International Journal of Engineering and Applied Sciences (IJEAS)

ISSN: 2394-3661, Volume-2, Issue-12, December 2015

The Performance of Tourism Sector and Economic

Growth in Nigeria (1996-2010)
Ojo. Thompson Olabode (Ph.D), Oyekunle. Janet Oluwafunmike, Olaleye. O.A.

in Nigeria by providing all supporting and essential

Abstract This study employs the Granger causality test infrastructure that are lacking, this sector will collectively
within a multivariate cointegration and error-correction enhance tourism experience of potential tourists.
framework to investigate the relationship between tourism
arrival, tourism receipts and economic growth in Nigeria. This At present, Nigeria is facing a worse energy cum oil crisis of
study covers the annual sample from 1996 to 2010. The main
its history; Nigerias petroleum oil revenue is expected to
findings of this study are that in the short-run there is
uni-directional Granger causality running from tourism arrival
decrease in the next few years. Tourism development can be a
and tourism receipts to gross domestic product in Nigeria. useful tool for obtaining efficiency in economic growth and
While, in the long-run tourism arrival, gross domestic product hence enhancing our GDP. This paper aims at determining the
and tourism receipts are bi-directional Granger causality. relationship between tourism arrival and tourism receipts to
Hence, sound and developed tourism sector growth that can gross domestic product in Nigeria.
attract investors, boost the stock market, industrialization and
urbanization and improved efficiency of economic activities via 2. Data and Model Specification.
tourism activities should not be left out from the process of
economic growth and development in Nigeria.
This paper presents the analysis of the empirical results of the
study Performance of Tourism Sector and Economic Growth
Index Terms Causality, Cointegration; Tourism Arrivals and
with specific reference to Nigeria for the period 1996-2010.
Receipts; Economic Growth.
The data used here is secondary time series data covering the
period 1996-2010. The analysis considers annual data for
I. INTRODUCTION
Nigeria. The source of data is the World Development
In 2006, the Nigerian Government instituted a new set of Indicators database. Tourism arrivals (TR), Tourism Receipts
reforms in the tourism industry with the aid of the United (TR) and GDP are the variable used in the estimation of the
Nations Development. This led to the creation of the Nigeria model. The period of study was chosen because of availability
Tourism Master Plan. The report identified amongst other of data for the specified years also because in Nigeria,
things that the number of international visitors to Nigeria had capturing of tourism data began around the year 1994.To
been static over the last 15 years because of factors like measure the long run relationship between the variables, it is
stodgy entry visa processes, lack of information on Nigeria necessary that the data should be integrated of same order.
amongst tour operators in the west, bad road networks, and Augmented Dicky Fuller unit root is used to check the order
the dilapidated state of many of Nigerias historic sites. of integration. Johnson co-integration test and vector error
Shamefully, six years after the master plan had been drawn, correction model are used to determine the relationship
the commentary on Nigerias tourism industry will probably between variables and granger causality test is used to
be worse. Though there has been a major influx of global determine the direction of causality among the variables. The
hotel brands into Nigeria in the past five years, especially in following log linear model is estimated.
Lagos, the commercial capital of Nigeria, the average cost of p q

these rooms puts it out of reach of the average holiday maker ln GDPt 0 1TAt 1 2 ln GDPt 1 1 ln GDPt 1 2 ln TAt 2 t
t 1 i o
from Europe and America especially in these hard times. The
(1)
tourism master plan advocated for partnership programs p q
between the internationally operated hotels in Nigeria and ln GDPt 0 1TRt 1 2 ln GDPt 1 1 ln GDPt 1 2 ln TRt 2 t
t 1 i o
tour operators in the west to use spare capacity in these hotels
at favorable rates. Five years after, the tourism and hospitality (2)
industry still represents a meager 0.5% of Nigerias GDP Where;
(NTDC, 2013). ln is Natural Logarithm GDP is Real Gross Domestic
Product for Nigeria in $US, TR is tourism receipts In $US,
If properly developed and encouraged, tourism can help TA is International Tourism Arrivals, TR is Tourism Receipts
refocused Nigeria economy globally, especially in developing for Nigeria in $US, is First Difference Operator and is
country like ours where attention is being directed to serve as the error term. This present study estimated these two
revenue for economic diversification (Awodele & equations separately where Tourists Arrivals and Tourists
Ayeni,2011) . This will benefit Nigeria as a whole, as well as Receipts are independent variables and GDP is the
the local economy in the face of the present global oil dependent variable as indicated by equation (1) to (2).
misfortune (Bankole & Odularu, 2006). In view of this, it is The short run dynamics of the models is specified as follows:
very important to note that if tourism is sustainably developed p q
ln GDPt 0 1 ln GDPt 1 2 ln TAt 1 t
t 1 t 0
Manuscript received. (3)

49 www.ijeas.org
 The Performance of Tourism Sector and Economic Growth in Nigeria (1996-2010)

p q variables are integrated I(1). The VECM is restricted form of

ln GDPt 0 1 ln GDPt 1 2 ln TRt 1 t unrestricted VAR (vector autoregressive) and restriction is
t 1 t 0
levied on the presence of the long run relationship between
(4)
the series. The system of error correction model (ECM) uses
Where the variables are as previously defined and p and q
all the series endogenously. This system allows the predicted
are ARDL orders which are selected. The corresponding variable to explain itself both by its own lags and lags of
ECM model is specified below; forcing variables as-well-as the error correction term and by
p q
ln GDPt 0 1GDPt 1 2 ln TAt 1 ECM t 1 t residual term. The VECM equations are modeled as follows:
t 1 t 0
(5)
Yt 1 a11i a12i a13i a14i ....a118i Yt 1
p q

ln GDPt 0 1GDPt 1 2 ln TRt 1 ECM t 1 t
t 1 t 0
I L TAt 2 (1 L) b21i b22ib23ib24i .....b118i TAt 1
TRt 3 i 1 TR
c31i c32i c33i c34i ....c118i t 1
(6)
Where:
1 , 2 are the short run dynamics coefficient of the model,
1 1t
is the speed of adjustment and ECM is the error
correction term estimated from equation (1). 2 ECTt 1 2t
3 3t
The optimum lag structure of the first difference regression is
selected using Alkaike Information Criteria (AIC).
Where it are random terms and supposed to be normally
In Johansen-Juselius approach, distributed with zero means and constant variances. The
' is (n n) coefficient matrix called the impact established long run relation between the series is further
matrix and contains information about the long-run confirmed by the statistical significance of lagged error term
equilibrium relationship between the said variables. is the i.e. ECTt 1 . The estimates of ECTt 1 also shows the speeds
parameter denoting the speed of adjustment to disequilibrium, of convergence from short run towards the long run
while is a matrix of cointegrating vectors. equilibrium path. The vector error correction method
Johansen-Juselius cointegration approach offered two ( VECM ) is appropriate to examine causality between the
likelihood ratio variables once series are found to be cointegrated and then
( LR ) test statistics, namely trace test, causality must be found at least from one direction. The
k

LR trace T i r 1 ln 1 i and maximum
VECM also distinguishes causality relationships between
short-and-long runs. The VECM is also used to detect the
eigenvalues test, LR max T
k
i r 1
ln 1 i 1 , causality in the long run, short run and joint i.e. short and long
runs respectively.
where T represents the total numbers of observations and i
t statistic of the estimate of lagged error term i.e.
The
are the eigenvalues 1 2 ... k . ECTt 1 with negative sign is used to test the long run casual
The common practice for the multivariate Johansen-Juselius
relation and the joint statistical significance of the
2
cointegration test is to determine the lag structure for the
VECM system. In this respect, Hall (1991) pointed out that estimates of the first difference lagged independent variables
the choice of lag structure in the VECM system is vital is used to investigate short run causality. Tourism receipt
because too few lags may lead to serial correlation problem, Granger causes economic growth if 22, j 0 i are
whereas too many lags specified in the VECM system will
found statistically significant. On the contrary, if
consume more degree of freedoms thus lead to small sample
problem. For this reason, the optimal VECM system for 22, j 0 i is statistically significant then causality runs
multivariate Johansen-Juselius cointegration test was from economic growth to tourism receipts. The rest of causal
determined by minimizing the system-wise Akaikes hypotheses can be inferred similarly. The joint causality i.e.
Information Criterion (AIC). The AIC was used because Liew long-and-short runs are investigated by using Wald or
(2004) and Ltkepohl (2005) found that AIC performed F test for the joint significance of the estimates of lagged
better than any other information criterions when the terms of the independent variables and the error correction
estimated sample size is relatively small (e.g. less than 60 term. The presence of short-and-long runs causality relation
observations). The AIC statistic indicates that 2 years lag is between the variables is known as measure of strong
the optimal lag length for the multivariate Johansen-Juselius Granger-causality (Shahbaz et al. 2011 & 2013).
cointegration test.
The direction of causality between the variables is 3. Emprical Findings and Discussion
investigated by applying the VECM Granger causality This study intends to investigate the relationship between
approach after confirming the presence of cointegration economic growths, tourism arrival (TA) and Tourism
between the variables. On the same lines, Granger (1969) Receipts (TR) in Nigeria from 1995 2010. Time series data
argued that vector error correction method (VECM) is more usually have a tendency to be non-stationary, and the
appropriate to examine the causality between the series if the estimated results may indicate spurious results. To determine

50 www.ijeas.org
 International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-2, Issue-12, December 2015
the order of integration between the variables, test of unit root At most 3* 0.29915 7.09874 3.752466 0.0168
has been carried out, there are several unit root test available 9 4
to solve the problem of stationerity, however, we have used Trace test indicates 4 co-integrating equations at the 5%
Augumented Dickey Fuller test at level and at first difference. level, * denotes rejection of the hypothesis at the 5% level
The results in table-1 indicate that the variables are and ** Mackinnon-Haug-Michelis (1999) p-value.
non-stationery at level, thus carry unit root. When the unit
root is tested at first difference, estimates show stationery Table 3: JOHANSEN CO-INTEGRATION TEST
properties, which mean the variables, are integrated of order (MAXIMUM EGIEN VALUES)
1, I 1 . Hypothesized Eigenvalue Max-Eige 5%Critica Prob.**
Number of n Statistic l Value
Table 1: Stationarity Tests: Unit Root Test CE(s)
None* 0.82954 61.81195 36.72407 0.0000
0
At most 1* 0.66483 32.59550 25.24301 0.0033
1
At most 2* 0.53556 24.46201 15.13657 0.0025
8
At most 3* 0.29915 7.088674 3.771346 0.0133
2
Max-eigenvalues test indicates 4 cointegrating equations
at the 5% : * denotes rejection of the hypothesis at the 5%
level : ** Mackinnon-Haug-Michelis (1999) p-value.

The presence of co-integration show long-run relationship

between the variables. This indicates that there exists an error
correction model, which combines the short run effects with
the long run and indicates how much of previous
disequilibrium is removed in the current year. The results of
the Vector Error Correction Model (VECM) given in Table
IV gives results of long run and short run causalities as well as
NOTE: ** Order of integration at 5% significant level; the ECM. The coefficients of ECM explain a one period lag of
*** Order of integration at 10% significant level the cointegrating vectors between GDP and TA and GDP and
TR. The ECM coefficient of TA is positive and insignificant
Since the variables in the model are now stationery and are indicating that TA has no long run causality on GDP or in
integrated of same order we now apply the Johansen other words, there is no long run causality between the two.
co-integration test to determine the long term relationship As for TR, the coefficient of ECM is negative and significant
between the variables. The VAR method is used to at one percent level, meaning there is a long run causality
determine the original lag length and stability condition. The running from TR to GDP. This also ensures that the
" AIC " criteria determine the lag length and Because of adjustment process from the short run deviation is small. In
the number of observations, the test could not go beyond two other words, only 3 percent of the disequilibrium in GDP
lags and thus lag two with a lower value becomes the optimal from the previous periods shock will converge into long run
lag length. This was also confirmed by the HQ criteria which equilibrium in the present period.
The short run causality is determined by the probability of
also estimated 2 lags, hence it support the lag 2 as the
chi-square where the null hypothesis indicates presence of
optimal choice, The rule of thumb here is that the lowest value short run causality and the alternative; no short run causality.
of AIC is chosen when trying out the number of lags. The test If the probability value is less than 5%, we fail to reject null
was done firstly with one lag which gave a value of -1.945323 and accept the presence of short run causality. From the
and then two lags which gave a value of -2.215089. Once the results, probability of chi-square for TA is (0.7142) which is
optimal lags were established, the second step was to apply greater than 5% meaning we reject the null and accept the
the Johansen co-integration in order to establish a alternative that there is no short run causality between TA and
cointegration relationship between the three variables. GDP. However, the results of TR show a chi-square
Johansen co-integration, in Table 2 and Table 3 presents both probability of (0.0002) indicating the presence of short run
the trace and maximum eigenvalues. causality running from TR to GDP. In other words, there is
both the short and long run causality running from TR to GDP
Table 2: JOHANSEN CO-INTEGRATION TEST and no causality from TA to GDP in both the short and long
(TRACE STATISTICS) run period.
The estimated coefficients of TA and TR are -0.210237 and
Hypothesized Eigenvalue Trace 5%Critica Prob.**
Number of Statistic l Value
-0.448882 respectively. These also confirm the short run
CE(s) impacts explained above. The coefficient of TA is negative
None* 0.82954 112.169 57.24487 0.0000 and insignificant, implying no short run impacts on the
2 0 economy of Nigeria. While on the other hand impact on the
At most 1* 0.66483 61.4673 35.01090 0.0000 economy of Nigeria is confirmed by significance of the TR
6 0
At most 2* 0.53556 29.5698 18.39771 0.0000
coefficient. A unidirectional relationship from TR to GDP is
7 8 also confirmed.

51 www.ijeas.org
 The Performance of Tourism Sector and Economic Growth in Nigeria (1996-2010)

Table 4 : VECM Granger Causality Anaysis model 1 and 2. Therefore we fail to reject null and conclude
that residuals are normally distributed.
Dependent Variable: DLNGDP
4.2: Arch Test under Heteroscedasticity
Under this test, the null hypothesis is there is no arch effect
ECM Prob.Chi-squar
e and the alternative is there is arch effect. A probability value
DLNTA -0.210237 0.170101 of more than 5 percent would mean there is no arch effect. In
(0.26824) (0.983616) this case, probability of chi-square is used under observed
[-0.78378 [0.172935] 0.7142 R-squared. Table 6 and 7 give the results for model 1 and 2
]
respectively.
DLNTR -0.448882 -0.400474 0.0002*
(0.12844) (0.125915)
[-3.49483 -[3.18050 Table 6 Heteroscedasticity Test: ARCH model 1 (GDP
] 5] and TA)

F-statistic 0.436186 Prob. F(2,8) 0.6610

Prob.
Standard errors in ( ) & t-statistics in [ ]; *indicates Obs*R-squared 1.081571 Chi-Square(2) 0.5823
significance at one percent level
The results of the unidirectional relationships confirms those
found by Isla, et.al (2011), Kaplan, et.al (2011), Mishral, et.al
(2009) and Bowden and Payne (2009) who used the method Table 7 Heteroscedasticity Test: ARCH, Model 2 (GDP
of OLS; Shahbaz, et.al (2010), Sadorsky (2010), who used and TR)
ARDL approach.
4. Diagnostic Test Results F-statistic 0.578224 Prob. F(2,8) 0.5827
Various tests have been undertaken to ensure that the model Prob.
used in this study are not spurious and misleading. This was Obs*R-squared 1.389286 Chi-Square(2) 0.4993
done to ensure efficiency. The residuals were first tested for
stability and the results are given on table 5 below. The
The results above indicate that for both models, the
analysis is based on the fact that if t statistic is greater than
probability is greater than 5 percent, therefore indicating no
critical value we can reject the null of unit root and accept the
arch effect. The probability values for both models are 0.5823
alternative of no unit root. The critical value used in this case
and 0.499 respectively
is that of Engel Granger which 3.04. Therefore since 3.04 is
less than the absolute value of t in this case; 3.14 at 5 percent
4.3 Serial Correlation Test
significance level it means unit root does not exist. Therefore
The Breusch-Godfrey Serial Correlation LM Test was used in
the estimated model is not spurious.
this study to test for autocorrelation. In the same way, the
probability of chi-square was used and the results are
Table 5: Residual Unit Root Test Results indicated in tables 8 and 9 for model 1 and 2 respectively. The
null hypothesis: there is no serial correlation was tested
t-Statistic Prob.*
against the alternative; model has serial correlation.
Augmented Dickey-Fuller test statistic -5.238798 0.0017
Table 8 Breusch-Godfrey Serial Correlation LM Test
Test critical
Results
values: 1% level -4.121990
5% level -3.144920
10% level -2.713751 F-statistic 2.245840 Prob. F(2,5) 0.2014
Prob.
Obs*R-squared 6.151898 Chi-Square(2) 0.0461
Variable Coefficient Std. Error t-Statistic Prob.

U(-1) -1.451343 0.277037 -5.238798 0.0004

C -0.003205 0.009835 -0.325854 0.7513

4.1: Normality Test Results

A good regression is supported by normal distribution of Table 9 Breusch-Godfrey Serial Correlation LM Test
residuals. This test was carried out under Jacque Berra (JB) Results
under the null hypothesis; residuals are normally distributed
and alternative; residuals are not normally distributed. If the F-statistic 0.400686 Prob. F(2,7) 0.6843
probability of JB is greater than 5 percent we fail to reject the Prob.
null and accept that residuals are normally distributed. In this Obs*R-squared 1.000159 Chi-Square(2) 0.6065
case, the results gave a probability value of 0.606 and 0.769
for Jacque Berra of which is greater than 5 percent for both
Table 10 Breusch-Godfrey Serial Correlation LM

52 www.ijeas.org
 International Journal of Engineering and Applied Sciences (IJEAS)
ISSN: 2394-3661, Volume-2, Issue-12, December 2015
TestResults [9] Lutkepohl, H. (2005). New Introduction to Multiple Time Series
Analysis, Berlin: Springer, Verlag.
[10] Mishra, V., Sharma, S.S., Smyth, R., (2009), Is economic development
F-statistic 1.018496 Prob. F(2,5) 0.4256 export-led or import-led in the Pacific Island Countries? Evidence
Prob. from panel data models. Pacific Economic Bulletin 25, 46-63.
Obs*R-squared 3.763099 Chi-Square(2) 0.1524 [11] Sadorsky, P., (2010), The impact of financial development on energy
consumption in emerging economies. Energy Policy 38,2528-2535.

[12] Shahbaz,M,. Islam, F., Islam, M.M, (2010), The impact of financial
The results of serial correlation test using the development on energy consumption in Pakistan: evidence from
Breusch-Godfrey test show that model 1 had serial ARDL bound testing approach to co-integration. In: Proceedings of
the Second International Conference on the role of Social Sciences and
correlation. This was confirmed by the probability value of Humanities in Engineering (COHSE) 2010, 12-14 November 2010,
0.0461 which is is less than 5 percent (table 9). This was Bayview Beach Resort, Penang, Malaysia.
corrected by running the model in first difference and removal
of intercept. The results of the test after this application are [13] Shahbaz M, Mutascu M, Azim P. (2013) Environmental Kuznets Curve
indicated in table 10. The probability value of 0.6065 was in Romania and the Role of Energy Consumption. Renewable and
Sustainable Energy Reviews 2013; 18: 165-173.
found, which means the model now had no serial correlation,
that is to say fail to reject null. Table 6 on the other hand show
that there is no serial correlation in model 2 as indicated by a
probability value of 0.1524 which is greater than 5 percent. Ojo. Thompson Olabode (Ph.D), Department of Statistics, School of
Therefore, we cannot reject the null hence the model has no Applied Sciences, The Federal Polytechnic , P.M.B. 231, Ede, osun-state.
Nigeria
serial correlation.
Oyekunle. Janet Oluwafunmike, Department of Statistics, School of
5. Conclusion Applied Sciences, The Federal Polytechnic , P.M.B. 231, Ede, osun-state.
This paper set out to investigate the need for the enhancement Nigeria
of the Nigerian tourism and tourist attractions for sustainable Olaleye. O.A., Department of Statistics, School of Applied Sciences, The
development. Results as explained above have confirmed Federal Polytechnic , P.M.B. 231, Ede, osun-state. Nigeria
tourism led growth hypothesis for Nigeria. In other words,
there is a possibility of the economy of Nigeria to grow from
injections from tourism sector. The results have also shown
that this tourism led growth dominates in the long run than in
the short run as shown by a negative unidirectional causality
from tourism receipts to GDP. The results are consistent with
the findings of Ercan (2011) who examined the potential
contribution of tourism to the economic growth of Istanbul
economies. The results indicated that tourism contributes
significantly to the current level of GDP and the economic
growth of Nigeria as a country. Similar conclusions were also
found by Ikein et.al., (2008) for the oil, democracy and the
promise of true federalism.

References
[1] Awodele, O.A & Ayeni, D.A. (2011). Managing risk in tourism
development project: A case study of Nigeria. Journal of Construction
Project Management and Innovation 1(2), 105 129.
[2] Bankole, A.S. & Odularu, G.O. (2006). Achieving the millennium
development goals: Issues and Options for the Nigerias tourism
industry. PASOS. Revista de Turismo y Patrimonio Culture 4(3),
429 436.
[3] Bowden, N., Payne, J.F., (2009), The Causal relationship between US
energy consumption and real output: a disaggregated analysis.
Journal of Policy Modeling 31, 180 188.
[4] Ercan, M.A. (2012). Challenges and conflicts in achieving sustainable
communities in historic neighborhood of Istanbul. Habitat
International, 35(2), 295 306.
[5] Hall,S.G. (1991) The effect of varying length VAR models on the
maximum likelihood estimates of cointegrating vectors. Scottish
Journal of Political Economy, 38(4), pp. 177-187.

[6] Ikein, A.A.,Alamieseigha, D.S.P., & Azaiki, S.S. (2008). Oil, democracy
and the promise of true federalism. Maryland: University Press of
America.
[7] Islam, F., Shahbaz, M., Alam, M., (2011) , Financial Development and
Energy Consumption Nexus in Malaysia: A Multivariate Time Series
Analysis, MPRA Paper 28403, University Library of Munich,
Germany.
[8] Kaplan. M., Ozturk,I., Kalyoncu, (2011), Energy Consumption and
Economic Growth in Turkey; co-integration and causality analysis.
Romanian Journal of Economic Forecasting 2, 31-41.

53 www.ijeas.org