The estimated model for the productivity effect of FDI on the primary, manufacturing and services sectors, based on a simple bivariate autoregressive framework, can be specified in the linear form for the primary sector, for instance, as follows:
LYprit = σୀଵȽLYprit-ί + σୀଵߚLFDIprit- j + μt (6.2) where LYprit refers to the log of productivity in the primary sector; LFDIprit is inflows to the primary sector; α, β are the parameters to estimate; and μt is the error term. The same model is applied to estimate the direction of causality between the productivity of the manufacturing sector and FDI inflows in this sector and between the productivity of the services sector and FDI inflows in this sector. In equation (6.2), testing for Granger causality between LYprit and LFDIprit-j involves testing whether the lagged information on the variable LFDIpri provides any statistically significant information on the variable LYpri, in the presence of lagged LYpri. If not, then LFDIpridoes not Granger cause LYpri. A priori, it is expected that βj>0 or <0, depending on the positive or negative effects that exist between the FDI inflows and productivity output of the sectors.
6.5 Data and Methodology
This section presents the data and methodology used to show the productivity effects of FDI on the primary, manufacturing and services sectors, and the causality between each
respective sector’s productivity and FDI inflows. The study employs annual data for the
period from 1985 to 2010. The methodology used includes the stepwise approach and the pair-wise Granger causality. The stationary test for time series data is also employed. Detailed discussion of the data and method is provided in the subsequent subsections.
6.5.1Data
The productivity model analyses the productivity effects of FDI inflows on the primary, manufacturing and services sectors and it incorporates other factors for the augmented productivity output framework. The constructing of accurate and comparable measures of FDI inflows by sector for the Solomon Islands, has resulted in data being collected from various sources, which include annual data series on gross value added in primary (LYpri), manufacturing (LYmanu) and services (LYser); and the ratio of school enrolments in primary and secondary are extracted from the World Bank, World Development Indicators (2011). 54 Data on foreign direct investment inflows and exports in the primary, manufacturing and services sectors; and gross domestic product per capita are sourced from the Central Bank of Solomon Islands. To capture institutional quality and stability, data for economic freedom index are from the Heritage Foundation offering the freedom of investment and levels of bureaucratic quality of the economy are used (see Appendix Table A6.1).
To capture the sectoral effects of both domestic and foreign exports on total output and productivity, data for log, fish, copra, cocoa and palm oil exports are from CBSI (various) were constructed for primary exports. Data for transportation and tourism receipts were computed for export services, and data on canned fish and beverages and tobacco exports are the manufacturing exports. Total export of each sector is calculated as a ratio to total exports. All variables are transformed into logarithmic value.55
The growth rate of output or productivity measured by gross value added in the primary, manufacturing and services sectors are in constant prices (Solomon Islands dollar). The assumption is predicated on the belief that higher FDI inflows may result in a positive effect and spillover into sectoral productivity. In order to capture the sectoral productivity base, which is the dependent variable, this study computes the productivity efficiency unit, by dividing the gross value added in each sector (i.e. primary, manufacturing and services) by the sectoral labour, following the study by Tondl and Fornero (2008).56
54 The gross value added in agriculture is used as a proxy for gross value added in the primary sector.
55See Appendix Table A6.1 for data sources and Appendix Tables A6.2- A6.4 for descriptive statistics
.
56The sectoral productivity base following Tondl and Fornero (2008) is calculated as; LY
it = ΣYit/ΣLabour,
where LYit is log in total output/ productivity for the respective sectors (that includes both the domestic firms
and foreign firms) at time t. Labour is the number of each worker in each sector. Several previous studies have used either the share of employment, or output, as a proxy to measure the productivity effects of FDI (see Alfaro, 2003; Prüfer and Tondl, 2008; Tondl and Fornero, 2008).
6.5.2Econometric Methodology
To estimate the productivity effects of FDI by key sectors (primary, manufacturing and services) and the possible spillover effects of FDI inflows the stepwise procedure has been utilised. The direction of the relationship between sector productivity and FDI inflows is examined using the Granger causality method. A test for unit root is also included. Detailed discussion of the method and process is provided next.
Stepwise Procedure
This study follows the stepwise approach noted by Faraway (2002) for models (6.1a) to (6.1c) to estimate the equations on the productivity effects of FDI inflows. One of the advantages of the stepwise procedure is that it is “the simplest of all variable selection
procedures and can be easily implemented” (Faraway, 2002, p.126). The process involves
either backward elimination or forward selection, or a combination of both. Backward elimination involves starting with all explanatory variables and testing them one by one for statistical significance and deleting the insignificant ones. Forward selection, in contrast, begins with no variables in the model and tests the variables one by one and only includes them if they are statistically significant (Faraway, 2002). Various studies which have used this approach include Akinlo and World Institute for Development Economics Research (2005), and Tondl and Fornero (2008), where they add the explanatory variables to the model one by one and retained only those variables that were statistically significant. The approach used in this study undertakes a combination of both backward elimination and forward selection. The advantage of using both backward elimination and forward selection is that the variables can be added or removed early in the process and it easily allows for changes later (Faraway, 2002). In other words, there is the flexibility of dropping or adding a variable, but has some major drawbacks. Faraway (2002) notes that adding or dropping variables one at a time may have the possibility of missing the optimal, including ambiguous results on the validity of p-values, due to multiple testing.
The possibility of biased and inconsistent results arising from endogenity and multicollinearity problems may also occur (Tondl and Fornero, 2008). To avoid these problems, this study follows the approach undertaken by Tondl and Fornero (2008) and Busse and Hefeker (2005), where the models include the lagged dependent variables in levels on the right side hand of each equation, as instruments to dramatically increase the
efficiency of the estimation.57 Furthermore, the sector specific effects were controlled for and an interaction term and a dummy variable were included in order to avoid bias and inconsistent results.
Although the approach is similar to Busse and Hefeker (2005) and Tondl and Fornero (2008) it differs in terms of country specific study here. Whilst Tondl and Fornero use panel data to examine the productivity effects of FDI on various sectors for the Latin American countries, Busse and Hefeker examine that for cross sectional study for 83 developing countries and this study uses time series data for the Solomon Islands.
Granger Causality
The next step is to estimate the direction of the relationship between each sector’s
productivity and FDI inflows, using the pairwise Granger causality method. This similar approach used in Chapters 4 and 5 will be applied to model (6.2), but this approach differs in two aspects. Firstly, whilst the Granger causality test in Chapters 4 and 5 examines a three variable model (trivariate), this chapter will examine two variable (bivariate) approach. The bivariate causality equation takes the following form:
∆LYprit =σୀଵȽοLYprit-i +σୀଵߚ݆∆LFDIprit-j + μ1t (6.3a) ∆LFDIprit = σୀଵߣI∆LFDIprit–i + σୀଵɁj∆LYprit-j + μ2t (6.3b) The disturbances or the error terms U1t and U2t are assumed to be uncorrelated. Equation (6.3a) shows that current productivity in the primary sector (LYpri) is related to both its own past values and that of foreign direct investment in that sector (LFDIpri) and equation (6.3b) also shows similar behaviour of FDI and current productivity of that sector. The hypotheses formulated from equation (6.3a) state that LFDIpri does not Granger cause LYpri if H0: ߚ1ί = 0 against H1:ߚ1ί ≠ 0. If the hypothesis is rejected, it can be said that FDI Granger-causes GDP. The reverse hypothesis of Granger causality from GDP to FDI, is given as: LYpri does not Granger cause LFDIpri if H0: δj1= 0 against H1: δj1 ≠ 0.
Theoretically, the priori expectation is that the line of causality, equation (6.3a) for instance, could be justified, based on the fact that higher FDI inflows in the primary sector
will have a positive effect on the sector’s productivity. Inversely, in equation (6.3b), a
priori expectation is that higher primary productivity will induce FDI inflows. A similar
57To reduce the problem of autocorrelation, Busse and Hefeker used the lagged dependent variable on the
right hand side of the regression equations. They note that this procedure is “theoretically plausible as foreign investment in the previous period is highly relevant for FDI in the current period” (Busse and Hefeker 2005, p.19). Such method provides valid results as shown by Busse and Hefeker (2005) and Tondl and Fernero (2008) where the lagged FDI variable is always highly significant in their regressions.
approach will test for the causality between productivity in manufacturing (LYmanu) and FDI inflows in manufacturing sector (LFDImanu) and between productivity in services (LYser) and FDI inflows in services sector (LFDIser).
Prior to applying the stepwise and Granger causality methods, it is important to check for the stationary of each variable used in the models. In the regression analysis and the Granger causality, the variables are only consistent and unbiased if they are stationary, or they do not have the problem of unit root (Gujarati and Porter, 2009). The use of non- stationary variables in the time series analysis may lead to misleading conclusions. The Augmented Dickey Fuller (ADF) test is employed to test for stationary.
6.6 Empirical Results
This section presents the results of the econometric estimation of the FDI-productivity nexus equation, as outlined in equations 6.1a to 6.1c and 6.2. To investigate the productivity effects of FDI on the primary, manufacturing and services sectors, four of these specifications, including the constant values of the independent variables and the lagged values (t-1) of the dependent variable, were estimated. The results for the FDI- productivity nexus model and the causality between each of the sectors are presented in the subsequent sections.