^{1}

^{*}

^{2}

^{3}

This is a causality study on Foreign Direct Investment (FDI), Unemployment and Economic growth in Uganda using time series Autoregressive Distributed Lag (ARDL) bounds approach based on FDI, unemployment rates and GDP datasets from 1993 to 2015. Curtailing levels of unemployment and simultaneously sparkling levels of economic growth are primary macroeconomic objectives of every country. This study intends to be instrumental in advising alternative policies that aim at reducing unemployment and sparkling economic growth; one of which is the attraction of FDI. Literature has no concrete premise for this hypothesis since evidences from some countries show positive results while others show negative results. Our study adopts dynamic ARDL bounds approach with efficient and successful record of undertaking empirical studies of this nature in time series. Using data series obtained from the world bank, our findings indicate no causalities between the variables considered. Therefore, there is no sufficient statistical evidence to suggest that FDI plays significant roles in reducing unemployment and sparkling economic growth. The short-run and long-run dynamics of the model do not point to any statistically significant relationships. Based on this specific country-case premise, the study recommends on the need to revitalize domestic industries and re-strategize FDI comprehensive policy frameworks to provide competitive edge to domestic firms and attract FDI in a pace consistent with the growth agenda of local industries.

Economic backwardness and poverty have for long consumed greater part of the economic history of Sub-Saharan Africa countries. This has continued to worsen over the recent decades. Uganda being part of this cluster of nations suffers from extreme levels of poverty, high unemployment, poor infrastructure, large fiscal deficits and very low levels of economic growth among others [

Recent statistics drawn from the world investment report indicate that global flows of FDI to developing countries declined from 1.5 trillion in 2012 to $1.2 trillion dollars in 2015 and FDI flows to Africa continued to decline in 2016, by three per cent to $59 billion, according to United Nations Conference on Trade and Development (UNCTAD) [

However, against all these backgrounds of both theoretical and empirical justifications about the contributions of FDI in curtailing unemployment and sparkling growth in Sub Saharan countries like Kenya, it is noteworthy that there is no any clear link between FDI and employment and economic growth in host countries since several other empirical evidences show negative results. Therefore, given the inconsistency with which FDI relates to unemployment and economic growth in various countries, this study focuses on Uganda to establish how FDI impacts on unemployment and economic growth.

Fluctuations in business cycles and subsequent spread of global economic crises have propelled scholars into devotion of huge amounts energy and anxiety in studying interdependence between macroeconomic variables and FDI inflows. These studies have considerably focused on labor markets imperfections by using employment or unemployment as a proxy variable for macroeconomic stability [

A study employing dynamic panel causality carried out by Yayli and Deger [

Studies by Jayaraman and Singh [

Causality relationships between FDI, exports, unemployment and GDP for Turkey for the period of 2000-2007 carried out by Aktar and Oztur [

Similar studies were done in Poland to examine the relationship between FDI, unemployment and GDP over the period of 1995-2009 by Balcerzak and Żurek [

Another similar study was on Sri Lanka by Velnampy, Achuchuthan & Kajananthan [

Studies done by Shaari, Hussain and Abralim [

Studies by Habib and Sarwar [

Based on the examination of the above scholarly works, it’s evident that while several studies have alluded to the direction of a positive relationship between FDI and Unemployment and economic growth, it’s worthy to make independent country-case studies since there are divergent results showing negative relationships for some countries.

The ARDL model is an interventional method for time series models that do not clearly feature the elements of simple and standard OLS as well as Error Correction models (ECM). This method works with series that are of varying integral order I (0) or I (1), but not I (2). It carries more preferential weight among econometric researchers because different variables assume different lags as they are fitted into the model in additional to its ease and simplicity to execute among others, Pesaran and Shin [

In estimating ARDL models, we clearly and consistently undertook the following procedural steps:

1) We estimated the unrestricted VAR error-correction model as a particular form of ARDL model.

2) We then determined the optimal lag structure of the model.

3) Model diagnostics was thereafter performed by testing whether the errors of the model were serially correlated. This was done using LM test.

4) Dynamic stability of the model was then examined to ensure that the parameters of the model have not changed over the sample period.

5) We tested the evidence of long term relationship among variables by performing bound test.

6) The next procedure required that in case our results from the bound test indicate significant relations, we proceed to estimate long-run levels models and restricted error correction model.

7) Finally, we make use of the results to formulate short-run and long-run dynamics about the variables we are studying.

ARDL models are defined by having lags of dependent variables and other variables as regressors. In our study, our basic formulation for the models can be illustrated from the traditional ARDL as follows:

Δ y t = β 0 + Σ β i Δ y t − i + Σ γ j Δ x 1 t − j + Σ δ k Δ x 2 t − k + φ z t − 1 + e t (1)

This study employs time series data of FDI, unemployment rates and economic growth using data obtained from the world bank data portal http://data.world.org/country/uganda between 1993-2015. This interval was selected because it had data available for the variables under study and the series was long enough to conduct a time series study.

Study of time series requires that the underlying series is stationary. A series is considered non-stationary when it has unit root. Such a series has no tendency to revert to its long-run deterministic path and its variance is dependent on time. There are many methods of testing unit root, but this study employs two superior methods: The Augmented Dickey Fuller (ADF) [

Δ Y t = α 0 + р 1 Y t − 1 + α 2 T + ∑ α Y Δ + u t (2)

р = 0 is tested and follows the same asymptotic distribution as DF statistic.

H 0 : р 1 = 0 ( р 1 ∼ I ( 1 ) ) , against H a : р 1 < 0 ( р 1 ∼ I ( 0 ) ) .

The null hypothesis for ADF reads that the series is non-stationary and it’s similar to that of PP. Our results of the test indicated that at initial levels, FDI and GDP were non-stationary while UNEM was stationary. FDI and GDP were made stationary by taking the first difference after which they became stationary. Our choice of the model for unit root was the one with intercept and no trend (

The general ARDL model is presented as follows;

Δ Y t = α 0 + ∑ Φ i Δ y t − i + ∑ β j Δ x t − j + ∑ δ k Δ z t − i + θ 1 y t − 1 + θ 2 x t − 1 + θ 3 z t − 1 + u t (3)

Augmented Dickey Fuller Test(ADF) | Philips-Perron Test (PP)A | |||
---|---|---|---|---|

Variables | Intercept | Intercept & Trend | Intercept | Intercept & Trend |

LN_FDI | −0.7409 | −3.6194 | −0.6604 | −2.3142 |

DLN_FDI | −4.8652* | −3.6887** | −4.8795* | −4.7451* |

LN_GDP | −0.7056 | −1.5561 | −0.7396 | −1.8186 |

DLN_FDI | −3.84601* | −3.7444* | −2.8208* | −3.7023** |

LN_UNEM | −4.6226* | −4.5646* | −4.6797* | −4.6069* |

DLN_UNEM | −5.0694 | −5.0084* | −14.5074* | −13.9953* |

Notes: 1. *, **, *** imply significance at the 1% 5% and 10* levels respectively. 2. PP test suggests a non-parametric method of controlling for higher order autocorrelation in a series. The null and the alternative hypotheses tested are the same as for the ADF test.

This is typical of ECM, where Ф_{i}, β and δ_{i} are the short-run coefficients while θ_{1}, θ_{2} and θ_{3} represent the ARDL long-run coefficients and the error-correction term, z_{t}_{-1} is replaced by the y_{t}_{-1}, x_{t}_{-1}, and z_{t}_{-1} terms. Lagged residuals series therefore become z t − 1 = ( y t − 1 − a 0 − a 1 x 1 t − 1 − a 2 x 2 t − 1 ) ._{ }

In our study, we specify our ARDL models as follows (based on first lag differences):

D L N _ F D I = α 0 + Φ 1 D L N _ F D I + δ 2 D L N _ G D P + θ 3 D L N _ U N E M + Φ 1 L N _ F D I + δ 2 L N _ G D P + θ 3 L N _ U N E M + u t (4)

D L N _ G D P = α 0 + Φ 1 D L N _ F D I + δ 2 D L N _ G D P + θ 3 D L N _ U N E M + Φ 1 L N _ F D I + δ 2 D L N _ G D P + θ 3 L N _ U N E M + u t (5)

D L N _ U N E M = α 0 + Φ 1 D L N _ F D I + δ 2 D L N _ G D P + θ 3 D L N _ U N E M + Φ 1 L N _ F D I + δ 2 L N _ G D P + θ 3 L N _ U N E M + u (6)

Based on appropriateness from the diagnostics, we estimated the following

model: (

DLN_FDI = α 0 + Φ 1 DLN_FDI + δ 2 DLN_GDP + θ 3 DLN_UNEM + Φ 1 LN_FDI + δ 2 LN_GDP + θ 3 LN_UNEM + u (7)

where DLN_FDI, LN_GDP & LN_UNEM are the short-run log differences/ changes for the study variables while the last three represent the long run dynamics.

The results below indicate initial regression results for both the long-run and short-run dynamics. The estimates guide in choosing appropriate dynamics of the model and they do not explain causality.

After proper specification of the ARDL model, the next critical step is the choice of appropriate lags to be included in the model. This is necessary to determine the optimum lag length(k) by using proper model order selection criteria such as; the Akaike Information Criterion(AIC), Schwarz Bayesian Criterion (SBC) or Hannan-Quinn Criterion(HQC) to obtain Gaussian error terms that are free from serial correlation and heteroscedasticity. These criteria are based on a high

Variable | Coefficient | Std. Error | t-Statistic | Prob. |
---|---|---|---|---|

C | −0.739908 | 7.139846 | −0.103631 | 0.9189 |

D(LN_FDI(−1)) | 0.009157 | 0.290757 | 0.031495 | 0.9753 |

D(LN_GDP(−1)) | −1.334785 | 1.088679 | −1.226060 | 0.2404 |

D(LN_UNEM(−1)) | −0.688412 | 0.497915 | −1.382590 | 0.1884 |

LN_FDI(−1) | −0.064723 | 0.227299 | −0.284746 | 0.7800 |

LN_GDP(−1) | 0.081406 | 0.440376 | 0.184855 | 0.8560 |

LN_UNEM(−1) | 0.389463 | 0.757263 | 0.514304 | 0.6151 |

log-likelihood value, with a “penalty” for including more lags to achieve this. The form of the penalty varies from one criterion to another. Each criterion starts with −2log(L), and then penalizes, so the smaller the value of an information criterion the better the result. As shown in

After estimating our appropriate model equation using the appropriate lag, we carried out model diagnostics by checking for autocorrelation using the Breusch-Godfrey Serial Correlation LM Test method under the null hypothesis that there is autocorrelation against that of the alternative that errors are either AR(p) or MA(q), for p and q = 1, 2, 3, ... This is a key element in the assumptions of the ARDL/Bounds Testing methodology of Pesaran et al. (2001). As presented, the P-value associated with the Chi-square statistic is way above 5% hence we cannot reject the null hypothesis that errors are serially independent (

The second most important model diagnostics tests is the dynamic stability of the model to assess whether the model’s parameters changed at one or more points in the sample period.

In our study, model stability was carried using Recursive OLS Estimates- CUSUM Test and our results indicates that the trend line lies within the boundaries as shown below. The CUSUM (Cumulative Sum) statistics are defined according to:

Lag | LogL | LR | FPE | AIC | SC | HQ |
---|---|---|---|---|---|---|

0 | −20.12573 | NA | 0.669700 | 2.434287 | 2.583409 | 2.459525 |

1 | −11.41311 | 13.75677* | 0.298480* | 1.622432* | 1.821262* | 1.656082* |

2 | −11.11522 | 0.438988 | 0.323403 | 1.696339 | 1.944876 | 1.738402 |

3 | −11.07017 | 0.061650 | 0.361075 | 1.796860 | 2.095104 | 1.847335 |

4 | −8.720429 | 2.968095 | 0.317668 | 1.654782 | 2.002733 | 1.713669 |

Notes: *indicates lag order selected by the criterion LR: sequential modified LR test statistic (each test at 5% level) FPE: Final prediction error AIC: Akaike information criterion SC: Schwarz information criterion HQ: Hannan-Quinn information criterion.

F-statistic | 0.209179 | Prob. F(1,13) | 0.6550 |
---|---|---|---|

Obs*R-squared | 0.332554 | Prob. Chi−Square(1) | 0.5642 |

Source: Researcher’s estimations.

CUSUM t = ∑ i = k t w i + 1 , i

for t = k , k + 1 , ⋯ , T − 1 , where k = 2 p + s + 1 is the minimum sample size for which we can fit the model. Under the null hypothesis, the CUSUM_{t} statistic is drawn from a CUSUM(t-k) distribution. The CUSUM(t − k) distribution is a symmetric distribution centered at 0. Its dispersion increases as t − k increases. We reject the null hypothesis at the 5% significance level if CUSUM_{t} is below the 2.5-percentile or above the 97.5-percentile of the CUSUM(t − k) distribution. The output will be a

Our choice for the ARDL model can be represented as follows;

DLN_FDI = α 0 + Φ 1 DLN_FDI + δ 2 DLN_GDP + θ 3 DLN_UNEM + Φ 1 LN_FDI + δ 2 LN_GDP + θ 3 LN_UNEM + u t _{ }

This test is carried out to examine the existence of long term relationships between variables under study. However, the statistical significance of the F-statistics is not based on the provided p-value because the distribution of the test statistic is totally non-standard. We therefore, use the Pesaran table since the exact critical values for the F-test are not available for arbitrary mix of I(0) and I(1) values and it provides bounds on the critical values for the asymptotic distribution of the F-statistic. For various situations (e.g., different numbers of variables, (k + 1)), they give lower and upper bounds on the critical values. In each case, the lower bound is based on the assumption that all of the variables are I

Test Statistic | Value | df | Probability |
---|---|---|---|

F-statistic | 3.472919 | (3, 14) | 0.0451 |

Chi-square | 10.41876 | 3 | 0.0153 |

(0), and the upper bound is based on the assumption that all of the variables are I (1).

Ho: Φ 1 = δ 2 = θ 3 = 0 (null, i.e. the long run relationship does not exist)

H1: Φ 1 = δ 2 = θ 3 ≠ 0 (Alternative, i.e. the long run relationship exists)

The task is a simple test of co-integration for the absence of a long-run equilibrium relationship between the variables. This coincides with zero coefficients for y_{t}_{-1}, x_{1t-1} and z_{t-1}. A rejection of H_{0} implies that we have a long-run relationship. From the Pesaran table, the Upper bound is I (1) = 4.35 and the Lower bound I (0) = 3.25 for our model with intercept and no trend (

Our results are as follows:

Lower bound I (1) =3.23.

Upper bound I (0) =4.35.

This study was carried out against the backdrop of varied experiences across the world on the relationship between FDI, unemployment and economic growth. While numerous theoretical and empirical evidences have seemed to suggest positive directional relationships in countries such as Kenya, Zimbabwe, Malaysia and Pakistan, other country cases found different results for the same kind of studies. Without any reliable claim to ascertain specific country experience, advising policy for such hot debates can be pretty much troublesome especially for impoverished countries like Uganda.

This study sought to reasonably provide unbiased understanding, insights and conclusions based on empirical assertions by employing time series bounds approach. Our findings did not provide any statistical evidence of cointegration among FDI and unemployment and economic growth. This suggests that FDI does less in curtailing unemployment and sparkling economic growth. These findings are similar to the experiences from Turkey (Yayli and Deger (2012) and Nigeria; Salami and Oyewale (2013). However, given the enormity of econometric tools employable to an intensive study of such debates, we cannot assertively conclude that Uganda is isolated from the rest of the world in terms of benefits from FDI. Findings of this nature, therefore, advise policy to reenergize the potential of domestic industries and first create suitable competitive environment for them instead of prematurely allowing in FDIs.

Uganda should, therefore, prioritize and revitalize its domestic potential in all sectors and re-strategize its FDI policy in a comprehensive framework that embodies and prioritizes competitiveness for its domestic industries.

Garang, A.P.M., Yacouba, K. and Thiery, K.K.Y. (2018) Time Series Bounds Approach to Foreign Direct Investment, Unemployment and Eco- nomic Growth in Uganda. Modern Economy, 9, 87-96. https://doi.org/10.4236/me.2018.91006