explanatory variable that has a significant effect on manufacturing wage growth. As consistent with the expected priori sign, it has a positive sign. However, LCAP (0.0167) and LLBP (0.1279) do not play any role in influencing the dependent variable.
Furthermore, the ECT is significant (in 5% significance level) and has a negative sign. This is in line with the conventional criteria in the ARDL short run estimation where the ECT must be significant and is negative in value for the variables to be converged as a result of a deviation from the long run. The the speed of adjustment is -0.1559, meaning that when the variables deviate from the fundamental (long run), they will converge by 15.59% per annum. As a result, they need 6.41 years to fully converge (100% convergence).
In the long run, it was found that all regressors including LFDI do not have impact on Malaysia's manufacturing wage growth. The result (especially for LFDI) is consistent with results of some previous empirical research (Axarloglou & Pournarakis, 2007; Waldkirch, 2010) where they found poor or even no FDI effect on manufacturing wage.
Table 4.10: Long Run Coefficients using the ARDL Approach Model 2,
Equation 1(1,0,0,0)
Variables
Coefficient
Standard Error
T-Ratio[Prob]
LFDI
0.4004
0.2587
1.5468[0.134]
LCAP
0.1074
0.2523
0.4257[0.674]
LLBP
0.8201
0.7671
1.0691[0.295]
INPT
-0.1718
3.0167
-0.0569[0.955]
The Short Run Dynamics using the ARDL Approach
dLFDI
0.0624*
0.0362
1.7224[0.097]
dLCAP
0.0167
0.0389
0.4308[0.670]
dLLBP
0.1279
0.1622
0.7881[0.438]
ECM(-1)
-0.1559**
0.0695
-2.2443[0.034]
dINPT
-0.0268
0.4773
-0.0561[0.956]
Diagnostic checking
Serial Correlation=1.7230[0.201]
Heteroscedasticity=2.4600[0.108]
Note: ***, **, and * refer to significant level at 1%, 5% and 10% respectively.
4.3.4: The Impact of US's and Japan's FDI on Manufacturing Wage Growth (Model 2, Equation 2)
Table 11 shows the ARDL model based on Model 2, equation 2, using the Hendry's general to specific method (Pesaran et al., 2001). The value of adjusted R-squared and standard error of regression is equal to 0.5183 and 0.0357 respectively. In addition, several diagnostic checkings like Breusch-Godfrey LM test, ARCH test, Jacque-Bera test and Ramsey RESET test were conducted to ensure the validity of the model which free from the problem of serial autocorrelation, heteroscedasticity, non-normality and model specification correspondingly. Generally, probabiblity value (p-value) for all tests shows insignificancy, suggesting that the model is less (or even free from) associated with those problem mentioned above.
In the second step (ARDL level relation), both the short run and long run elasticities of manufacturing output growth with respect to total FDI inflows, capital, and labor were reported in Table 4.
In the short run (second column), it was found that LFDI (0.0519) is the only explanatory variable that has a significant effect on manufacturing output in the short run. As consistent with the expected priori sign, it has a positive sign. However, LCAP (-0.0237) and LLB (0.1790) do not play any role in influencing the dependent variable.
In addition, the error correction term (ECT) is significant (in 1% significance level) and has a negative sign. This is in line with the conventional criteria in the ARDL short run estimation where the ECT must be significant and is negative in value for the variables to be converged as a result of a deviation from the long run. The coefficient of the ECT, also known as the speed of adjustment is -0.1619, meaning that when the variables deviate from the fundamental (long run), they will converge by 16.19% yearly (note : this is an annual data based study). As a result, they need 6.18 years to fully converge (100% convergence).
In the long run, it was found that two regressors significantly and positively influence Malaysia's manufacturing output growth namely LFDI and LLB, with the respective estimated coefficients of 0.3203 and 1.1056. The result (especially for LFDI) is consistent with results of some previous empirical research (Chandran & Krishnan, 2008; Al-Zu'bi et al., 2012). The estimated coefficients imply that in the long run, a 1% increase in total FDI inflows and number of labor employed in manufacturing sector will lead to a 0.3203% and 1.1056% respective increase in Malaysia's manufacturing output growth. However, it was found that LCAP does not have a significant impact on Malaysia's manufacturing value added.
Table 4.11: The Estimated ARDL Model based on Model 2, Equation 2
(USFDI & JPFDI)
Variable
Coefficient
t-statistic
Probability
C
LWG(-1)
LUSFDI(-1)
LJPFDI(-1)
LCAP(-1)
LLBP(-1)
D(LWG(-1))
D(LWG(-2))
D(LWG(-3))
D(LUSFDI)
D(LJPFDI)
D(LJPFDI(-1))
D(LCAP)
D(LCAP(-1))
D(LCAP(-2))
D(LLBP)
D(LLBP(-1))
D(LLBP(-2))
Adjusted R-squared
Standard Error of Regression
F-statistic
Prob(F-statistic)
-0.7804
-0.2284
0.0201
0.1526
0.4911
-0.3187
-0.3931
-0.2381
-0.5208
-0.1239
0.0337
-0.0865
0.2115
-0.2716
-0.0838
0.7663
0.4617
0.7905
-1.1237
-1.3387
0.2409
1.5164
4.2343
-0.7594
-1.1318
-0.8063
-1.9581
-2.6517
0.6252
-1.7853
4.7800
-3.9653
-1.8003
2.5236
1.3283
2.4605
0.5183
0.0357
2.6456
0.0700
0.2902
0.2135
0.8150
0.1637
0.0022
0.4671
0.2870
0.4409
0.0819
0.0264
0.5474
0.1079
0.0010
0.0033
0.1053
0.0326
0.2168
0.0361
II. Diagnostic Checking
Autocorrelation (Breusch-Godfrey Serial Correlation LM test)
F(2)= 0.3646[0.7069] F(4)=3.3159[0.1104]
Heteroscedasticity ARCH Test:
F(2)=0.0803 [0.9231] F(4)=0.2283 [0.9189]
Jacque-Bera Normality Test : χ2 (2)= 1.3677[0.5072]
Ramsey RESET specification Test :
F-statistic = 0.5208[0.6154] Number of fitted terms=2
Note: ***, **and * denote significancy at 1%, 5% and 10% significance levels.
Figures in squared parentheses [ ] refer to marginal significance level. For both Breusch-Godfrey LM test and ARCH test, we are testing for serial correlation and heteroscedasticity at the significance level ranging from the first to the fourth order.
With respect to the first step (bounds test), the result of bound cointegration test for model 2, equation 1 was shown in Table 12. The test clearly demonstrates that the null hypothesis of γ21= γ22= γ23= γ24= γ25=0 against the alternative hypothesis of γ21≠γ22≠γ23≠γ24≠γ25=0 is rejected at 5% significance level. The computed F-statistic (Wald test) 5.1826 is greater than the upper bound critical value (4.774). Therefore, based on the test results, it is concluded that there exists a cointegration among manufacturing wage growth, US's FDI, Japan's FDI, total capital, and number of labor employed, suggesting that these variables cannot deviate substantially from each other and indirectly, it can be concluded that disequilibrium between them is just a short run phenomenon.
Table 4.12: Bound Test based on Model 2, Equation 2
Computed F-statistic: 5.1826**
Null hypothesis: No Cointegration
Critical value
(K=3)
Critical Value
(K=4)
Lower Upper
Lower Upper
1% significance level
5.333 7.063
4.768 6.670
5% significance level
3.710 5.018
3.354 4.774
10% significance level
3.008 4.150
2.752 3.994
Note: The critical values were taken from Narayan (2001), Table CI (iii):
Unrestricted intercept and no trend.
***, ** and * denote significancy at 1%, 5% and 10% significance level.
In the second step (ARDL level relation), both the short run and long run elasticities of manufacturing wage growth with respect to US's FDI, Japan's FDI, total capital, and number of labor employed were reported in Table 13.
In the short run (second column), it was found that LCAP (0.1134) and LLBP (0.6675) have significant effect on manufacturing wage growth. As consistent with expected priori sign, they have a positive sign. However, LUSFDI (-0.0337) and LJPFDI (0.0076) do not play any role in influencing the dependent variable.
In addition, the error correction term (ECT) is significant (in 5% significance level) and has a negative sign. This is in line with the conventional criteria in the ARDL short run estimation where the ECT must be significant and is negative in value for the variables to be converged as a result o
